Datasets:

naivete5656

/

strogen-arxiv

like 0

[ "There are decentralized models {{cite:6d30d451e9c3e38783de7e2329bba28a96c3373c}} and systems {{cite:9dc373a8f244f117e8d23c0e05b4394c813dd25f}}to support collaborative data sharingwithout a single schema.", "According to Wu et al., it is difficult to navigate an OLAP schema without help; they have proposed a keyword-driven OLAP model {{cite:854f3607614c99d277d38e60fbcd96552015ecbb}}.There are several OLAP visualization techniques including theCube Presentation Model (CPM) {{cite:8145d8e722c98647ef6ac9b782290db6ed0f661b}}, Multiple CorrespondenceAnalysis (MCA) {{cite:5f291d975ee4d1d596766dcd02ee486fecd18828}} and other interactive systems {{cite:5d80c018a1c20ea43e9ce033024230545a0d5ac8}}.", "Tag clouds have been popularizedby the Web site Flickr launched in 2004. Severaloptimization opportunities exist: similar tags can beclustered together {{cite:e6e5e72ee0e24e687923a0c5b804f4c2da61c47a}},tags can be pruned automatically {{cite:482463eb468372343e66113758b8619226c5dc15}}or by user intervention {{cite:57df1f63b3a055bb17208c888038465fd7d0c767}},tags can be indexed {{cite:57df1f63b3a055bb17208c888038465fd7d0c767}}, and so on. Tag clouds can be adapted to spatio-temporal data {{cite:a9c957297619f65506bbc46719df69da153141ab}}, {{cite:04023e202579ef4fbcf648e4c47f9289d0ead88d}}." ]

[]

[ "Batson, Spielman, and Srivastava {{cite:31b0dfcfbda4f1fbcd31fa57bab15faab548a782}} have given a deterministicalgorithm that constructs sparsifiers of size {{formula:9f139acc-b091-4e11-a81c-6e33f0c52757}} in{{formula:69ee8f78-f3a0-45b6-a6f9-fbdbebb0faa0}} time.While this is too slow to be useful in applications, it isoptimal in terms of the tradeoff between sparsity and quality ofapproximation and can be viewed as generalizing expander graphs.Their construction parallels ours in that it reduces the task of spectral sparsification to approximating the matrix{{formula:7467dcbc-d654-4078-bc83-0a6a6b49c1fc}} defined in Section 3; however, their method for selecting edges is iterativeand more delicate than the random sampling described in this paper.", "In addition to the graph sparsifiers of {{cite:a43e35ffff8cd60a00f0b78a07712def95abaa75}}, {{cite:31b0dfcfbda4f1fbcd31fa57bab15faab548a782}}, {{cite:b2b5a8f730ef4c373645713ed8e4eaa1393d94ae}},there is a large body of work on sparse {{cite:c211f5168620f47aed7f42d2d8da083e95e992b0}}, {{cite:4aecebad11b6dbdfc442852ece83a111fd34ffe1}} and low-rank{{cite:5b73954239e11e201e5946afa8c93ce533f9716c}}, {{cite:4aecebad11b6dbdfc442852ece83a111fd34ffe1}}, {{cite:bd2297989914b1918c5d8c055b9a3bc3d589b7ca}}, {{cite:dc5a8c27cc87c2ea894cb9e561ac38b54c2951f8}}, {{cite:88e6364bb365567066467dab2babdc2f99522112}}approximations for general matrices. Thealgorithms in this literature provide guarantees of the form {{formula:80753b01-4b1a-4918-af33-e22a2b34fb4e}} , where {{formula:9c42b08c-c4c6-449f-ab74-70bc2fcaf6d6}} is the originalmatrix and {{formula:02cac628-dd94-46bb-98be-ae2db88ccf4b}} is obtained by entrywise or columnwise sampling of {{formula:a289f69f-3f02-404d-864d-2dd20664259a}} . This isanalogous tosatisfying (REF ) only for vectors {{formula:a70ada77-fe5f-44be-8db7-7d48cfc7fe66}} in the span of the dominant eigenvectors of {{formula:c31df31b-7689-47a6-9e13-da8918d02c45}} ; thus,if we were to use these sparsifiers on graphs, they would only preserve the large cuts. Interestingly, our proofuses some of the same machinery as the low-rank approximation result of Rudelson and Vershynin {{cite:bd2297989914b1918c5d8c055b9a3bc3d589b7ca}} — thesampling of edges in our algorithm corresponds to picking {{formula:6cb0fd6e-357b-4961-9739-873c0fb085b1}} columns at random from acertain rank {{formula:4b6ddfe0-6d28-4564-a1fc-e5e797bf4fe5}} matrix of dimension {{formula:220d39cb-caa3-4113-ab9b-258e514353ef}} (this is the matrix {{formula:d5e92fa6-59ab-4f87-a04b-3c4301b07602}} introduced in Section 3).", "The use of effective resistance as a distance in graphs has recently gained attentionas it is often more useful than the ordinary geodesic distance in a graph.For example, in small-world graphs, all vertices will be close to one another, but those with a smallereffective resistance distance are connected by more short paths.See, for instance {{cite:b9a7b531c3dae1e09c1a5b52300ff409da4f81f9}}, {{cite:c3ac809c9218b1fd5a9b8e0849f13da1190de930}}, which use effective resistance/commute timeas a distance measure in social network graphs." ]

[]

[ "In this paper we have shown that multiple heads augment the expressive power of CHR. Indeed we have seen that the single head CHR language, denoted by CHR{{formula:a6774294-6c04-4e00-adb4-873a7a535500}} , is not Turing powerful when the underlying signature (for the constraint theory) does not contain function symbols, while this is not the case for CHR.Moreover, by using a technique based on language encoding, we have shown that CHR is strictly more expressive than CHR{{formula:bc3bc3ed-2167-453d-bf60-ccdf109d000f}} also when considering a generic constraint theory, under some reasonable assumptions (mainly, compositionality of the translation of goals).Finally we have shown that, under some slightly stronger assumptions, in general the number of atoms in the head of rules affects the expressive power of the language. In fact we have proved that CHR{{formula:80e9a985-9ce7-4966-b1e1-26b11ecf5206}} (the language containing at most {{formula:fda50291-8ee9-44e3-8bee-1129bad798a7}} atoms in the heads of rules) cannot be encoded into CHR{{formula:cd7b3cb7-c4bd-4b2f-8da1-7a949b066011}} , with {{formula:2864e873-09bb-4bbb-bde4-0a774f9510d2}} .", "There exists a very large literature on the expressiveness of concurrent languages, however there are only few papers which consider the expressive power of CHR. A recent one is {{cite:7d89cd833054a856d626bc033d0efce78592c190}}, where Sneyers shows that several subclasses of CHR are still Turing-complete, while single-headed CHR withouthost language and propositional abstract CHR are not Turing-complete. Moreover{{cite:7d89cd833054a856d626bc033d0efce78592c190}} proves essentially the same result given in Theorem by using Turing machines rather than Minsky machines. Both Theorems and were contained in the short version of this paper {{cite:a0f89d7e8e7e34aef9672aedcab05d7c67936a4b}}, submitted before {{cite:7d89cd833054a856d626bc033d0efce78592c190}} was published and both these results, including the encoding of the Minsky machine, were suggested by Jon Sneyers in the review of an older version ({{cite:066dc7485591e7b69277f4e00470b2e36ce4f17c}}) of {{cite:a0f89d7e8e7e34aef9672aedcab05d7c67936a4b}}. It is worth noting that very similar encoding exists in the field of process algebras. For example, in {{cite:cfba5298cac41ba4b18bb54f1e34467c27153342}} an encoding of Minsky machines in a dialect of CCS is provided which represents the value {{formula:ac9fcf10-9972-45c6-a614-9ad24b0b6b97}} of a register by using a corresponding number of parallel processes connected in a suitable way. This is similar to the idea exploited in Section , where we encoded the value {{formula:a06b872b-37c3-4876-8659-8d80f1094254}} of a registers by using using a conjunction (the CHR analogous of CCS parallel operator) of {{formula:f19f3866-952b-41da-b909-e7fcd45354ec}} atomic formulas.", "Another related study is {{cite:d573a45ee7b386ba5b57e76e1499cdd4d84e34fc}}, where the authors show that it is possible to implement any algorithm in CHR in an efficient way, i.e. withthe best known time and space complexity. This result is obtained by introducing a new model of computation, called the CHR machine, and comparing it with the well-known Turing machine and RAM machine models. Earlier works by Frühwirth {{cite:b36d339d5d14f2c63e042466da27d4b671ff2f80}}, {{cite:eff4dd5b2bc208530f4f0446fdf23d378e7662c9}} studied the time complexity of simplification rules for naive implementations of CHR.In this approach an upper bound on the derivation length,combined with a worst-case estimate of (the number and cost of) rule application attempts, allows to obtain an upper bound of the time complexity. The aim of all these works is clearly completely different from ours, even though it would be interesting to compare CHR and CHR{{formula:c58d56cb-a2fe-4264-a8e9-3dc1b3e8fff3}} in terms of complexity.", "When moving to other languages, somehow related to our paper is the work by Zavattaro {{cite:5cc4840b471ca7e79a2fbaf9a7ffce0373973025}} where the coordination languages Gamma {{cite:a96c99ab636fa9d1d610b81243d2e5a361642e02}} and Linda {{cite:8ad0ebb30e5ed1e38bf4dc3b31cd09143e17765f}} are compared in terms of expressive power. Since Gamma allows multiset rewriting it reminds CHR multiple head rules, however the results of {{cite:5cc4840b471ca7e79a2fbaf9a7ffce0373973025}} are rather different from ours, since a process algebraic view of Gamma and Linda is considered where the actions of processes are atomic and do not contain variables. On the other hand, our results depend directly on the presence of logic variables in the CHR model of computation.Relevant for our approach is also {{cite:832ce14dddd3d402116e0f5d2d5a5aa65fbe1440}} which introduces the original approach to language comparison based on encoding, even though in this paper rather different languages with different properties are considered.", "In {{cite:0212ace5f743469569d9a10d577ba5da38a887c2}} Laneve and Vitale show that a language for modeling molecular biology, called {{formula:89820e52-5aea-47cf-98ba-6f347a9f8dde}} -calculus, is more expressive than a restricted version of the calculus, called nano-{{formula:14a57b92-bbbf-4461-9940-c85c56fdc1ac}} , which is obtained by restricting to “binary reactants” only (that is, by allowing at most two process terms in the left hand side of rules, while {{formula:02415ce8-7bf5-42c4-bc58-cf9299dc1e32}} terms are allowed in {{formula:0ee3fc38-1434-4c51-8429-149bc463866d}} ). This result is obtained by showing that, under some specific assumptions, a particular (self-assembling) protocol cannot be expressed in nano-{{formula:ea732597-c7d6-4834-a83d-966a2e013b46}} , thus following a general techniquewhich allows to obtain separation results by showing that (under some specific hypothesis) a problem can be solved in a language and not in another one (see also {{cite:3124ea4bae5e225549edb5586822feb7272d9d0e}} and {{cite:c6df39c80298a0d4d34d312651096b3f52e14a0c}}).This technique is rather different from the one we used, moreover also the assumptions on the translation used in {{cite:0212ace5f743469569d9a10d577ba5da38a887c2}} are different from ours. Nevertheless, since {{formula:a18afe92-58fd-4f33-aa01-d1f67380b411}} (and nano-{{formula:d00b3aa0-4a5d-48e0-bfdc-abefb60f3ae8}} ) can be easily translated in CHR, it would be interesting to see whether some results can be exported from a language to another. We left this as future work.", "We also plan to investigate what happens when considering the translation of CHRsince many CHR implementations are built on top of a Prolog system, by using a compiler which translates CHR programs to Prolog. Our technical lemmata about CHR{{formula:f9505435-45c3-40c6-b165-1dbde43516eb}} can be adapted to what is called {{cite:28a02d36778452587091dd5042f132403e2b4489}} “pure Prolog”, that is, a logic programming language which uses the leftmost selection rule and the depth-first search. Hence it is easy to show that, under our assumptions, CHR cannot be encoded in pure Prolog. However, implemented “real” Prolog systems are extensions of pure Prolog obtained by considering specific built-ins for arithmetic and control, and when considering these built-ins some of the properties we have used do not hold anymore (for example, this is the case of LemmaREF ). Hence it would be interesting to see under which conditions CHR can be encoded in real Prolog systems, that is, which features of real Prolog (which are not present in pure Prolog) are needed to obtain an acceptable encoding of CHR. Finally we plan to extend our results to consider specific constraint theories (e.g. with only monadic predicates) and also taking into account the refined semantics defined in {{cite:dec3c5ef8eb32bddcabf00480df0634288091e84}}. This latter semantics requires further work, because it allows an improved control on computations and some properties that we used do not hold anymore in this case." ]

[]

[ "The performance of rateless coding over fading channels has alsobeen considered in {{cite:e1acb52bd30fc53945a7ac1952356c8ad27e38f4}}, in which the throughput and errorprobability are discussed. However, the tradeoff between these twowas not analyzed explicitly. For example, the results in{{cite:e1acb52bd30fc53945a7ac1952356c8ad27e38f4}} shows that increasing the value of {{formula:e7563fb9-3bcc-4aec-8386-854df07d706a}} will decreasethe system error probability in certain scenario and is thereforedesirable. In this paper we show that while this discovery is true,the system throughput, i.e., multiplexing gain might decrease when{{formula:d18362f0-fb62-40ae-b2d5-747aaff20ea9}} becomes larger for every fixed value of {{formula:5e1f79a5-eac5-469c-b0a7-1b5e83913fdb}} . Overall, ourresults reveal that the optimal design of rateless codes requiresthe consideration of both {{formula:76f687c3-10dd-440e-b12a-66d7a191a971}} and {{formula:ed90ffd4-a1e3-41a5-90ce-5dd003440611}} .", "Rateless coding may be considered as a type of Hybrid-ARQ scheme{{cite:bc6936c26b9b04eb6f52b0998a8c8589474e37cb}}. The DMT for ARQ has been revealed in {{cite:bc6936c26b9b04eb6f52b0998a8c8589474e37cb}}.However, it will be shown in the paper that this DMT curve wasincomplete and represents the performance only when {{formula:3c09e247-60eb-4311-909d-26a3e8b44a6b}} in which {{formula:9266f497-f7fd-418f-bd78-104347965b6d}} and {{formula:14a85646-61a7-431d-a0fa-291d965fedda}} are the number of transmit and receiveantennas. The complete DMT curve for rateless codingincluding those parts for higher {{formula:987693c2-de1d-4149-9e8e-e9e4fc513a96}} has never been revealedbefore, and will be shown in this paper. In addition to this, theresults in this paper also offer a relationship between the designparameter (i.e., {{formula:b462aa1c-4226-491c-83ce-5481d25ac144}} and {{formula:8a387b9d-ceb3-4d98-bcb3-b69b6cad7cc3}} ) and the effective multiplexing gain{{formula:3b9ff3ac-a8dc-4516-8397-c451077b057c}} of the system, thus offer further insights into system designand operational meaning compared to conventional coding schemes.Furthermore, we suggest new design solutions for rateless codes.Previous work on finite-rate feedback MIMO channels relies on eitherpower control or adaptive modulation and coding (e.g., {{cite:769842698ffbb4b923b64d0bacacb2daed3d5b52}}),which are not necessary for our scheme.", "The rest of this paper is organized as follows. The system model isproposed in Section II. In Section III, the DMT performance ofrateless codes is studied. In Section IV, design of specificrateless codes over fading channels is discussed. Finally,concluding remarks are made in Section V." ]

[]

[ "Dominating sets and cliques are basic structures in graphs andthey have been investigated very intensively. To determine whetherthe domination number of a graph is at most {{formula:6e5858ab-0d81-4f97-b92f-a74229fa085e}} is an NP-completeproblem {{cite:9ba7966d812db19ef75bbf10ab2877b0bdeb5d66}}. The maximum-clique problem is one of thefirst shown to be NP-hard {{cite:659740909e6d69c21cd249b1bedcb807c71d6c15}}. A well-known result of B.Bollobás, P. Erdös et al. states that the clique number inrandom graphs {{formula:447f7ac0-e727-4b14-b514-9e55ae58b9e5}} {{formula:20214961-5204-437b-afb8-65f85fac8dcd}} {{formula:6fe1f622-2339-4aa2-a7f3-932992bcb67c}} (n,p){{formula:979d0207-f3f6-437b-a7c2-52c913c45cf6}} b=1/p{{formula:dae52c94-85ef-4554-96be-dae10b2dc0c2}}{{formula:a21aacf8-e55b-4d62-aad8-3b9005e5c581}} {{formula:7e60a683-1ef8-4458-ba3c-5ae7620b7bbe}} {{formula:62277781-e187-4de4-83a9-d6862114487c}} {{formula:294a241c-5226-4b83-b085-9b58a475d394}} (n, p){{formula:5618775d-cb48-4140-99ab-5230fd3d04d3}} r1 {{formula:aea3b24b-f7aa-4dee-92cd-f0e9ff8d29b9}} r0 {{formula:2fa7a9d2-f3a4-401a-9112-de5b5e4b9c62}} The phase transition of dominating clique problem in random graphswas studied independently by M. Nehéz and D. Olejár in, and J. C. Culberson, Y. Gao, C. Antonin {{cite:67c9bba5f9d7ce0a335df5511fa16822eb7216a9}}. It was shown in {{cite:67c9bba5f9d7ce0a335df5511fa16822eb7216a9}} that the property ofhaving a dominating clique is monotone, it has a phase transitionand the corresponding threshold probability is {{formula:13a17e4a-8084-4e77-8368-c2decfa35627}} . The standard first and the second moment methods(based on the Markov's and the Chebyshev's inequalities,respectively, see , ) were used to prove thisresult. However, the preliminary result of M. Nehéz and D.Olejár {{cite:8c67026b8b63bea9c1670fe9878676e5550f11f2}} pointed out that to complete the behaviorof random graphs in all spectra of {{formula:63e308e1-0d3e-44aa-ba40-b55ee90fa53b}} needs a more accurateanalysis, namely in the case when {{formula:349a59f0-85d6-479b-bdb5-738bf86d630b}} .The main result of this paper is the refinement of the previousresults from , , . Let usformulate this as the following theorem.", "Theorem 1 Let {{formula:41ea9acc-fcdb-4424-89d2-273d91692b21}} be fixed and let {{formula:8a0d3954-1031-44e5-a103-25c6511615da}} denote{{formula:d3801cca-8119-4ff5-bd3d-a9c0985dfc48}} . Let {{formula:1dd91374-91f9-4feb-8667-0755f489fbe9}} be order of a clique such that{{formula:98596f95-0490-4b8e-b7f1-f777fd3dabb9}} . Let{{formula:8d4aab2b-d498-4786-8d6d-0fd6e7e0ac00}} be an arbitrary slowly increasingfunction such that {{formula:c1d0ea5c-95aa-4bdf-9c80-c8ee2b643596}} and let {{formula:7b1bb690-06c9-4cdb-9737-e971dde5040d}} {{formula:d19f62a2-baef-4042-a357-8df861e8861c}} (n,p){{formula:2e96f3e0-5129-4326-a4eb-7bf88ad8cc41}} To prove Theorem REF the first and the second moment methodwere used. The leading part of our analysis follows from aproperty of a function defined as a ratio of two random variableswhich count dominating cliques and all cliques in random graphs,respectively.The critical values of {{formula:cc311620-e96b-44ab-b705-1b3cbc65e504}} : {{formula:df7c126e-00b7-4971-ad4a-c0ad9bf3c659}} and {{formula:725ca8ff-2ee3-4304-8ff6-e2b2db3dd3a3}} ,respectively, are obtained from the bounds (), ()see , .", "The rest of this paper contains the proof of the TheoremREF . Section 2 contains the preliminary results. Anexpected number of dominating cliques in {{formula:cc88e648-fec3-4da6-94f6-c4892d776789}} {{formula:f86ca40c-d345-45fc-beed-ea6cca2c0285}} (n,p){{formula:c579cbde-6b55-4f6e-a70e-6cfdf5bc72c9}}" ]

[]

[ "The problem considered here isa non-interactive version of the contextual banditproblem (see {{cite:6eb28ad24a6fb344fcaba1fd90bee6cd52477ca0}}, {{cite:689104a179198faaa68c3d4a14c65fdf5bf9120a}}, {{cite:0c005a1e26e5e35f71c9585af774cddd8ce7a61c}}, {{cite:bcc4f3bf6fa97df09eac3364f08c847b0e66319e}}, {{cite:37aa6d304801f27e6cd87a446b44c21d22be99cf}} for backgroundon the bandit problem).The interactive version has been analyzed under various additionalassumptions {{cite:77e0b9a5c1d7eb7c315434413377b215860097ea}}, {{cite:7e9e21bed8a59bfe65295f97440a094afe9e6c0e}}, {{cite:4681ded22591474ae3cb40d3eed8badb31d09b9a}}, {{cite:37167e50798707bd2fedb36c991cb143c2913818}}, {{cite:89806877d1610faae89387be0e3b6046d574128a}}, {{cite:348c975efcda6c8f4ea904cefba9179afb4504c5}},including payoffs as a linear function of the sideinformation {{cite:19293209882b6c3c2dd049e434ddf44b06f11464}}, {{cite:6eb28ad24a6fb344fcaba1fd90bee6cd52477ca0}}. The Exp4 algorithm {{cite:689104a179198faaa68c3d4a14c65fdf5bf9120a}} has anice assumption-free analysis. However, it is intractable when thenumber of policies we want to compete with is large. It also relies oncareful control of the action choosing distribution, and thus cannot beapplied to historical data, i.e., non-interactively.", "Sample complexity results for policy evaluation in reinforcementlearning {{cite:faf08478600624cd920f94d759c426a4f1ea20d5}} and contextual bandits {{cite:4681ded22591474ae3cb40d3eed8badb31d09b9a}} showthat Empirical Risk Minimization type algorithms can find a goodpolicy in a non-interactivesetting. The results here are mostly orthogonal to theseresults, although we do show in section  that a constantfactor improvement in sample complexity is possible using the offsettrick.", "The Banditron algorithm {{cite:77e0b9a5c1d7eb7c315434413377b215860097ea}} deals with a similar settingbut does not address several concerns that the {{formula:e47280a7-fa52-4356-a34a-bd3b7e555a23}} addresses: (1)the Banditron requires an interactive setting; (2) it deals with aspecialization of our setting where the reward for one choice is 1,and 0 for all other choices; (3) its analysis is further specializedto the case where linear separators with a small hinge loss exist; (4)it requires exponentially in {{formula:26cbb4da-5b96-47d5-a155-1887faedd2ed}} more computation; (5) the Banditronis not an oracle algorithm, so it is unclear, for example, how tocompose it with a decision tree bias.", "Transformations from partial label problems to fully supervisedproblems can be thought of as learning methods for dealing with sampleselection bias {{cite:560bbbba8817b6e5365a8406ef03ec3fe9a72667}}, which is heavily studied in Economics andStatistics." ]

[]

[ "In this section we apply our results to light moving across a 2Dsquare array of infinite square dielectric prisms with diagonalsaligned with the sides of the square primitive cell, a systempreviously proposed byMilton etal.{{cite:52926aec4898d29c9bee9b02f7fa7ba5e50188b1}}We chose the parameters{{formula:8369d940-a55b-42f7-b7b3-6aca3dec1d0b}} ,{{formula:54910a1e-fa6c-4ba0-b5f1-317dfe34d5fc}} , and {{formula:7370abbe-37ff-418b-820b-a37b0a04cbf8}} . We takea finite free-space wavelength {{formula:ea3d4864-3f41-4fba-80e8-99bc9900cdbc}} , with {{formula:7e310fd5-64a0-4b16-a7dc-784734d9c811}} the lattice parameter, sothat, according to Eq. (REF ) we expect only small retardationeffects of the order of {{formula:d728b511-1a63-4b0a-b4d1-fa7890ad9817}} .We choose the polarization normal to the prisms axis so that in ourlocal limit the system is effectively isotropic in 2D.We truncated our matrices in reciprocal space by setting a maximumvalue {{formula:e19964af-8929-4d36-9161-d4bdd1cdae60}} for the magnitude {{formula:9f476dcd-4cb5-424a-805b-ef673b9e16c2}} and {{formula:f27024be-95c8-4698-8450-dfadf83dce88}} ofthe components of the reciprocal vectors, so for a field polarizationwithin the planethe numberof rows and columns for the matrix {{formula:b5a262fe-d56e-4a61-b29d-f92dfab7614b}} inEq. (REF ) is given by{{formula:450333d2-b7a7-4580-8bc4-2916bf8db2ad}} .To test the convergence of our computational procedure, inFig. REF we show our results for {{formula:88111eac-370b-4311-8ff0-fbd7966eb978}} as a functionof the maximum index {{formula:e43d95d1-fcf4-4e9a-96fb-ca18d0311e3b}} .{{figure:56fc6a11-490e-4bf7-8ea6-dd70bbad2e1d}}", "FromFig. REF we see that {{formula:0156805a-6c34-4855-8c32-7e3ca8981a30}} converges approximately as{{formula:b8b64f9d-9aa4-4183-b767-4ba92472487f}} , andvalues of the order around {{formula:6d3f297e-f2a8-462c-a732-7b1c39270854}} are needed to obtainan accuracy better than 0.5% without extrapolating,yielding large matrices of more than {{formula:c63b5fe3-ad17-4f1c-bb97-d4c460f41c8d}} elements. In the same figure we have indicated the response obtainedby Milton etal.,{{cite:ab65a242db4b6f96a28d39ebd3bb2a5444d0b59f}}, Tao et al.,{{cite:1b574e39669250de7304ba22f9731f0afc00e3c3}} and Bergman etal.{{cite:88152c0954aecc4c84ca086e0f49ddbcc540098f}}which studied the same composite. As we see, linear extrapolation ofour results towards {{formula:73058fc9-9975-4f4b-85c5-9149b8cb83d7}} converge tothose Bergman et al. and of Milton et al., whereas theresult of Tao et al.differs slightly.Finally, we also compare our results with those of mean-field theory,as embodied in Maxwell-Garnet's (MG) formulae{{formula:365df2ba-c08c-4802-b852-b4447a7761ca}} ", "with {{formula:435cd4f5-62c3-48ab-86c5-3cbfa01a0979}} for our 2D system.{{cite:d5ff77e7414f9709423e1dcd22b134da7c244178}}As we see in the figure, the MG results differ from ours and the otherauthors' results, mainly due to its intrinsic limitations.{{cite:d5ff77e7414f9709423e1dcd22b134da7c244178}}We havechecked our results with other set of parameters also reported bythe same cited authorsand we have obtained similar agreementas mentioned above. The rate of convergence of our method is similarto that reported in Ref. Sozuer(1992) when written in terms of{{formula:6214eff1-0ae0-42ee-9092-46d2101d8b4b}} .", "We can also test the convergence of our results above using Keller'stheorem,{{cite:06e170ba7bca2feefb575b966203ac9679dd8c9d}}, {{cite:b1c4d273be490d81ea7da6f6be6d075c5266df45}}, {{cite:8717ac29b78884c48971e69127db1fe32e756cb2}} which we maywrite as{{formula:f33085b1-39aa-4648-87ae-8683d4fbb6bc}} , where we define Keller's coefficients along principalaxes {{formula:1014a318-4eca-4db3-a6df-11fd97d23752}} as {{formula:cb051eac-1d42-4ec5-a904-69ef47f174ef}} and{{formula:b023e1b3-872e-48e7-807b-c0f60e67752b}} . Here, we introduced themacroscopic response {{formula:b5d7395c-2989-417c-979d-4279f5fdf27d}} and {{formula:1805adad-0d01-496e-9112-2bbae3bb0810}} ofthe reciprocal system that is obtained from the original system byinterchanging{{formula:d0547285-3a03-409b-aa45-31377de6dad6}} . Indeed, for our isotropic systemwe expect {{formula:e5ec8a65-db7a-405c-b10d-5cae2f9841f5}} as {{formula:158a5dce-2a1d-43ef-8b6c-229ed3c940a6}} .In Fig. REF we show {{formula:aa31f525-8315-4c11-a5dd-eb3b5ae2ab7d}} vs. {{formula:ce0ca4fc-adce-4e01-984f-caf6b740bef5}} forthe system corresponding to Fig. REF .{{figure:c4525403-a1bf-4238-be0d-9c94ee53ebc5}}", "We see clearly that {{formula:289296fa-2a1f-473b-8787-1199956c8cb7}} decreases linearly in{{formula:e26cfe9c-3d3f-4719-888a-847fe2a4eb1b}} . However, its extrapolation towards{{formula:c8c8acc8-9fd1-4afc-9a14-286c6b5b5697}} does not attain the value {{formula:b08b6902-fd65-491d-a41a-a22d00794503}} asexpected from Keller's theorem. The reason for the small discrepancy isthat our calculation includes retardation effects which we expect tobe of order {{formula:4f7a46af-81f8-4ad9-8937-09362ce99e3d}} , while Keller's theorem is strictlyvalid only in systems with no retardation. To confirm this statement,we also display in Fig. REF the results of a calculationfor {{formula:18417c20-ea68-4c46-b145-49199f6cb557}} , showing that in this case, the discrepancybetween the extrapolated and the expected value is negligible. Thus,we have verified that our calculation is consistent with Keller'stheorem inthe absence of retardation and has an error that goes to zero as{{formula:793cb05a-cb79-4706-9535-09000bf3595f}} when{{formula:75ee6de6-fddc-4319-976b-dbb187deaab6}} .{{figure:c32ea90d-8ac0-4322-8429-aaf84875a5f5}}", "In Fig. REF we show nearly converged (error {{formula:273c9d2c-73f2-4029-ac22-1375d2a9dd2b}} )results for {{formula:98bf1390-4f8a-4fe0-af59-b253573257d3}} as a function of the filling fraction {{formula:1fa6222d-8346-43bf-8d4b-9931e2ab035a}} for the same system as inFig. REF . We can see again an excellentagreement of our results with those obtained by Milton et al., and Bergmanet al., and, to a lesser extent, with those of Tao et al.The MG results are noticeably lower, with a discrepancy that increaseswith filling fraction.We have obtained results identical to ours in Figs. REF andREF using Eq. (35) of Ref. Perez(2006),confirming that our formalism is equivalent to that of Halevi andPérez-Rodríguez. In conclusion, our approach does indeedreproduce the resultsof other works, and thus we have validated our numerical scheme andcan be confident on the accuracy of our results." ]

[]

[ "The problem of social contagion or spread of influence was firstformulated by the sociological community, and introduced to the computerscience community by Domingos and Richardson {{cite:8d636642ad9affa5f2c58dc07f1c4f4f800e829d}}. An influentialpaper by Kempe, Kleinberg and Tardos {{cite:b653f32aa3d87a157a01e4bce8a4ccec7ed23d09}} solved the target setselection problem posed by {{cite:8d636642ad9affa5f2c58dc07f1c4f4f800e829d}} and sparked interest in this areafrom a theoretical perspective (see {{cite:530992e480d8997822ba0fd8019c57a2b584e422}}). This work hasmostly been limited to the influence maximization paradigm, whereinfluence has been taken to be a proxy for the revenue generated through asale. Although similar to our work in spirit, there is no notion of price in thismodel, and therefore, our central problem of setting prices to encourage influencespread requires a more complicated model.", "A recent work by Hartline, Mirrokni and Sundararajan {{cite:795eca3eca67497af38f7a6bf2e81ad9599bf919}} is similarin flavor to our work, and also considers extending social contagion ideaswith pricing information, but the model they examine differs from our modelin a several aspects. The main difference is that they assume that the selleris allowed to approach arbitrary nodes in the network at any time and offertheir product at a price chosen by the seller, while in our model the cascadeof recommendations determines the timing of an offer and this cannot bedirectly manipulated. In essence, the model proposed in {{cite:795eca3eca67497af38f7a6bf2e81ad9599bf919}} is akinto advertising the product to arbitrary nodes, bypassing the networkstructure to encourage a desired set of early adopters. Our model restrictssuch direct advertising as it is likely to be much less effective than adirect recommendation from a friend, especially when the recommender has anincentive to convince the potential buyer to purchase the product (forinstance, the recommender might personalize the recommendation, increasingits effectiveness). Despite the different models, the algorithms proposed byus and {{cite:795eca3eca67497af38f7a6bf2e81ad9599bf919}} are similar in spirit and are based on an influence-and-exploit strategy.", "This work has also been inspired by a direction mentioned byKleinberg {{cite:530992e480d8997822ba0fd8019c57a2b584e422}}, and is our interpretation of the informalproblem posed there. Finally, we point out that the idea of cashbacks hasbeen implemented in practice, and new retailers are also embracing theidea {{cite:a1c13860929510447ff3ecad8d57983bcf478ff6}}, {{cite:4771c1c084e71f63440063a0de804504f0a10d12}}, {{cite:0a118da4bb9c93dbdb468c64afbb1bf78a7558b4}}. We note that some of the systems beingimplemented by retailers are quite close to the model that we propose, andhence this problem is relevant in practice." ]

[]

[ "A brief account of MaxSAT and PBO solvers is provided inSection .The use of unsatisfiability for solving MaxSAT was first proposed in2006 {{cite:82e2f97aac12b1665eed2913be92883d27953f01}}. This work was laterextended {{cite:e53315d5f7bbf2526e648c846e2b5f6b1adf882d}}, {{cite:ccc28352aae05aa512346bf1edee35c57fdee544}}, {{cite:439276b3910d7239a8aa5bc03c3bd90fde00c614}}, to accommodateseveral alternative algorithms and a number of optimizations to thefirst algorithm.To the best of our knowledge, MSUnCore is the first algorithm forsolving (Partial) Weighted MaxSAT with unsatisfiable sub-formulaidentification. Also, to the best of our knowledge, WBO represents anew modeling framework, and the associate algorithm is new.", "The use of optimization variants of decision procedures has also beenproposed in the area of SMT {{cite:0cce8452034b4602c27cd7bd49e595b3b4a18d5a}}, and a few SMTsolvers now offer the ability for solving optimization problems. Theapproaches used for solving optimization problems in SMT are based onthe use of relaxation variables, similarly to the PBO approach forsolving MaxSAT {{cite:432f86f940d5afeb301d4405b5033749a14d177a}}." ]

[]

[ "The best known deterministic and randomized algorithms for generalinstances have competitive ratios at most{{formula:d4dc3297-f1ae-458b-9458-b8d8e4738269}}  {{cite:0b4176c20a007af53f0b51f46439fe1fe4e46c8f}} and{{formula:e0a00853-b7a6-44b3-99b3-2a4b4df0b514}}  {{cite:044891f32e11d937d3d57bc4c756a34913ccf7b3}},respectively. A recent analysis of the latter algorithm shows that itretains its competitive ratio even against adaptive-onlineadversary {{cite:44a91652adee7c0265acafb1494c09e4334f24a4}}.", "The best known lower bounds on competitive ratio against either adversary typeuse rather restricted 2-bounded sequences in which every packet haslifespan (deadline {{formula:d833deb1-0642-4ac5-98d6-c53e41a364dd}}  release time) either 1 or 2. The lower bounds inquestion are {{formula:f8de19b6-4fc9-4218-a531-9c88ac57e28e}} for deterministic algorithms {{cite:a38da7f30309202a69301a7dd1c28d83150afe05}},{{formula:6a8887f6-360c-44a0-8ed7-c4046f3f6741}} for randomized algorithms against adaptiveadversary {{cite:7be71c77299741c8aa07a897fdf510c6eff6b8ac}},and {{formula:26a8e91a-7676-45c0-9a79-4c13d5db410f}} for randomized algorithms against obliviousadversary {{cite:f58b49b89a42eca622ae16fc3b64c7b24f2c0c56}}.All these bounds are tight for 2-boundedsequences {{cite:6bb9a820513ba45ece3bc200feb842d10c3d1a68}}, {{cite:7be71c77299741c8aa07a897fdf510c6eff6b8ac}}, {{cite:044891f32e11d937d3d57bc4c756a34913ccf7b3}}.", "We restrict ourselves to sequences with agreeable deadlines, in whichpackets released later have deadlines at least as large as those released before({{formula:1088f929-7a8f-4a5c-86da-1a4fcfcc6949}} implies {{formula:e0d6ef22-5ed0-40e1-8a4f-13049b848cb8}} ). These strictly generalize the2-bounded sequences. Sequences with agreeable deadlines also properly contains-uniform sequences for all {{formula:15469d47-4b6f-404d-8b9f-8d014ec396a6}} , i.e., sequences in which every packet haslifespan exactly {{formula:7e4acfa4-33f4-40db-b068-52dce99060f2}} . An optimal {{formula:52ff7019-4160-4f41-8b08-2f364ee07ce0}} -competitive deterministicalgorithm for sequences with agreeable deadlines is known {{cite:7dea4b903be4da72f90bb017b8160de130d3a1ab}}.", "Jeżabek studied the impact of resource augmentation on the deterministic competitiveratio {{cite:a51cea2163dabf75b46ab3f2885d31f9c393b8cb}}, {{cite:25b87699de3772eee7955bab7305c1d6dfbe4086}}. It turns out that whileallowing the deterministic algorithm to transmit {{formula:bfe7bbe0-50c2-485f-800f-6ff23e92f3d1}} packets in a single step forany constant {{formula:fd9c56c5-984b-4ca5-951c-302c9ecd9ead}} cannot make it 1-competitive (compared to the single-speedoffline optimum) on unrestricted sequences {{cite:a51cea2163dabf75b46ab3f2885d31f9c393b8cb}},{{formula:98654d20-eb69-436b-ad6f-3bdf724dd338}} is sufficient for sequences with agreeable deadlines {{cite:25b87699de3772eee7955bab7305c1d6dfbe4086}}." ]

[]

[ "Formal methods for reasoning about events based on explicit representation of events date back at least to work on situation calculus {{cite:fbcf4e17033c4c16735702394a6a3b0cf80fef8a}}. Situation calculus treats situations as snapshots of the state of the world at some time instants. Actions change one situation to another. These actions are instantaneous, have no duration, and have immediate and permanent effects upon situations. Another formal method is the event calculus {{cite:17e86007b6487fa8b17492a10ba747456c5eef6e}} which explicitly represents events (including actions) that belong to an event type and generate new situations from old ones. Predicates in event calculus are defined over fluents which are time-varying property of a domain, expressed by a proposition that evaluates to true or false, depending on the time and the occurrence of relevant events. Predicates on fluents include: Occurs(event,time), HoldsAt(fluent,time), Initiates (event,fluent,time), Terminates (event,fluent,time).", "Reactive systems (including active databases and large system monitoring applications) also explicitly store and reason about events. Here, an event is defined as a system generated message about an activity of interest and it belongs to an event type (situation). In addition to representing events, these applications are also interested in detecting the occurrence of events (e.g. {{cite:f4d37d09e1418e51490a0c98f4818cdc5892ca91}}, {{cite:e9f176145f54537c27acd2cba38b7ee726938b63}}, {{cite:fd9a7e4065f578f7750eac4ad781fcaf55b0220f}}). Besides detecting and storing primitive events, these applications detect and represent composite events which are defined as some sequence of primitive events using an event algebra (e.g. {{cite:85f2276873c4aacf49ecb0127a1b8dec31081b90}}). As in situation calculus, an event is considered to occur at a precisely determined point in time and has no duration. Although composite events span a time interval they are typically associated with the time-point of the last component event. However, events can be mapped to time intervals to apply queries over the duration of the sequence that a “compound composite\" event matches. All event types consist of core attributes like time point of occurrence, event identifier, event type label, event source identity, and so on. There are a number of commercial products that support such applications including IBM's Tivoli Enterprise Console (and its Common Base Event Infrastructure) and iSpheres EPL Server.", "Our focus is different from this body of work in several ways, namely:(1) Events in reactive systems are well-defined structured messages with restricted variations. In contrast, real-world events are communicated in diverse formats like text, video, audio, etc., (2) Since we deal with real world events, we consider spatial aspects of events which are not dealt with in reactive systems, (3) Information about real-world events can be much more imprecise (as it is derived from potentially noisy source like human reports) and much more complex. (4) Relationships between events (e.g. causation) in reactive systems are typically strong and easier to detect due to the static nature of the environment (system configuration) in which the events occur. Real-world events have weaker relationships and include temporal, spatial and domain relationships.", "Recent work in video content representation has also considered events as foundations of an ontology-driven representation {{cite:a7f55d9e2e666072ff259c173317a5a2e6e57cc4}}, {{cite:508f9012acc049cc1f72f4a4e716e759bebe5169}}, {{cite:f98b4543851dc5fd3d700cdc1fff589f5d6168ad}}. The goal of this body of work thus far has been on producing a video event mark-up language that can facilitate data exchange and event recognition. As in situation calculus, an event is defined as a change in the state of an object. A state is a spatio-temporal property valid at a given instant or stable within a time interval. Events can be primitive or composite. Primitive events are state changes directly inferred from the observables in the video data. Primitive events are more abstract than states but they represent the finest granularity of events. As in situation calculus {{cite:fbcf4e17033c4c16735702394a6a3b0cf80fef8a}}, time is the critical distinguishing factor between states and events. For example, two identical states with different time values represent two different events. A composite event is a combination of states and events. Specifically, a composite event is defined by sequencing primitive events in a certain manner. This sequencing can be single-threaded (single-agent based) or multi-threaded (multi-agent based). Events, states and entities can be related to each other using predicates. Spatial and temporal relationships are defined as predicates on members of the time and space domains linked to events. In general this body of work is object-centric, i.e. assumes knowledge of objects precedes knowledge of the event as it defines events as changes in object states. As discussed in Section , we adopt an event-centric approach. Besides, the constructs in this body of work are tailored to automatic recognition of events from video while we focus on facilitating queries on event data.", "Event-oriented approaches have also been studied in spatio-temporal data management. The goal here is to represent events associated with geographical/spatial objects. As noted in {{cite:1af6b0a969cff6dac0b9d7e86ab4bda9df19c918}}, the effort on spatio-temporal event representation has evolved in three stages: (1) Temporal snapshot of spatial configurations of events, (2) Object change (captured in terms of change primitives such as creation, destruction, appearance, disappearance, transmission, fission, and fusion) stored as a sequence of past states, and recently (3) Full-fledged representation of changes in terms of events, actions (initiated occurrences), and processes. An example of stage 1 representation is [11] where, starting with an initial state (base map), events are recorded in a chain-like fashion in increasing temporal order, with each event associated with a list of all changes that occurred since the last update of the event vector. The Event-based Spatio-Temporal Data Model (ESTDM) {{cite:e42b9e449fdf8fa74c96699cf31927828ea91a0a}} is an example of stage 2. ESTDM groups time-stamped layers to show observations of a single event in a temporal sequence. The ESTDM stores changes in relation to previous state rather than a snapshot of an instance. An event component shows changes to a predefined location (a raster cell) at a particular point in time. The SPAN ontology {{cite:dc13e4e74162c161e5741a0f3b89e548587bd788}} that defines an event/action/process view and the process calculus based approach of {{cite:1af6b0a969cff6dac0b9d7e86ab4bda9df19c918}} that can also represent event-event relationships are examples of stage 3. Basically, in stage 3, rather than the sequence of past states of each object, the events that caused the state changes are modeled resulting in a a more richer representation. As such, stage 3 can: (1) tell us “why” a state exists, and (2) enable us to represent which event caused a state change when multiple events (or sets of events) can potentially cause the same state change." ]

[]

[ "Probabilistic databases have been an area of activity since the 1980s with foundational works such as {{cite:040bad85e9d0ad79104fd8092475e9305d15370b}}, {{cite:79d5898ad1fdddbc324cb5b15882a9a279078a56}}extending the relational model and algebra to represent and support uncertainty in databases.Current active projects - MystiQ{{cite:3383674b718762f710e3333824e2d5148261cd6a}}, MayBMS, Trio, or Orion {{cite:f1a2652c0137c0a38899129be01d98a11c85b92a}} employ different underlying uncertain databaserepresentation formalisms that either vary subtly, or in some cases significantly across each other, for instance MystiQusing \"or-tuples\", Trio using or-sets but with additional \"lineage\" information, and MayBMS using more expressiveworld set decompositions (WSDs). MayBMS has considers conditioning probabilisticdatabases with ICs which is motivated from a data cleaning perspective, dealing with \"equalitygenerating dependencies\" (equivalent to the tuple level ICs) and just functional dependencies (FDs)from amongst relation level ICs (as opposed to the larger class of relation ICs that we address).Their approach to resolving ICs is quite different from ours. Instead of applying ICs to an uncertain database as wedo, they augment queries with the ICs so that the ICs are resolved at query time.The approach to factor in FDs using a chase based procedure {{cite:19b80c2ff519ab951ac7e833cd691ceebc788767}} can result inan exponential blow up even with a single FD. Each relation is represented as decomposedinto multiple \"components\" the product of which yields the entire relation. Each component essentiallycontains the values of an attribute or a set of attributes.Their algorithm is to consider pairs of tuplesviolating the FD, take each attribute in the FD and merge the components containing those attributesfor the pair of tuples into a new component, and then clean the new component by eliminatingattribute value combinations that are inconsistent with the FD. In the case of a relation R,with FD {{formula:fb660feb-7d88-47db-ad47-391cb3de8964}} , and pair-wise violations {{formula:b6220b84-3b72-4495-823d-b53770869340}} with thisFD, we will end up with a component that has as columns ({{formula:4b48f3e0-10c0-40e0-a601-e07df7cc18db}} ) andin the rows of this component have all consistent combinations of attribute A and B values.The size of this component is {{formula:9c213fb4-e7e8-4b42-b6b7-0f41d2d16afb}} where M is the degreeof uncertainty (choices) in the attributes. Further, the chase based procedure must select the consistentcombinations only and its compelxity is also {{formula:120c1064-4ae4-4944-b684-c6cea3697f44}} . Even with modest values of say M=2 and K= 30, {{formula:1378af6e-8379-43bd-96f7-b3fb1bed689c}} is extremely large.While we observe that their approach is exponential, we note that the authors essentially meant the technique to be used in the context of data that has onlyvery few violations, in which case their approach will work fine. This is substantiated by their experimentswhich have been done with a degree of data dirtiness as low as 0.001% - 0.005% and also stated as a validassumption by them given the focus on data cleaning applications. In contrast, our approach is applicable todatabases with a much higher degree of data dirtiness, for applications such as information extractionwhere literally all fields in the data can be uncertain i.e., with a degree of dirtiness of 100% !Also in our synthetic data evaluation we have used an {{formula:4a90a3ac-a4e3-46d5-82ca-c88a337a30c6}} (dirtiness) factor of at least 5% (Table REF ).To the best of our knowledge our work is the first to a) Provide an approach to factoring a large class of ICs, including many kinds of relation level ICssuch as FDs, aggregation constraints, inclusion dependencies, and set constraints in a correct manner into an uncertain database, b) Provide an approachto incorporating ICs that makes no assumptions on factors such as the degree of data dirtiness and is thus applicable to applications where the degreeof data dirtiness can in practice be quite high.", "In information extraction, the approach developed in {{cite:8e45657fc97cfbbd9101d55a304af15b3dae5361}} is to approximate a complex CRF distribution that represents text segmentation possibilities into a probabilistic relational model. This work however does notconsider any dependencies across different extracted segments, where each extracted segment is treated as a tuple.We address such dependencies as relation level ICs. In {{cite:8e45657fc97cfbbd9101d55a304af15b3dae5361}} the probability distribution beingapproximated is known to be generated from a CRF and an efficientforward-backward-message-passing algorithm is employed for marginal computation, vs our setting where marginal probabilities must be estimated.We compared with database repair {{cite:8b43871a9f40ad7ed78312934dd968781680fec5}}, {{cite:6008d555049a832c3f7b4147fc7c1bf512088099}}, {{cite:079bae6aa2d76fb92ae942bb03b575f4fa5227b3}} earlier and further note that most prior work on repair has considered only a limited set of constraints, such as {{cite:6008d555049a832c3f7b4147fc7c1bf512088099}} which deals with only functional (FD) and inclusion (IND) dependencies whereas our paper addresses a large class of attribute, tuple, and relation level ICs. Workon consistent query answering (CQA) deals with a related but different problem of answering queries over a dirty database considering constraintsover the database - this is established as a hard problem in general {{cite:079bae6aa2d76fb92ae942bb03b575f4fa5227b3}} with practical approaches {{cite:d68fb683c8b06778da590cd07ebd3a1c01bd1ece}} providedconsidering only primary key constraints." ]

[]

[ "The goal of adding a temporal aspect to web navigation has been explored in projects that focus on user interface enhancement. The Zoetrope project {{cite:4ae2df43555fbcd9f6f3843f4b6aee904e322e58}} provides a rich interface forquerying and interacting with a set of archived versions of selected seed pages. The interface leverages a local archive that is assembled by frequently polling those seed pages. The Past Web Browser{{cite:088df1174b81752ed81450ff7f4bb9e3ce18fe24}} provides a simpler level of interactionwith changing pages, but it is restricted to navigating existing web archives such asthe Internet Archive. And DiffIE is a plug-in for Internet Explorer that emphasizesweb content that changed since a user's previous visit by leveraging a dedicated client cache {{cite:4bc99b5376dea85d15c53cc7a738f14d5c5d1069}}. None of these projects propose protocol enhancements but rather use ad-hoc techniques to achieve their goals. All could benefit from DT-conneg as a standard mechanism for accessing prior representations of resources.", "Some projects have dealt with the problem of disappeared web pages and finding archived or replacement copies on the Web. The use of lexical signatures as search engine query terms was proposed as a way to find content that had moved from its original URI {{cite:dac3422954601cea5934b5f2f269946f7f5d22b8}}, {{cite:894219d29daef6941f58931c3cd9b24efafd7445}}. This approach was later applied to search for content in web archives {{cite:194fefba72c560a1530be74a03d9a44da26536cc}}, {{cite:ed56057cdccf6ab53e970b75f24584ffb3eb3b07}}. Also, when a “HTTP 404 Not Found” occurs, the ErrorZilla FireFox plug-inhttps://addons.mozilla.org/en-US/firefox/addon/3336 presents a user with a search page allowing her to find disappeared pages in web archives, and the UK National Archive's server plug-in redirects the client to an archive of its choice. As suggested (Section ), in the DT-conneg framework a client could intelligently react to 404s, and when doing so leverage available re-finding approaches.", "To the best of our knowledge, very little research has explored a protocol-based solution to augment the Web with time travel capabilities. TTApache {{cite:79a23e7897cb11d269daa89c40fe5dab1ef1bfb8}} introduced a modified version ofApache that stored archived representations in a local transactional archive(similar to the configuration illustrated in Figure REF ).Ad-hoc RPC-style mechanisms were used to access archived representations given the URI of their original, e.g. “page.html?02-Nov-2009” and", "“page.html?now”. This approach reveals the local scope of the problem addressed by TTApache, as opposed to the global perspective taken by the proposed DT-conneg framework. Indeed, the query components are issued against a specific server, and are not maintained when a client moves to another server as is the case with the X-Accept-Datetime header of DT-conneg. TTApache also allowed addressing archived representations using version numbers in query", "components rather than datetimes. This capability is similar to the deprecated “Content-Version” header field from RFC 2068 {{cite:a176598f79057036aefa0cfc135dd275f97e385b}} and other, similar expired proposals (e.g., {{cite:cf95370d6d3adffc4925cd7a1eedaf77657096db}}). Such versioning features have not found wide-spread adoption, presumably because their address space is tied to a specific resource or server, and not universal like the datetime of DT-conneg." ]

[]

[ "This section is devoted to review the related literature.FS can be classified as a Markov Chain Monte Carlo (MCMC) method.Other MCMC-based methods have been applied to characterize complex networks.Applications include, but are not limited to estimating characteristics of a population {{cite:2ec11f040a837d762296ce021d6f6b2e1aad90d4}} (e.g. estimation of HIV seroprevalence among drug users {{cite:d3f734b2bf2e028b28c28584e7b697b88e9e5b57}}), content density in peer-to-peer networks {{cite:63284a805726f9012f8504a186629695e4908bd9}}, {{cite:14142d826d92da6059f2d3e90fb48517e2623a96}}, {{cite:f02496846070308c8f494590d1d01543437abe1a}}, {{cite:94003f891c9ec9fde2aa98b171fcdd554e85144a}}, uniformly sampling Web pages from the Internet {{cite:e8dd71c724283a14348d57088bc2a07ac1782fdb}}, {{cite:92f20a41a86d511c13c1331990819229b8b4dfa9}}, anduniformly sampling Web pages from a search engine's index {{cite:75a1e3948c760a0d185e063e5dfe6399495e6e93}}.The above literature is mostly concerned with random walks that seek to sample vertices uniformly (also known as Metropolized Random Walks or Metropolis-RW) {{cite:63284a805726f9012f8504a186629695e4908bd9}}, {{cite:e8dd71c724283a14348d57088bc2a07ac1782fdb}}, {{cite:92f20a41a86d511c13c1331990819229b8b4dfa9}}, {{cite:75a1e3948c760a0d185e063e5dfe6399495e6e93}}, {{cite:94003f891c9ec9fde2aa98b171fcdd554e85144a}}.The accuracy of RW and Metropolis-RW (MRW) is compared in {{cite:bba33f242f24bb35c2d39da418a825c2d1cf4cfe}}, {{cite:f02496846070308c8f494590d1d01543437abe1a}}, and in a variety of experiments RW estimates are shown to be consistently more accurate than or equal to MRW estimates.", "The above literature does not consider the use of multiple random walks to address the problem of estimating characteristics of disconnected or loosely connected graphs.While multiple independent random walkers have been used as a convergence test in the literature, our simulations in Section  show that independent walkers are not suited to sample loosely connected graphs when the starting vertices are selected uniformly at random.", "A number of real complex networks are known to have disconnected or loosely connected components.A large body of MCMC literature is dedicated to overcome the locality problem described in Section REF .However, the literature either assumes that the graph is very structured, e.g., a 2 dimensional lattice, or that the graph is completely known. These assumptions make the solutions inapplicable to our problem.A comprehensive list of MCMC methods and their characteristics can be found in {{cite:e71430ab930af03df04a57af291baf84f45f1d08}}.", "Projecting a RW onto a higher dimensional space has been used in {{cite:edd2459a93cf99942cd5d77d463f9f7148a8366b}} to make the Markov chain associated to the random walker nonreversible, which can speed up the mixing of the original RW.Unfortunately, it is unclear if this method can be successfully used to estimate characteristics of complex networks.", "In networks that cannot be crawled (e.g., the Internet topology), samples must be obtained along shortest paths, and vertex degrees cannot be queried, {{cite:b840b7e397171773db49d662ad2b2fe9673f8c54}} shows that observed vertex degrees are biased.Our work, however, assumes a graph can be crawled and vertex degrees queried.Our scenario admits a RW with an unbiased estimator.Multiple random walks also find other applications besides the one presented in this work.They are used to collect Web data {{cite:b43b2141ee801aa7999232ba48c45eba94ecc552}}, search P2P networks {{cite:71161fc20cb9548e596e16bcc801b539998d0fb5}}, {{cite:d4b6ff006046dc80e9d5aa3d57619eaaef2cc147}}, and decrease the time to discover “new wireless nodes” {{cite:aca5dfa7c22cc719b632d6fec29773488f91751d}}.Dependent multiple random walks are also used in percolation theory {{cite:65a60365376c29866b57ec95f8b5544f2a923681}}." ]

[]

[ "Many CIDNs were proposed in the literature, such as Indra {{cite:2a0becc7507bf7f23c613606499e1fe44e19e5fb}}, DOMINO {{cite:41f9955089ddb91f28303121b7a7b3dec4b9fbca}}, and NetShield {{cite:3506b9cef39920d91a4a6171b350453ed01cd494}}. However, these works did not address the problem that the system might be degraded by some compromised insiders who are dishonest or malicious.", "Simple majority voting {{cite:4b22d410230b9577268c43712f7f08f3da2844ea}} and trust management are commonly used to detect malicious insiders in CIDNs. Existing trust management models for CIDN are either linear as in {{cite:91bedd2bc87b20983da37b19473bd0b68f813718}}, {{cite:461518c8c57fd5cc1e93631a165a2edfe5f6d0a9}}, or Bayesian model as in  {{cite:65f8769a30456508489fe78e75aced95b225af40}}. They are based on heuristic where the feedback aggregation is either a simple average  {{cite:9b5ba57845f940c3f62bc3396596563baac40b97}} or a weighted average {{cite:65f8769a30456508489fe78e75aced95b225af40}}. Moreover, no decision cost is considered in these models. In this paper, we use a sequential hypothesis testing model aiming at finding cost-minimizing decisions based on collected feedback. Existing work that applies hypothesis testing for intrusion detection includes{{cite:8dcfc71fde026ba4487fb6a58473775b5b560369}} and {{cite:ab359632c7541b24ce646959887fbb41ad768364}}, where a central data fusion center is used to aggregate results from distributed sensors in a local area network. However, their methodologies are limited to the context that all participants need to engage in every detection case. While in our context, IDSs may not be involved in all intrusions detection and the collected responses may be from different groups of IDSs each time.{{figure:9a96e506-b0cf-4295-949d-c64e77d122ba}}" ]

[]

[ "Erdös and Rényi have proposed a simple model that generates equilibrium random graphs, calledErdös-Rényi random graphs {{cite:514f35540db7c39d6416f724d53c8fc28e5fdd3d}}.In this model, given a fixed number of vertices, a graph is constructed by connectingrandomly chosen vertices with an edge repeatedly until the predetermined number of edges areobtained. This model is restrictive in that it produces only Poisson degree distributions.", "Dorogovtsev et al. proposed a model that can generate graphs with fat-tailed degreedistributions {{cite:0301e97893b07c9d15557af160b936fae38e5ae3}}.Given a random graph, this model restructures the given graph byrewiring a randomly chosen end of a randomly chosen edge to a preferentially chosen vertex andalso moving a randomly chosen edge to a position between two preferentially chosen verticesat each step of the evolution.", "The model proposed by Watts and Strogatz {{cite:69c8408d866960664185df2a74b54ff4b9938037}} generates random structureswith small diameter, which has been named as small-world graphs. This model transforms aregular one-dimensional lattice (with vertex degree of four or higher) by rewiring each edge,with certain probability, to a randomly chosen vertex. It has been found that, even with the smallrewiring probability, the average shortest-path length of the resulting graphs is of the order ofthat of random graphs, and generate graphs with fat-tailed degree distributions.", "The majority of recent models uses a method called preferential attachment {{cite:13bd0ef03de63b67546412163aefc6e173df72d9}}.In a representative model among these, proposed by Barabasi and Albert {{cite:3b8fa15978ec81913ffe77d0837dabc44cd1e08c}}, a new vertexjoins the graph at each time step and gets connected to an existing vertex with probabilityproportional to the vertex degree. With preferential attachment, these models canemulate the dynamic growth of real graphs.", "Leskovec et al. {{cite:6bced70bfe7bcacbf29c359b00456b45c3c2514e}} have proposed a graph generation model that addressessome of recently discovered properties of time-evolving graphs: densification andshrinking diameter. The main idea of their model is to recursively create self-similar graphswith certain degree of randomness. The self-similarity of the graphs is achieved by usingthe Kronecker product (also known as tensor product) {{cite:f65b26e2797498aa5dbacea3bc942fc55abaf6b2}}, which is a naturaltool to construct self-similar structures.Given a seed graph, at each step this model computes theKronecker product of two matrices that represent the seed graph and the graph generated in previousstep respectively.The graphs generated with this method have regular structure. The model changes the entries inthe target matrix with a certain probability before each multiplication toadd randomness to the graph." ]

[]

[ "There has been extensive research on fast similarity search due to its central importance in many applications.For a low-dimensional feature space, similarity search can be carried out efficiently with pre-built space-partitioning index structures (such as KD-tree) or data-partitioning index structures (such as R-tree) {{cite:ca838cb460151855f7e446ad7cef0b2c345b4f74}}. However, when the dimensionality of feature space is high (say {{formula:5065f3a9-0e61-4fac-abf2-9655e9a3c93f}} ), similarity search aiming to return exact results cannot be done better than the naive method — a linear scan of the entire collection {{cite:bffd0c30c93a423b2144cfd1382dce30020a9365}}. In the IR domain, documents are typically represented as feature vectors in a space of more than thousands of dimensions {{cite:850acbe3d8f4cad84396b5de6cf981d8de4a7bce}}. Nevertheless, if the complete exactness of results is not really necessary, similarity search in a high-dimensional space can be dramatically speeded up by using hash-based methods which are purposefully designed to approximately answer queries in virtually constant time {{cite:e405e16b3f5535f26dd554c943e33fa9f7a75438}}.", "Such hash-based methods for fast similarity search can be considered as a means for embedding high-dimensional feature vectors to a low-dimensional Hamming space (the set of all {{formula:91b99edf-53a2-4496-97fe-74550e350715}} binary strings of length {{formula:c485f242-7d46-4261-947a-89218adfa6b5}} ), while retaining as much as possible the semantic similarity structure of data. Unlike standard dimensionality reduction techniques such as Latent Semantic Indexing (LSI) {{cite:4cb644231d7832d2af2e29d834be88edfe29beb1}}, {{cite:1924914565406b0331c7d678e2e7305f97317dfe}} and Locality-Preserving Indexing (LPI) {{cite:6983bf0708f048b4743bd4159bd95b1b9abd67cd}}, {{cite:91a7a82965cf7dadc1816e13b23d243d31913afc}}, hashing techniques map feature vectors to binary codes, which is key to extremely fast similarity search (see Section ).One possible way to get binary codes for text documents is to binarise the real-valued low-dimensional vectors (obtained from dimensionality reduction techniques like LSI) via thresholding {{cite:70cb018f31480b7be16ea3eace40d30427328bf7}}.An improvement on binarised-LSI that directly optimises a Hamming distance based objective function, namely Laplacian Co-Hashing (LCH), has been proposed recently {{cite:fd96b58ec17238e3ef8d1cf13653391923ddc9ed}}.", "The most well-known hashing technique that preserves similarity information is probably Locality-Sensitive Hashing (LSH) {{cite:7953108a2b7fee7d9fa42a6f2012853b4fdd8131}}. LSH simply employs random linear projections (followed by random thresholding) to map data points close in a Euclidean space to similar codes. It is theoretically guaranteed that as the code length increases, the Hamming distance between two codes will asymptotically approach the Euclidean distance between their corresponding data points. However, since the design of hash functions for LSH is data-oblivious, LSH may lead to quite inefficient (long) codes in practice {{cite:70cb018f31480b7be16ea3eace40d30427328bf7}}, {{cite:ea06e280abea126f58f9204708d821ad7b43f4ea}}.", "Several recently proposed hashing techniques attempt to overcome this problem by finding good data-aware hash functions through machine learning.In {{cite:70cb018f31480b7be16ea3eace40d30427328bf7}}, the authors proposed to use stacked Restricted Boltzmann Machine (RBM) {{cite:45eafc925ed92850838e674589ad6dc51bd779b9}}, {{cite:84bbd6161f082146efd00e8001ca1be4e97963ca}}, and showed that it was indeed able to generate compact binary codes to accelerate document retrieval.Researchers have also tried the boosting approach to Similarity Sensitive Coding (SSC) {{cite:8020bb079f6372294a8a900e1f16b9b343e594ad}} and Forgiving Hashing (FgH) {{cite:b07b4af1477e8107f2732daf2b16b7274abbd087}} — they first train AdaBoost {{cite:2d2f8a593fd54d74fe2e435688c852c784f8c4a1}} classifiers with similar pairs of items as positive examples (and also non-similar pairs of items as negative examples in SCC), and then take the output of all (decision stump) weak learners on a given document as its binary code.In {{cite:1f67fb9947af7f7314954a73f946bb2968a5d757}}, both stacked-RBM and boosting-SSC were found to work significantly better and faster than LSH when applied to a database containing tens of millions of images.In {{cite:ea06e280abea126f58f9204708d821ad7b43f4ea}}, a new technique called Spectral Hashing (SpH) was proposed. It has demonstrated significant improvements over LSH, stacked-RBM and boosting-SSC in terms of the number of bits required to find good similar items.There is some resemblance between the first step of SpH and the unsupervised learning stage of our STH approach, because both are related to spectral graph partitioning {{cite:15d612f335fecf89378125576b35a9620235a29b}}, {{cite:f7030e49e420d07087f856481a7e9aaa10dae001}}, {{cite:bce5c764852c8361eb61a80cbbd0294b09a6c288}}. Nevertheless, we use a different spectral method and take a different way to address the entropy maximising criterion (see Section REF ). More importantly, in order to process query documents, SpH has to assume that the data are uniformly distributed in a hyper-rectangle, which is apparently very restrictive. In contrast, our proposed STH approach can work with any data distribution and it is much more flexible (see Section REF ). The superiority of STH to SpH has been confirmed by our experimental results (see Section ).", "A somewhat related, but different, line of research is to use hashing representations for machine learning {{cite:41c7af64d1300dac8e22270c46bb7a71f46b080e}}, {{cite:de69112802224f999b0bb4bcc98d2c04bb6a9cc9}}. The objective of such techniques is to accelerate complex learning algorithms, but not similarity search. Our work is basically the other way around." ]

[]

[ "The idea to use sequential scanning of documents to research new retrieval approaches is certainly not new: We know of at least one researcher who used sequential scanning over ten years ago for his thesis {{cite:3d7bcd8bb0346ac3ab2859d244ca4b50f3300547}}. Without high-level programming paradigms like MapReduce, however, efficiently implementing sequential scanning is not a trivial task, and without a cluster of machines the approach does not scale to large collections.", "Lin {{cite:0a09b5ba741b223d0a3d9b6182ae6cb3f37cae01}} used Hadoop MapReduce for computing pairwise document similarities. Our implementation resembles Lin's brute force algorithm that also scans document representations linearly. Our approach is simpeler because our preprocessing step does not divide the collection into blocks, nor does it compute document vectors." ]

[]

[ "Although there are numerous graph algorithms reported in the literature,little work has been done in graph-centric benchmark development.The most relevant work in this area is probably the recently developedbenchmark suite, called DARPA High Productivity Computer Systems (HPCS) Scalable Synthetic Compact Application (SSCA)Graph Analysis Benchmark, which is commonly known as SSCA#2 benchmark {{cite:0ac1b856c036de203dcefe3bbbbea2ff51d32000}}.The SSCA#2 benchmark suite is comprised of a synthetic scale-free graph generator based on RMATmethod {{cite:d5c842437c8b33e19ce6893414034cd2d5cff1ba}}and four kernels. Each of the kernels is designed based on a small set of fundamental graph-related algorithms.It is the only benchmark currently available for architectural evaluation for graph-theoretic problems.However, the SSCA#2 benchmark suffers significant shortcomings.One of the biggest drawbacks is that the benchmark is not comprehensive.Also, among the four kernels in the benchmark, kernels 1 and 2 are not directly graph-related.Remaining kernels suffer design flaws that prevent the benchmark from accurately modeling the real execution-timecharacteristics of the targeted algorithms.In addition, its use of RMAT as a synthetic graph generator results in longer graph generation time andgraphs that do not resemble common real-world scale-free graphs.", "Gokhale et. al. {{cite:7351c7297bc9457c941718e0fef927a123031e75}} have evaluated the use ofexisting and emerging computing architectures, storage technologies,and programming models to solve data-intensive problems in various scientific and engineering disciplines.One of the data-intensive applications considered in their research was graph algorithms. In particular, they wereinterested in the graph algorithms that access very large graphs stored in external storage devices.In an effort to investigate the use of the new storage devices and identify what is needed for the efficient and scalablecomputation of graph applications, they developed a graph-centric benchmark that operates on out-of-core graphs.This benchmark measures the performance and scalability of the graph ingestion and search.", "A library that contains a set of the most commonly used graph algorithms hasbeen developed {{cite:852c3830f29d016252c18e490721244146110799}}.The library, called Boost Graph Library (BGL), enables the reuse of graph algorithms and datastructures by providing the users with a generic interface that allows access to a graph's structure whilehiding the details of the implementation.Therefore, any graph algorithms or libraries that implement this interface will be able toaccess the BGL generic algorithms and other algorithms that use the interface.The BGL offers some of the fundamental graph algorithms that includebreadth- and depth-first search algorithms, Dijkstra's shortest path algorithm, Kruskal's and Prim'sminimum spanning tree algorithms, and find strongly connected components algorithms.In addition to the graph algorithms, the BGL supports common graph representations: adjacency_list andadjacency_matrix.The BGL itself can be used as a benchmark, since it supports a wide range of fundamentalgraph algorithms and common data structures to represent graphs.However, it lacks a means to fine control the behavior of these graph algorithms, which any realbenchmark is expected to provide." ]

[]

[ "The Gaussian Z-interference channel has been extensively studied inthe literature. It is one of the few examples of an interference channel(besides the strong interference case {{cite:af6b71c861c284a4417dc863a80cabccda778b2a}}, {{cite:2c5f741d381a568345276f9de50e39d2f76119e4}}, {{cite:a652a614f427661b802a21c5fc8d5db90e271086}}and the very weak interference case {{cite:0095f89608cfdab4bdc2f73f921dd9d9bc612474}}, {{cite:25c1cdb24b6f0abe75b2089bb75f4d4aa90e0562}}, {{cite:2ac412688819a8ed730983132accc7dee86c10ac}}) forwhich the sum capacity has been established. The sum capacity of theGaussian Z-interference channel in the weak interference regime isachieved with both transmitters using Gaussian codebooks and with theinterfered receiver treating the interference as noise{{cite:e207a6a79e91eacf759563368831bba85f98bcb1}}, {{cite:a7b1f9e4e2e715bac563a467c8006463d6f2cf19}}.", "The fundamental decode-and-forward and compress-and-forward strategiesfor the relay channel are due to the classic work of Cover and El Gamal{{cite:2c8d9674d3a5a81667a6a8f803eb0168ec6d4a15}}. Our study of the interference channel with a relaylink is motivated by the more recent capacity results for a class ofdeterministic relay channels investigated by Kim {{cite:59cc8e8ce3a91312f1b734f1cb2b7a74dca1c2c9}} and aclass of modulo-sum relay channels investigated by Aleksic et al.{{cite:491b1725c211a2633d81c963bd94bc3c36eea90d}}, where the relay observes the noise in the directchannel. The situation investigated in {{cite:59cc8e8ce3a91312f1b734f1cb2b7a74dca1c2c9}}, {{cite:491b1725c211a2633d81c963bd94bc3c36eea90d}} is similarto the Type I Gaussian Z-relay-interference channel, where theinterference-free receiver observes a noisy version of theinterference at the interfered receiver and helps the interferedreceiver by describing the interference through a noiseless relaylink.", "The channel model studied in the paper is related to the work of Sahin et al. {{cite:b56f4498b44a2a2b5b7743d50ac8922f063560af}}, {{cite:dc1755a5e8d84f9d17e5dbdda356db7808889030}}, {{cite:6085a029f9876ef4f404f116732229e301225a75}}, Marić et al.{{cite:6beca44a35a44759a2b8432b32910dfae7a5f922}}, Dabora et al. {{cite:e0fe2fc1dd89a778945610be16fbcb814648247e}}, and Tian and Yener {{cite:2a2995f2afc87882cc08eef7b53abe7e3a6b3892}}, where theachievable rate regions and the relay strategies are studied for an interference channel with an additional relay node, and where the relay observes the transmitted signals from the inputs and contributes to the outputs of both channels. In particular, {{cite:6beca44a35a44759a2b8432b32910dfae7a5f922}}, {{cite:e0fe2fc1dd89a778945610be16fbcb814648247e}} propose an interference-forwarding strategy which is similar to the one used for the Type I channel in this paper.In a similar setup, the works of Ng et al.{{cite:29ca81649955fb9b1404a0d09e761fdc2b013358}} and Høst-Madsen{{cite:531ce803863cac3b230fb72060b57d028c95eb38}} study the interference channelwith analog relay links at the receiver, and use thecompress-and-forward relay strategy to obtain capacitybounds and asymptotic results.", "This paper is closely related to the work of Wang and Tse{{cite:e126564aea14c573f112d9c950ad5c8944c333c7}}, Prabhakaran and Viswanath{{cite:c4b7ee645be2c1517ddcdfed4ce5a6f0441fa078}}, and Simeone et al.{{cite:a8b5ba805a29d1f39c619191468240e42778143a}}, where the interference channel with limited receivercooperation is studied. In {{cite:a8b5ba805a29d1f39c619191468240e42778143a}}, the achievable rates of aWyner-type cellular model with either uni- or bidirectionalfinite-capacity backhaul links are characterized. In{{cite:e126564aea14c573f112d9c950ad5c8944c333c7}}, a more general channel model in whicha two-user Gaussian interference channel is augmented with bidirectionaldigital relay links is considered, and a conferencing protocol basedon the quantize-map-and-forward strategy of {{cite:e8dbc50974bf9b18c8dc68d86d22fe4c7b69bbd7}}is proposed.", "The present paper considers a special case of the channel model in{{cite:e126564aea14c573f112d9c950ad5c8944c333c7}}, i.e.,a simplified Gaussian Z-interference channel model witha unidirectional digital relay link. Byfocusing on this special case, we are ableto derive concrete achievability results and upper bounds and obtaininsights on the rate improvement due to the relay link.For example, while {{cite:e126564aea14c573f112d9c950ad5c8944c333c7}} adopts a universalpower splitting ratio of {{cite:e207a6a79e91eacf759563368831bba85f98bcb1}} at the transmitter to achievethe capacity region to within 2 bits, this paper adapts thepower splitting ratio to channel parameters, and shows thatin the weak interference regime a relay link from the interference-freereceiver to the interfered receiver is much more beneficial than arelay link in the opposite direction for a Z-interference channel." ]

[]

[ "We briefly review recent studies on {{formula:6afc3de0-83a5-42d4-bc08-e0c43a3663ed}} -regularization,most of which focus on {{formula:39c290a7-aca4-4885-9260-5452d3fe5d10}} -regularization and/or{{formula:4e90eb7b-656d-4c16-99fd-a8206eaa8c4b}} -regularization.", "{{formula:b67b1ad1-866c-4c52-9fa4-2e0fed57e15b}} -Regularization: The group Lasso was proposedin {{cite:666cd9b841ce59bfa32b4ea2988ee534e54903ce}} to select the groups of variables for predictionin the least squares regression. In {{cite:38be93ed19eade5dec9b93ed2952849a4956e139}}, the idea ofgroup lasso was extended for classification by the logisticregression model, and an algorithm via the coordinate gradientdescent {{cite:d847710a1bfd84aa3e4e1947de3f73242581cccb}} was developed. In {{cite:9592bc1ff3b821e5868f7e347e5b555595f603cc}},the authors considered joint covariate selection for groupedclassification by the logistic loss, and developed a blockwiseboosting Lasso algorithm with the boosted Lasso {{cite:33b0d3f4d1613344bdfd884c2348c820f4cb0919}}.In {{cite:045484f1762bf64e95fe132aef421d9ba3a5c3f7}}, the authors proposed to learn the sparserepresentations shared across multiple tasks, and designed analternating algorithm. The Spectral projected-gradient (Spg)algorithm was proposed for solving the {{formula:cc467912-53b1-4e37-8e79-20a733f19117}} -ballconstrained smooth optimization problem {{cite:35dea6998e856d2cdccb3b5e4f9f2c071e8e040c}},equipped with an efficient Euclidean projection that has expectedlinear runtime. The {{formula:b54a8290-ca14-48ac-ab95-96cdd6335471}} -regularized multi-task learningwas proposed in {{cite:f4e7cea3eb152c31c69e42533ca0406b6d186b51}}, and the equivalent smoothreformulations were solved by the Nesterov'smethod {{cite:71b848dcffa1019d020b95886f5d83c2a1bd5d5e}}.", "{{formula:880423b2-a44f-4f50-8357-fb958eb14ed0}} -Regularization: A blockwise coordinate descentalgorithm {{cite:f1ca8f344ccb8c8669dab5a29921b9ccd8246109}} was developed for the mutli-taskLasso {{cite:6465c603a124711dbac5200e962d788c0d038e99}}. It was applied to the neuralsemantic basis discovery problem. In {{cite:ef29eece38b4ffa6f04e7e18623f9f7fbe22a7f5}}, theauthors considered the multi-task learning via the{{formula:256e6b6d-3080-4457-8c2f-b865965411e5}} -regularization, and proposed to solve theequivalent {{formula:3ed04e70-d5d6-4831-a965-3f1bfe37f6fa}} -ball constrained problem by theprojected gradient descent. In {{cite:e842f18d5089558121887eb764bc0d09a11689d4}}, the authorsconsidered the multivariate regression via the{{formula:d56fb249-34e9-4ebc-b86a-78fd44d50c84}} -regularization, showed that thehigh-dimensional scaling of {{formula:be75c95c-c43e-4232-9df9-41d4270cf146}} -regularization isqualitatively similar to that of ordinary {{formula:28697f86-2178-4e44-a332-b5cb3e68032d}} -regularization,and revealed that, when the overlap parameter is large enough({{formula:3c687f41-6787-44c0-949c-d27f0f3bceae}} ), {{formula:09244e5b-316d-44aa-980a-8a0e8309ae5b}} -regularization yields the improvedstatistical efficiency over {{formula:e330324c-c197-49ac-9bde-7ff57047620f}} -regularization.", "{{formula:655348ab-12da-4c69-9f71-7cfd2ead1191}} -Regularization: In {{cite:b2b59eee222da890c9d3e01cd3639ec541134395}}, theauthors studied the problem of boosting with structural sparsity, anddeveloped several boosting algorithms for regularization penaltiesincluding {{formula:8b44efb4-cd3f-434c-af91-cc878627659f}} , {{formula:f82f0acb-e5e8-4196-bc2d-9404f8fa41ad}} , {{formula:075e740a-43dd-4de3-9e5a-1f980ca6a3d3}} , and{{formula:884be1b1-094b-4831-bfb3-8d5ecf379e0f}} . In {{cite:27a208aec9a07e7d21d562be84833d54137a178a}}, the composite absolutepenalties (CAP) family was introduced, and an algorithm called iCAPwas developed. iCAP employed the least squares loss and the{{formula:cda1092a-0eaa-407a-9de2-ae522ae8627a}} regularization, and was implemented by theboosted Lasso {{cite:33b0d3f4d1613344bdfd884c2348c820f4cb0919}}. The multivariate regression with the{{formula:0e113360-8e3d-418c-a84c-0462b03b3edb}} -regularization was studiedin {{cite:652f8c07f7cb48abd3535fc648f1bd629166b6d1}}. In {{cite:1109311f8eaf9dcd6ed11206ad526e153deee85a}}, a unifiedframework was provided for establishing consistency and convergencerates for the regularized {{formula:0ee7bb63-7a13-4bfe-8097-71b40e61d93e}} -estimators, and the results for{{formula:38b291ff-d1b0-4ca0-abae-54c07932cb06}} regularization was established." ]

[]

[ "This paper is concerned with the efficient evaluation of higher-order derivatives of functions {{formula:131b3eac-4f50-41a7-8922-6dd998fc578f}} which are implemented as computer programs that contain numerical linear algebra functions like the QR or the real symmetric eigenvalue decomposition.", "Traditionally, Algorithmic Differentiation (AD) tools like ADOL-C {{cite:f6a242a669dc9e6bde786f53124af2802355c7b8}} or CppAD {{cite:a7aa740d80612f6d364fc6e27468f4077b949232}} regard the functions defined in the C header file math.h as elementary functions. In the forward mode of AD, their approach to compute higher-order derivatives is to generalize from real arithmetic to univariate Taylor polynomial (UTP) arithmetic {{cite:f6a242a669dc9e6bde786f53124af2802355c7b8}}, {{cite:66cb1d4ec10db133abb8ccd388be564c3c6348c9}}, {{cite:4624e48a7113692e656efe202c6e339fed2a4d01}}. For the reverse mode of AD, the program evaluation is traced and stored in a computational graph or on a sequential tape. During the so-called reverse sweep the stored intermediate values are retrieved and used to compute derivatives (c.f. Section ).", "As explained in Section , the functions in math.h suffice since all computable functions are a concatenation of these functions. However, working only at the expression level has also disadvantages since no global knowledge of the function's structure can be used. A particularly important class of algorithms in science and engineering are numerical linear algebra (NLA) functions. Though NLA functions are typically locally smooth, their implementations often contain non-differentiable operations and program branches. If no special care is taken, this may result in incorrect computations of derivatives. Also, many NLA functions on {{formula:988407db-fe60-4a36-be59-87ce9781f5d1}} matrices require {{formula:4bcd4b22-cc84-42fc-b21e-9c58a19b641c}} arithmetic operations. Since during the reverse mode intermediate results are required, this would yield an {{formula:016c05ee-8709-4a45-8f92-eef24b27ea3c}} memory requirement. Though it may be possible to adapt codes to yield reduced memory requirements, as for instance reported for the LU decomposition {{cite:32abf7318be081b42089a0e697f42391c5fd4132}}, in practice it can be a cumbersome and error-prone process. Also one would like to reuse existing, high-performance implementations of NLA algorithms. Adding the NLA functions to the list of elementary functions circumvents this problem. This has been realized before {{cite:10cc54e278cac8ecb500c2414529842b5bc90492}}, {{cite:2efae808f0b1e657954668fcdbfe61b4c28a9727}} and also UTP algorithms for some NLA functions (e.g. the solution of linear equations) have been implemented in software {{cite:73e22ead829e6ba6dd0ce479c35a0083d4ce821d}}.", "The contribution of this paper is to provide explicit algorithms for UTP arithmetic applied to the QR decomposition and the real symmetric eigenvalue decomposition. Note that our approach to the real symmetric eigenvalue decomposition is similar to {{cite:326dcd4affe57503aa6a73fbee3c32a1cb527069}}, {{cite:a9f97b685629f079fa8181fd3b2812356a3fcd91}} but our algorithmic result differs. In addition, we also treat the reverse mode of AD.", "The document is structured as follows: In Section we give two application examples for the algorithms presented in this document, followed by a brief review of the underlying computational model in Section . We shortly describe the basics of AD in Section where we make use of the results from . In Section we describe the general approach of NLA functions. After that, we apply the results from Section to find extended functions for the QR and eigenvalue decomposition in Section and and also provide pullback algorithms that are necessary in in the reverse mode of AD. Finally, we present some numerical results in Section ." ]

[]

[ "Several distance functions are used for time series classification, such as", "Dynamic Time Warping (DTW) {{cite:edafc5def43a4cec31407ee68c48578e5848a579}},DISSIM {{cite:c3304790974f2d54be2d7192f38a133a5d97b4dc}},ThresholdQueries {{cite:65d7ca02b17ddd7a384265a00d73f93be38ff3e1}},Edit distances {{cite:21b01bd0cfffa07b1fd0457fca86619b627686a6}}, {{cite:3be4c92db57eb6da108e4130a5c722c8b84addf5}},Longest Common Subsequences (LCSS) {{cite:8c5e54d190902d9032f95eedf967fb5924e776f6}},Swale {{cite:04688d37ac7055b0359652c25b64e2419e5cfcba}},SpADe {{cite:b9116874ec819695d2cb46378c75c265102d459f}},and Cluster, Then Classify (CTC) {{cite:9eafc00b1a5e0348295967712cd7d6075317720b}}.", "{{cite:ef4231577f481c6764b07f497a9e6ab5acd09c38}} presented an extensive comparison of these distance functions andconcluded that DTW is among the best measures and that the accuracy of the Euclidean distanceconverges to DTW as the size of the training set increases.", "In a general Machine Learning setting, {{cite:ad10ea0da68bf725251b7bb32788f06a015f0895}}, {{cite:c3aa6e3ff7190315f3fc2900e053bf845aeae682}} compared Euclidean distance with the conventional and class-based Mahalanobis distances.One of our contribution is to validate these generic results on time series: instead of tens of features, we have hundreds or even thousands of values which makes the problem mathematically more challenging: the rank of our covariance matrices areoften tiny compared to their sizes.", "More generally, distance metric learning has an extensive literature {{cite:811bb32169fcc31f62c56ccfba7db2b5d0aeaf21}}, {{cite:e8f3bdc2fc5bd14049348f1f78cfd960f4e5fae5}}, {{cite:d31f0a6bd837ee90b520e965a0c43e96b1840d00}}, {{cite:bb633cd1fed7c25083bab126de21463a7e7cd380}}.We refer the reader to {{cite:7b3426c9378ce15f08e8f88ab1eaa2165a73e44a}} for a review.", "A conventional distance-learning approach is to find an optimal generalized ellipsoiddistance with respect to a specific loss function.The LMNN algorithm proposed by {{cite:7b3426c9378ce15f08e8f88ab1eaa2165a73e44a}} takes a different approach. It seeks to force nearest neighbors to belong to the same class and it separates instances from different classes by a large margin.LMNN can beformulated as a semi-definite programming problem.They also propose a modificationwhich they call multiple metrics LMNN as it learns different distances for each class.", "There are many extensions and alternatives to NN classification. For example,{{cite:03ba381dda558e4d88c062b3ec8b085d3f18acbf}} use instance weights to improve classification.Meanwhile, {{cite:f482d7aa11071cf733368e35049c643978ed5df6}} learn a distance per instance." ]

[]

[ "Related work (to various degree of relevance)can be found below (this list is not exhaustive):", "Taxonomy of Linux kernel vulnerability solutions in terms ofpatches and source code as well as categories for both are found in{{cite:5c011da7e818d0328b95c2ac9b8340637310e4d3}}.The core ideas and principles behind the MARF's pipeline and testingmethodology for various algorithms in the pipeline adapted to this caseare found in {{cite:12fa7112a6e1a85ca690a4c19f0959c57b734631}}. There also one can find the core optionsused to set the configuration for the pipeline in terms of algorithms used.A binary analysis using machine learning approach for quick scans forfiles of known types in a large collection of files is described in{{cite:90074dbd70419fb1c627c06e1f4135573cdc4d61}}.The primary approach here is similar in a way that was done forDEFT2010 {{cite:022e7d7cb3aaa29540190623c387855043dc0c35}}, {{cite:e62f198812e71f23ace13e82c9c5685eef57faf8}} with thecorresponding DEFT2010App and its predecessorWriterIdentApp {{cite:b1255f866f53443d8a10846b1a9ea17ed8865744}}.Tlili's 2009 PhD thesis covers topics on automatic detection of safety and securityvulnerabilities in open source software {{cite:c815299fe0e9a66cb9921a0e42de4b83ef9c0bf5}}.Statistical analysis, ranking, approximation, dealing with uncertainty,and specification inference in static code analysis are found in theworks of Engler's team {{cite:39ff7d6d92ad3fede6066cc81fda6662df54821c}}, {{cite:9f916c483a53ccaee47c8d8d45c89a5769bea9f8}}, {{cite:ec19209e06065139727d86a7a508879e78d8fae2}}.Kong et al. further advance static analysis (using parsing, etc.) and specificationsto eliminate human specification from the static code analysis in{{cite:88572f6e7f9311d17be34593572d6f1b0658c67c}}.Spectral techniques are used for pattern scanning in malware detection by Eto et al. in{{cite:ac5d0d74a494ee791b1e6405ebd869aef7fdb400}}.Researchers propose a general data mining system for incident analysis with data miningengines in {{cite:6b79c9215d86898908f8b5a75a60edc86a0fe7cf}}.Hanna et al. describe a synergy between static and dynamic analysis for the detectionof software security vulnerabilities in {{cite:b4561f1d2b2b50c4b13d9590a4386c380a5d6fe8}} paving the wayto unify the two analysis methods.The researchers propose a MEDUSA system for metamorphic malware dynamic analysisusing API signatures in {{cite:6ed84d6e176ee7e57a4af020b17aafb629319b97}}." ]

[]

[ "BFS used in practice. BFS is widely used today for exploring large networks, such as OSNs.In {{cite:d16d04a4221fa95a7d2e9affc225382e955c529a}}, Ahn et al. used BFS to sample Orkut and MySpace. In {{cite:9b9a8002cb9355adb598c6be7c55ea7cb93a74e7}} and {{cite:b4a549cc4efb49ecba4c9c445635ea89d5739056}}, Mislove et al. used BFS to crawl the social graph in four popular OSNs: Flickr, LiveJournal, Orkut, and YouTube.{{cite:9b9a8002cb9355adb598c6be7c55ea7cb93a74e7}} alone has about 380 citations as of December 2010, many of which use its highly biased Orkut BFS sample.In {{cite:f7f49438fc3cb6641d148b82a5cba17b80213139}}, Wilson et al. measured the social graph and the user interaction graph of Facebook using several BFSs, each BFS constrained in one of the largest 22 regional Facebook networks. In our recent work {{cite:38b6555ca4bf7d681ad190eea9f9afa7a27a93a3}}, {{cite:65323233c2e865ca9bb3495a5123e53f73186d29}}, we have also crawled Facebook using various sampling techniques, including BFS, RW and MHRW.", "BFS bias.It has been empirically observed that incomplete BFS and its variants introduce bias towards high-degree nodes {{cite:863e2f895fda44f17a9682aa2defbb54718bbf81}}{{cite:dfa53efe127599f8cee97c312db3a703ff1778f3}}, {{cite:2a42a1dfd8f95f484ee11d3dce8537e1a7855402}}, {{cite:47edcb051ce9db13e8acd8586a40c5aae3ece2a4}}.We confirmed this in Facebook {{cite:38b6555ca4bf7d681ad190eea9f9afa7a27a93a3}}, {{cite:65323233c2e865ca9bb3495a5123e53f73186d29}}, which, in fact, inspired and motivated this paper.Analogous bias has been observed in the field of social science, for sampling techniques closely related to BFS, i.e., Snowball Sampling and RDS {{cite:eb00a72c57a45a58da082a962a95493adc6e5827}}, {{cite:6a2d48cda2946f963edd94bb5f4e3d7a4bf3264d}}, {{cite:dfa1e67ebf321a93148f1d2752fe9b5221438159}} (see Section REF ).", "Analyzing BFS. To the best of our knowledge, the sampling bias of BFS has not been analyzed so far.{{cite:98eb22af34973ae5035a45bf7f07af3f717cdb28}} and {{cite:b7bedeaa7a2fd51fe049b47fe5fd3a1c55f56ccb}} are the closest related papers to our methodology. The original paper by Kim {{cite:98eb22af34973ae5035a45bf7f07af3f717cdb28}} analyzes the size of the largest connected component in classic Erdös-Rényi random graph by essentially applying the configuration model with node degrees chosen from a Poisson distribution. To match the stubs (or “clones” in {{cite:98eb22af34973ae5035a45bf7f07af3f717cdb28}}) uniformly at random in a tractable way, Kim proposes a “cut-off line” algorithm. He first assigns each stub a random index from {{formula:eb9f7941-de51-4403-a372-b83684e7f9ae}} , and next progressively scans this interval. Achlioptas et al. used this powerful idea in {{cite:b7bedeaa7a2fd51fe049b47fe5fd3a1c55f56ccb}} to study the bias of traceroute sampling in random graphs with a given degree distribution.The basic operation in {{cite:b7bedeaa7a2fd51fe049b47fe5fd3a1c55f56ccb}} is traceroute (i.e., “discover a path”) and is performed from a single node to all other nodes in the graph.The union of the observed paths forms a “BFS-tree”, which includes all nodes but misses some edges (e.g., those between nodes at the same depth in the tree). In contrast, the basic operation in the traversal methods presented in our paper is to discover all neighbors of a node, and it is applied to all nodes in increasing distance from the origin. Another important differenceis that {{cite:b7bedeaa7a2fd51fe049b47fe5fd3a1c55f56ccb}} studies a completed BFS-tree, whereas we study the sampling process when it has visited only a fraction {{formula:fe82f41f-f685-4c02-8225-b4761c32916e}} of nodes. Indeed, a completed BFS ({{formula:ebf33f29-570f-45e1-aa0d-d92a9decf64f}} ) is trivial in our case: it has no bias, as all nodes are covered.", "In the field of social science, a significant effort was put to correct for the bias of BFS's close cousin - Snowball Sampling (SBS) {{cite:eb00a72c57a45a58da082a962a95493adc6e5827}}. SBS together with a bias correction procedure is called Respondent-Driven Sampling (RDS) {{cite:6a2d48cda2946f963edd94bb5f4e3d7a4bf3264d}}. The currently used correction technique {{cite:dfa1e67ebf321a93148f1d2752fe9b5221438159}}, {{cite:d11f53a459ecb4a23927e0e97436bde42b3c3a98}} assumes that nodes can be revisited, which essentially approximates SBS by Random Walk (see Section REF ).In this paper, we formally show that this approximation is valid if the fraction {{formula:a62a0bcd-e73e-426e-9314-a8563b9c003c}} of sampled nodes is relatively small. However, as {{cite:fd21b76f4cab95ec29638701823809aecf76359f}} points out, the current RDS methodology is systematically biased for larger {{formula:bc8a7c81-fd38-49cc-b844-228c8c6271ac}} . Consequently, {{cite:2d2e4819a1a782e6e4e0aeb2d4eb2f17a9e35180}} proposed an SBS bias correction method based on the random graph {{formula:3c10a542-3909-405c-ad2a-4745ff740178}} . This is essentially the same basic starting idea as used in our original paper published independently {{cite:1d9bb95cb3fc352383198ca81519454e5223d1d0}}. However, the two papers fundamentally differ in the final solution: {{cite:2d2e4819a1a782e6e4e0aeb2d4eb2f17a9e35180}} proposes a simulation-aided approach, whereas we solve the problem analytically.", "Another recent and related paper is {{cite:19d8c55e7cfca9ab9588250b4631ac59009c54b2}}.The authors propose and evaluate a heuristic approach to correct the degree bias in the {{formula:f01a461e-c454-4cbc-a55f-a35d6e60c6a0}} th generation of SBS, based on the values measured in the generation {{formula:682227f1-39cd-48de-918c-86f728c0cc80}} .In practice, this generation-based scheme may be challenging to implement, because the number of nodes per generation may grow close to exponential with {{formula:2d8eafde-1637-4162-82cb-1c0d5c65c826}} . Consequently, we are likely to face a situation where collecting the next generation is prohibitively expensive, while the current generation has much fewer nodes than our sampling capabilities allow for.", "Probability Proportional to Size Without Replacement (PPSWOR).At a closer look, our {{formula:c7452159-7ec2-4232-97d4-75c13465336c}} -based approach reduces BFS (and other graph traversals) to a classic sampling design called Probability Proportional to Size Without Replacement (PPSWOR) {{cite:a7e241ab57ac93e7a43f1bc5ef6108e4a9b8421a}}, {{cite:047131937998d3f941d1c495c129f9e71688b491}}, {{cite:7ce956ef7c9bc5b6f5f9705508d980881161c440}}, {{cite:7e425605cbc5c9430e0da8d4173b0d80a4fae64c}}, {{cite:6848de37a9b222230f8ec9feb4222e7a1ab7c013}}, {{cite:4a86490de46e2db7293e93f7dfb44c7a6269ad02}}, {{cite:c1db3848db70f4a6d310c84daa899b00bdf67cd4}}, {{cite:80ead95d2bf000d14fc5012570cca4248ec30fca}}.Unfortunately, to the best of our knowledge, none of the existing results is directly applicable to our problem.This is because, speaking in the terms used later in this paper, the available results either (i) require the knowledge of {{formula:607e33c6-dbbb-41a5-ac64-2ac9e4f8d7a2}} (expected, not sampled) as an input, (ii) propose how to calculate {{formula:c6d38e1d-9afa-40dd-a364-2ed1df2d6712}} for the first two nodes only, or (iii) calculate {{formula:5b217859-3d74-424f-a67a-e12207903151}} as an average of many simulated traversals of the known graph (in contrast, we only have one run on unknown graph) {{cite:80ead95d2bf000d14fc5012570cca4248ec30fca}}.In fact, this work can be naturally extended to address the problems with PPSWOR.", "Previous version of this paper.This work is a revised and extended version of our recent conference paper {{cite:1d9bb95cb3fc352383198ca81519454e5223d1d0}}.The main changes are:(i) a successful application of our {{formula:39746796-c3ac-4ca2-9fa6-b1cc48115624}} -based correction procedure to a wide range of large-scale real-life Internet topologies (Table REF , Fig. REF , Fig. REF (d), Section REF ),(ii) bias correction procedures for arbitrary node properties (Section ),(iii) a complementary BFS-bias correction technique (Section ), and(iv) a publicly available ready-to-use python implementation of our approach.", "Finally, we would like to stress that our two other JSAC submissions {{cite:65323233c2e865ca9bb3495a5123e53f73186d29}}, {{cite:aaf2b2bfb3f1308d1ae6a9349d118a0d16c823b0}} focus on sampling techniques based on random walks, which differ in fundamental aspects (sampling with replacement vs without, sampling of nodes vs of topology) from the BFS sampling addressed here." ]

[]

[ "Probabilistic bisimulation was first introduced by Larsen and Skou{{cite:666836408f9092d5ffaa2ad5e1c3361f1318ca47}}. Later on, it was investigated in a great manyprobabilistic models.An adequate logic for probabilistic bisimulation in a setting similar to our pLTSs hasbeen studied in {{cite:7658097f60c258c81bc760d0c3bc9f9cd1a02128}}, {{cite:c92da52af7d6922fd31ba84dbea385bda1836e4f}}. It is also based on an probabilisticextension of the Hennessy-Milner logic. The main difference from ourlogic in Section REF is the introduction of theoperator {{formula:9c652ccd-65a0-4686-9367-d69703d1510d}} . Intuitively, a distribution {{formula:cd2471a2-ad34-4dee-9afe-e5a6ed6d4a9c}} satisfies the formula {{formula:20bf693a-52be-43f9-9ae7-b72d72c38f41}} when the set of states satisfying{{formula:f3c49522-eeca-4d87-92cf-56cdbf093d1c}} is measured by {{formula:9eff8e1c-f1d3-409c-8099-631a8e6a00de}} with probability at least {{formula:7ad70c76-61b4-43e8-8d6b-e7784c239f33}} . So theformula {{formula:62d374c8-f516-4f5d-89cf-c5d9cde97663}} can be expressed by our logic in terms of theprobabilistic choice {{formula:46e0ec9a-2805-4773-995c-301b4fe0d810}} by setting{{formula:576634dc-1f3e-47c1-8dfc-640e7cbcf4d0}} , {{formula:81ad43de-6c3c-4a97-8d23-5bd62021e598}} , {{formula:d27b1b36-e22f-41ce-9521-38fe5d9310eb}} , {{formula:4834dbd9-5936-4df8-ac69-975c49384d09}} , and {{formula:d3e3be9f-bcc2-4027-8bb6-74bbaa7505ce}} .When restricted todeterministic pLTSs (i.e., for each state and for each action, thereexists at most one outgoing transition from the state),probabilistic bisimulations can be characterised by simpler forms oflogics, as observed in {{cite:666836408f9092d5ffaa2ad5e1c3361f1318ca47}}, {{cite:fa6c7e6a35963934d19ecdac2b2b924de65ee109}}, {{cite:c92da52af7d6922fd31ba84dbea385bda1836e4f}}.", "An expressive logic for nonprobabilistic bisimulation has been proposed in {{cite:aeaf93ffede047bd6807538dd8981b4677eec0c8}}.In this paper we partially extend the results of{{cite:aeaf93ffede047bd6807538dd8981b4677eec0c8}} to a probabilistic setting that admits bothprobabilistic and nondeterministic choice. We present aprobabilistic extension of the modal mu-calculus {{cite:69996de659a8d498677240eba1f02ec7f3ed9156}}, wherea formula is interpreted as the set of states satisfying it. This isin contrast to the probabilistic semantics of the mu-calculus asstudied in{{cite:3812026da5b8b308d30a424c33dd8e1ebd2ee836}}, {{cite:2e257c62e1a3edd0c8cd0d79f38c791d7d9db2db}}, {{cite:ae9f7672edffb95e873b0d9f4c71455779b98bcd}} where formulae denote lower bounds ofprobabilistic evidence of properties, and the semantics of thegeneralised probabilistic logic of {{cite:b0756462975382fb8e889d5ff059791beaa0c00c}} where a mu-calculusformula is interpreted as a set of deterministic trees that satisfyit.", "The Kantorovich metric has been used by van Breugel et al.for defining behavioural pseudometrics on fully probabilisticsystems {{cite:251bbd213a6bf3c4bfd270c44fca9932089a1a72}}, {{cite:305db3ffb8a2eb4b5a6ebe5b02da164766fd1961}}, {{cite:130d0f07f7c032c2a812be1109f5884a07dccef0}} and reactive probabilistic systems{{cite:c69b8442273febbc3ca4147875b04de9a6ca2986}}, {{cite:8c303c1d62b17b464f60d966ba0b7801506467ab}}, {{cite:245bf25fa94bd3f2cf4b0ebf4a77567d3ebec56d}}, {{cite:d44f511b30372dd1b6dd3886e40f79e330009dd0}}; and by Desharnais et al. forlabelled Markov chains {{cite:a0ad4a73267a0089018e7185da765f2468f23215}}, {{cite:d18c40282bd5c5f78beee821bbea90d9b84eae5d}} and labelled concurrentMarkov chains {{cite:fcae1b4fee68657219ba636e3a4969310d598114}}; and later on by Ferns et al. forMarkov decision processes {{cite:a034fc98f7115ec5722e763f54aa68806bf95a90}}, {{cite:85e03e30f5cfc1b80a58d58c606b58d48ae2e87d}}; and by Deng etal. for action-labelled quantitative transition systems{{cite:ef5a81fb28cc472a9c10674978578d0173539130}}. One exception is {{cite:673b39f952067f8c43e82ba58af6343b84d90d66}}, which proposes apseudometric for labelled Markov chains without using the Kantorovichmetric. Instead, it is based on a notition of {{formula:a5435913-f030-4456-b628-5ebc5c09de7f}} -bisimulation, whichrelaxes the definition of probabilistic bisimulation by allowing smallperturbation of probabilities.In this paper we are mainly interested in thecorrespondence of our lifting operation to the Kantorovich metric.The metric characterisation of probabilistic bisimulation inSection  is merely a direct consequence of thiscorrespondence.", "Decision algorithms for probabilistic bisimilarity and similarityhave been considered by Baier et al. in {{cite:54cb57fbee52d48d5e446bf5dc94732e47e28372}} and Zhang etal. in {{cite:12ee29ce425d8893dbe74d68d29e5623ead2bf70}}. Their algorithms are global in the sense thata whole state space has to be fully generated in advance. Incontrast, “on the fly\" algorithms are local in the sense that thestate space is dynamically generated which is often more efficientto determine that one state fails to be related to another. Ouralgorithm in Section  is inspired by {{cite:54cb57fbee52d48d5e446bf5dc94732e47e28372}}because we also reduce the problem of checking if two distributionsare related by a lifted relation to the maximum flow problem of asuitable network. We generalise the local algorithm of checkingnonprobabilistic bisimilarity {{cite:54ed00f18a727eac027bcaadfa22140b1f0e0020}}, {{cite:6bbf927b191952265b802bd02f8e49cec9bab517}} to the probabilisticsetting.", "This paper provides a relatively comprehensive account ofprobabilistic bisimulation. Some of the results or their variants were mentioned previouslyin {{cite:ef5a81fb28cc472a9c10674978578d0173539130}}, {{cite:05ea6eb600755903881c5752828fc1d95116f7e0}}, {{cite:a931b38ebf6e44d87c95ec0de9ea42187d810403}}, {{cite:34d3caa2a5903d5619b7fa6d207022779b2cb94d}}. Here they are presented in a uniformway and equipped with detailed proofs." ]

[]

[ "Schenato et al. {{cite:c469e9eb7344a215148fcae715c1288e4745de86}} and Ishii {{cite:90fc5a47a4b748394240972defc4881ea8ddcd5e}} study stabilityand controller synthesis for different control system architectures,but they both model networks as i.i.d. Bernoulli processes that droppackets on a single link. The information passed through theinterface between the network and the control system is the packetdrop probability of the link, which is assumed to be known and fixed.Seiler and Sengupta {{cite:714b6d125d15daa617d9744d9d37c8216aff530e}} study stability and {{formula:4ddbadaf-5e1f-400c-a947-64596a81b689}} controller synthesis when the network is modeled as a packet-droppinglink described by a two-state Markov chain (Gilbert-Elliott model),where the information passed through the network-controller interfaceare the transition probabilities of the Markov chain. Elia {{cite:31d99dd0c6980bad15f1a306e9d11f1e3b5135dc}}studies stability and the synthesis of a stabilizing controller whenthe network is represented by an LTI system with stochasticdisturbances modeled as parallel, independent, multiplicative fadingchannels.", "Some related work in NCSs do use models of multihop networks. Forinstance, work on consensus of multi-agent systems{{cite:e94146884b075acc1dedf4fa798c8c1322e2d2ca}} typically study how the connectivitygraph(s) provided by the network affects the convergence of thesystem, and is not focused on modeling the links.Robinson and Kumar {{cite:29526740679ab160495349131e5b0556a23006cb}} study the optimal placement of a controller ina multihop network with i.i.d. Bernoulli packet-dropping links, wherethe packet drop probability is known to the controller.Gupta et al. {{cite:2594673179f59209c38b1cc2493db205f5a201ed}} study how to optimally process and forward sensormeasurements at each node in a multihop network for optimal LQGcontrol, and analyze stability when packet drops on the links aremodeled as spatially-independent Bernoulli, spatially-independentGilbert-Elliott, or memoryless spatially-correlatedprocesses.Here, “spatially” means “with respect to otherlinks.” Varagnolo et al. {{cite:40ce7714e9fce3b5e8cd98afa89452d0a94423ee}} compare the performance of atime-varying Kalman filter on a wireless TDMA mesh network underunicast routing and constrained flooding. The network model describesthe routing topology and schedule of an implemented communicationprotocol, TSMP {{cite:cf542763dee69e199345639b1b664c1421473dd0}}, but it assumes that transmissionsuccesses on the links are spatially-independent and memoryless.Both Gupta et al. {{cite:2594673179f59209c38b1cc2493db205f5a201ed}} and Varagnolo et al. {{cite:40ce7714e9fce3b5e8cd98afa89452d0a94423ee}} are concernedwith estimation when packet drops occur on the sensing channel, and theestimators do not need to know network parameters like the packet lossprobability." ]

[]

[ "In single-path routing, the path metric is often defined as the sum ofthe link metrics along the path. Examples of link metrics include thenegative logarithm of the link probability (for path probability){{cite:d83e7d46fa3d65f560f51178dfcfcf330bbaed58}}, ETX (Expected Transmission Count), ETT(Expected Transmission Time), and RTT (Round Trip Time){{cite:7b735931d0def13994cc7372a69b7c68384f70f9}}. Most single-path routing protocols find minimumcost paths, where the cost is the path metric, using a shortest pathalgorithm such as Dijkstra's algorithm or the distributed Bellman-Fordalgorithm {{cite:bc1183d5e23e6dc2608ccc773729a9e2b51a9185}}.", "In multi-path routing, one wants metrics to compare collections ofpaths or entire routing topologies with each other. Simply definingthe multi-path metric to be the maximum or minimum single-path metricof all the paths between the source and the sink is not adequate,because such a multi-path metric will lose information about thecollection of paths.", "Our FPP metric is a generalization of the reliability calculationsdone in {{cite:5d29370494ffcbf870a59dc9cf44b5223ae8619e}} for the M-MPR protocol and in {{cite:bf99c2307e49688613ec2a1cb6a11c99d258571d}}for the GRAdient Broadcast protocol. Unlike{{cite:5d29370494ffcbf870a59dc9cf44b5223ae8619e}}, {{cite:bf99c2307e49688613ec2a1cb6a11c99d258571d}}, our algorithm for computing the FPP metricdoes not assume all paths have equal length.", "Our URF metric is similar to the anypath route metric proposed bydubois-Ferriere et al. {{cite:d83e7d46fa3d65f560f51178dfcfcf330bbaed58}}. Anypath routing,or opportunistic routing, allows a packet to be relayed by one ofseveral nodes which successfully receives a packet {{cite:38b54602b72f9f19cfb592dc40125099eed67a95}}.The anypath route metric generalizes the single-path metric bydefining a “links metric” between a node and a set of candidaterelay nodes. The specific “links metric” is defined by thecandidate relay selection policy and the underlying link metric (e.g.,ETX, negative log link probability). As explained later inSection REF , although the packet forwardingmodels for the URF and FPP metrics are not for anypath routing, avariation of the URF metric is almost equivalent to the ERS-best E2Eanypath route metric presented in {{cite:d83e7d46fa3d65f560f51178dfcfcf330bbaed58}}.", "One of our earlier papers, {{cite:be664e02b9c65f3bca9e9b12a2a4d381329e7fc7}}, modeled theprecursor to the WirelessHART protocol, TSMP {{cite:c58866ccc6ac20e8c692e8d0a3ee7ef1e80748b6}}. Wedeveloped a Markov chain model to obtain the probability of packetdelivery over time from a given mesh routing topology and TDMAschedule. The inverse problem, trying to jointly construct a meshrouting topology and TDMA schedule to satisfy stringent reliabilityand latency constraints, is more difficult. The approach taken inthis paper is to separate the scheduling problem from the routingproblem, and focus on the latter. The works{{cite:2900c980fdd4bc65884f48d5c8b2c2a95bcf6d7d}}, {{cite:21f4ed96e935a317302fd478cdcd0690e4d36827}} find the optimal schedule and packetforwarding policies for delay-constrained reliability when given arouting topology.", "Many algorithms for building multi-path routing topologies try tominimize single-path metrics. For instance, {{cite:9957e480d159d74c27218a41c3281161e9ac7733}}extends Dijkstra's algorithm to find multiple equal-cost minimum costpaths while {{cite:943bc0691b7ce03e331f138218d39aef983e9d86}} finds multiple edge-disjoint andnode-disjoint minimum cost paths. RPL {{cite:25ceb1e485103396fbc5b97f0cbdfdb10b4db544}}, arouting protocol currently being developed by the IETF ROLL workinggroup, constructs a DAG routing topology by building a minimum costrouting tree (links from child nodes to \"preferred parent\" nodes) andthen adding redundant links which do not introduce routingloops.The primary design scenario considered by RPL usessingle-path metrics. Other extensions to consider multi-pathmetrics may be possible in the future. In contrast, our URF-DTalgorithm constructs a reliable routing topology by locally optimizingthe URF metric, a multi-path metric that can express the reliabilityprovided by hop-by-hop routing over interleaved paths.", "Another difference between URF-DT and RPL is that URF-DT specifies amechanism to control the order which nodes connect to the routingtopology, while RPL does not. The connection order affects thestructure of the routing topology.", "Finally, the LCAR algorithm proposed in {{cite:d83e7d46fa3d65f560f51178dfcfcf330bbaed58}}for building a routing topology cannot be used to optimize the URFmetric because the underlying link metric (negative log linkprobability) for the URF metric does not satisfy the physical costcriterion defined in {{cite:d83e7d46fa3d65f560f51178dfcfcf330bbaed58}}." ]

[]

[ "There have been a number of earlier works that have, directly or indirectly,shown lower bounds for various models of sparse recovery and certain classesof matrices and algorithms. Specifically, one of the most well-known recoveryalgorithms used in compressed sensing is {{formula:3f30f8c2-157e-4672-ae72-0a6f40c81b4f}} -minimization, where asignal {{formula:b4147c7d-3729-4567-84f2-803ef49985f3}} measured by matrix {{formula:94a0f3b9-00f2-491c-a148-4a20d3d93c82}} is reconstructed as{{formula:61b90d45-8ee5-4c80-8dfb-d739499b40d1}} ", "Kashin and Temlyakov {{cite:8b0700a156118e066faae47fa74b57332d0764a2}} (building on prior work on Gelfand width {{cite:ea10e6cde60100d2eef1b2b6c87a8fab3a71bde0}}, {{cite:b8e6ac9ab707984ef1aa9b0d4a63a7d83288b10e}}, {{cite:2f4d3b5457162a6bcfbbdae8112566c3833ca1f7}}, see also {{cite:5a1d153d48b76c5b5ea863f2ddda00dbf263a587}}) gave a characterization of matrices {{formula:082b5689-2ac5-4d48-a3a8-cb882d0cba2c}} for which the above recovery algorithm yields the{{formula:260179fc-a6f1-40b2-bd7f-fc9659761b70}} guarantee, i.e.,{{formula:b53446ce-f1c0-4e9b-b191-10d4cfc6a0a9}} ", "for some constant {{formula:57764a3a-1040-461d-9596-b539901503a1}} , from which it can be shown that such an {{formula:165a8cc6-3398-4f57-9590-3749f21d3516}} must have{{formula:64dc8f68-564e-465f-8feb-30d0a2e784db}} rows.", "Note that the {{formula:75ef0bf6-7ade-41d0-bf77-b052f7b2f6c7}} guarantee is somewhat stronger than the {{formula:560f4790-266b-454b-8d2e-def2eb630fb8}} guarantee investigated in this paper. Specifically, it is easy to observe that if the approximation {{formula:aa24da97-7471-4af1-9cc3-db570e2ac394}} itself is required to be {{formula:996c6a19-f9fa-4376-860e-f775755b90a9}} -sparse, then the {{formula:859d8059-295b-4520-9a59-cb27687b238a}} guarantee implies the {{formula:1dcf6e55-17c3-4f2e-a745-b93e1d821268}} guarantee (with a somewhat higher approximation constant).For the sake of simplicity, in this paper we focus mostly on the {{formula:2f2e3f22-2966-4b96-b060-29da0b383d9d}} guarantee.However, our lower bounds apply to the {{formula:d9de8eea-d6f5-4010-9f3d-fcc838ec9c9a}} guarantee as well: see footnote on pageREF .", "The results on Gelfand width can be also used to obtain lower bounds for general recovery algorithms (for the deterministic recovery case), as long as the sparsity parameter {{formula:69902691-3993-4f63-b90e-3db49afa36d2}} is larger than some constant. This was explicitly stated in {{cite:40854d9d0d502cd039f54e66bfe77e755345a696}}, see also {{cite:5a1d153d48b76c5b5ea863f2ddda00dbf263a587}}.", "On the other hand, instead of assuming a specific recovery algorithm,Wainwright {{cite:2715917273990b79c3df9ac4c802c16f765ef900}} assumes a specific (randomized) measurement matrix. Morespecifically, the author assumes a {{formula:3caee180-1e12-43bd-8b28-321764e19232}} -sparsebinary signal {{formula:b9a72adf-0b30-42ed-bd20-4828cd6e342e}} , for some {{formula:1e4b7b01-5279-4555-93e4-27d2e077eba8}} , to which is added i.i.d.standard Gaussian noise in each component. The author then shows thatwith a random Gaussian matrix {{formula:4fd636f1-1635-4b0f-8fe2-b1c06b9a4e8a}} , with each entry also drawn i.i.d. from thestandard Gaussian, we cannot hope to recover {{formula:3518fc30-1959-4032-9771-04d25775ddf3}} from {{formula:56f804d5-aaae-449c-a423-87f9ff033c62}} with any sub-constantprobability of error unless {{formula:0d8c352a-227b-48e5-b1dd-abe61512f65f}} has {{formula:679e4c21-279d-4a5f-af36-0c7c80292c27}} rows. The author also shows that for {{formula:ea2089b2-f2a4-487b-ad46-9b9dbbb11130}} , this istight, i.e., that {{formula:e8810d8f-5393-4611-a936-18819e8b3efe}} is both necessary and sufficient. Althoughthis is only a lower bound for a specific (random) matrix, it is a fairly powerfulone and provides evidence that the often observed upper bound of {{formula:ff6815e8-789d-4047-920b-2721a35da20a}} islikely tight.", "More recently, Dai and Milenkovic {{cite:b21cdb1432eb3cef01c271f6b4c6c2effe0cd4ad}}, extending on {{cite:500c55695c19f8f8c5cb20a930c464fad5fde770}} and {{cite:121384b8acf165e20cc51f476f38251264131b4e}},showed an upper bound on superimposed codes that translates to a lower bound on thenumber of rows in a compressed sensing matrix that deals only with {{formula:de5ff6bc-60fd-48c1-8d2d-cd851baabda3}} -sparsesignals but can tolerate measurement noise.Specifically, if we assume a {{formula:247f4ffa-19b7-4853-a6be-60b1b99af223}} -sparse signal {{formula:df57f575-4cdd-4150-9e7a-e5c08381cdbc}} ,and that arbitrary noise {{formula:233e9ac5-d189-4527-9098-3c6b5ed488f8}} with {{formula:08538b1b-cd2c-4cec-a4d0-b46fe7281803}} isadded to the measurement vector {{formula:813095d7-b315-462e-8c37-54f81c513bb0}} , then if exact recovery is still possible, {{formula:2207e2e9-04df-45e5-b231-29ed79131d03}} must have had {{formula:682c8bbb-8556-4712-80a4-ca89d1e10018}} rows, for some constant {{formula:e0df4a98-1520-4df8-b0a4-b1fd52f1c9e4}} andsufficiently large {{formula:dc19ea15-3afb-4d5f-b722-f34d3f0bc1dd}} and {{formula:7e778d5a-a367-49e4-9cfa-e21c0577601d}} .Here {{formula:169e5f7d-cb03-4a0f-bfad-3f4a62db69f4}} is assumed tohave its columns normalized to have {{formula:b2a28f1e-b990-4a69-be4e-5bb57a89ac46}} -norm 1. This is natural sinceotherwise we could simply scale {{formula:182de2ab-bb85-474e-922a-adfb18319478}} up to make the image points {{formula:981b3da3-b923-4c17-b6bf-5c3b915f5ec5}} arbitrarilyfar apart, effectively nullifying the noise." ]

[]

[ "We have greatly improved the performance of real number computation in using 's new machine integers. We produced highly structured and abstract code using type classes with no apparent performance penalty. Moreover, 's notation mechanism combined with unicode characters gives nicely readable statements and proofs. Type classes were a great help in our work. However, the current implementation of instance resolution is still experimental and at times too slow (at compile time).", "Canonical structures provide an alternative, and partially complementary, implementation of type classes {{cite:c79eb664a1e572be9b8e9c88fab498f2cb8ffcf2}}. By choice, canonical structures restrict to deterministic proof search, this makes them more efficient, but also somewhat more intricate to use. The use of canonical structures by the team {{cite:91aff1ddc482050d3dd84eb49fb97dd59d23aafa}} makes it plausible that with some effort we could have used canonical structures for our work instead. However, the -library is currently not suited for setoids which are crucial to us. The integration of unification hints {{cite:88be0387286ad548e4200c093dee1ae4e35303d7}} into may allow a tighter integration of type classes and canonical structures.", "We needed to adapt our correctness proofs to prevent the virtual machine from eagerly evaluating them. Lazy evaluation for would have allowed us to use the original proofs. Moreover, setoid rewriting over relations in would have made our work much easier.", "The experimental |nativecompute| by Boespflug, Dénès and Grégoire {{cite:565ee9249d06584a512469609c5fbc78b874ea27}} performs evaluation by compilation to native code. This approach uses the compiler available and is interesting for heavy compilation. Our first experiments indicate an additional speed up of 3 times compared to |vmcompute|.", "The project {{cite:9c0489d714b9bdb0e8373e61b773c0d6a9b78d73}} formalizes infinitary floating-points in . It provides a library of theorems on multi-radix multi-precision arithmetic and supports efficient numerical computations inside . However, the current library is still too limited for our purposes, but in the future it should be possible to show that they form an instance of our approximate rationals. This may allow us to gain some speed by taking advantage of fine grained algorithms instead of our more straightforward ones.", "The encoding of real numbers as streams of `bits' is potentially interesting. However, currently there is a big difference in performance. The computation of 37 decimals of the square root of 1/2 by Newton iteration {{cite:26fc3cdfd73f2e1ef0340f97d36b1952899878b8}}, using the framework described in {{cite:335bf4074a73406711c60aaa97bdd297f839a738}}, {{cite:0645f56a9008a2911f2a2f585eccccffbb62dc94}}, took 12s. This should be compared with our use of the Wolfram iteration, which gives only linear convergence, but with which we nevertheless obtain 3,000 decimals in a similar time. On the other hand, the efficiency of {{formula:97e3f872-23c1-44cb-97f6-fadf82d8d7bd}} in their framework is comparable with ours. Berger {{cite:274d3361a42c88d434d6ddedb23e680f62cdce59}}, too, uses co-induction for exact real computation.", "The present work is part of a larger program to use constructive mathematics based on type theory as a programming language for exact analysis. This should culminate in a numerical ODE-solver. To do so we need to extend the current technology to functional analysis. For instance we will build a type class interface for metric spaces in order to treat various function spaces.", "Cohen and Mahboubi {{cite:f0b6c308b0bff7151e86afca80120eac91d3ede8}}, {{cite:404a147ce0d8accef11db99a994345383b7c52a2}} provide a formalization of quantifier elimination for the theory of decidable real closed fields, and an implementation of the algebraic real numbers. Quantifier elimination will automate many proofs in constructive analysis which only involve algebraic real numbers.Conversely, our implementation could be used for efficiently evaluating a Cauchy representation of an algebraic real number." ]

[]

[ "Annotations have a long research history, and unsurprisingly the research perspectives and interpretations of what an annotation is supposed to be vary widely. Agosti et al. {{cite:367218acd61c94b2acb4eef452e766d57ac13149}} provide a comprehensive study on the contours and complexity of annotations. A representative discussion on how annotations can be used in various scholarly disciplines in given by Bradley {{cite:ba0d4991ce79d45b19f0af22e45601af371110d9}}. He describes how annotations can support interpretation development by collecting notes, classifying resources, and identifying novel relationships between resources.", "Besides Annotea other annotation models have been proposed: {{cite:fbb87b2f39326e6c6cdf809912b673e86a990e69}} built the MPEG-7 compliant COMM Ontology. OAC, in contrast, is a resource-centric annotation model, which is more light-weight because it doesn't have a background in the automated feature extraction and representation. The M3O Ontology {{cite:afccc219cd79509f394554a2a5316a92fdfd48b6}} allows the integration of annotations with SMIL and SVG documents. OAC, in contrast, treats annotations as first-class resources on the Web, which would not be part but about a presentation.", "Early related work on the issue of describing segments in multimedia resources can be traced back to research on linking in hypermedia documents {{cite:a7467549ea3e596fc94ed7dc61abf8903b56b915}}. For describing segments using a non-URI based mechanism one can use MPEG-7 Shape Descriptors (cf. {{cite:81dd3324d5cc812c41d48b2ea4c9d16a9c0f0152}}) or terms defined in a dedicated multimedia ontology. SVG {{cite:3cbb1294dab518ad7640ff54c15c80395139ceb5}} and MPEG-21 {{cite:b2d223bb1a8e4cb97852011fa35d55e36ad8f93b}} introduced XPointer-based URI fragment definitions for linking to segments in multimedia resources." ]

[]

[ "This work is the first to measure the UV/optical colors of a large sample of AGN host galaxies at {{formula:254a978d-cc49-4d3e-b6c5-7d8baf3cf0c8}} while quantitatively, reliably correcting for point source contamination in individual galaxies. Ignoring or insufficiently correcting for point source contamination results in systematic error in comparing the colors of hard and soft X-ray sources (see Figure REF ).", "Direct comparisons of our results with those obtained for Seyfert galaxies in the local universe are problematic, as numerous AGN selection techniques are commonly used in the literature. At low redshift, the effects of point source contamination on host galaxy color measurements are negligible because the galaxies are well-resolved. Locally, {{cite:bf72f478528d0642db770f833dc78ebe2d1cc46b}} find little host color difference between obscured and unobscured X-ray-selected AGN, in agreement with our results. Although the sample is small, their selection techniques and use of X-ray-derived obscuration measurements are more analogous to those utilized in this study. Given the similarity of our selection criteria, the agreement of our results suggests that the mean color offset between soft and hard sources has not evolved since {{formula:a39b1f42-bd42-49fa-aeef-048fcd1f5941}} .", "{{cite:8068c65b2796bd18d5c2d95ef32925da00a65040}} find that unobscured broad-line AGN hosts at low-redshift are observed to possess large reservoirs of HI gas. This result is interpreted as being evidence against a feedback scenario, in which unobscured AGNs are thought to be caught in the act of expelling gas {{cite:8068c65b2796bd18d5c2d95ef32925da00a65040}}. No comparison is made between Type I and II sources. Considering the use of different selection methodologies and measurements, there is no inconsistency between this study and our own. For a sample of mid-infrared Type II QSOs at {{formula:3b43ebf7-618c-4ec6-95ee-7e5bc40ef3f3}} , {{cite:863e524dfb79a964acb298327331c6abe44210b7}} find that the host galaxies exhibit significant star formation rates ({{formula:6880ba0c-a7e1-40a3-b026-9e467d55c0d1}} yr{{formula:c8f895c9-f913-4372-95d6-43768360ca2f}} ) and that disk inclination correlates with silicate features, implying that at least some of the reddening arises from the host galaxy. However, it is not clear how these star formation rates would compare to similarly selected Type I QSOs at similar redshift. Moreover, the AGN selection techniques, selected luminosity range, and redshift range are different from those in the present study, preventing direct comparison.", "{{cite:3964638362d1392abf10d1081d8c92fda3bad36a}} compares the {{formula:c224b7e9-e565-4144-a691-3f8258f77f20}} m fluxes of luminous ({{formula:75dec30e-3580-4d97-9516-4ba50fe9bbb4}} ergs s{{formula:2ab7fdf8-414f-4421-afbc-38b5a88ad5e2}} ), X-ray-selected AGN at {{formula:be72b612-1bea-4058-abdb-4b0890e87548}} , finding that obscured sources have significantly more submillimeter emission than unobscured sources. As a function of redshift, the mean flux difference between obscured and unobscured sources is consistent with zero at redshifts below {{formula:9bc436c4-8550-4462-b924-96629e565853}} and rises significantly at higher redshifts. Contrastingly, {{cite:acb478c09c24fd89b7a3bfa515d3c7518d010b3b}} measures the far-infrared-derived star formation rates of X-ray selected intermediate-luminosity AGNs ({{formula:0835f2ea-ef95-4e8b-9714-0bebafc53d2e}} ergs s{{formula:eb536e4a-7d5d-4b10-bf92-d4f3eab4f96b}} ) at {{formula:5d224216-7d45-49c2-a44e-fbb0be1b2e08}} in GOODS-North, finding no dependence of star formation rate on X-ray absorbing column density. It appears that the mean star formation rate of obscured sources diverges from that of unobscured sources at either the highest X-ray luminosities ({{formula:46c39dbd-a9d2-47e5-97c8-e94df130da64}} ergs s{{formula:626f75eb-0c1e-4df1-9436-5e249743fbae}} ), at higher redshifts beyond {{formula:28dd6f5c-f405-4eac-b81c-cfd8ee227a8e}} , or in both of these regimes.", "CP10 measures the rest-frame U-V colors of intermediate-luminosity, X-ray-selected AGN in the Extended Groth Strip {{cite:07ebd180b37c95096fe97f9300d0a3bab7ac2201}}. They find that soft sources are systematically bluer than hard sources in both nuclear ({{formula:1fadfeed-8ce4-458c-9b4d-f96eba18048a}} ), extended ({{formula:883745da-63d3-4d4d-9c7f-45742df9b290}} ), and integrated apertures. The mean difference in extended color between soft and hard populations in our GOODS sample differs from the CP10 value by {{formula:5e543a84-5378-4d88-b0da-f398f6883794}} ({{formula:06d67f99-becd-4e25-a956-5039ca9c8604}} mag in {{formula:93442cac-8766-470c-95bd-91dc37bf5815}} ). Our color offsets are consistent to within {{formula:c34419e7-d8eb-4f7f-ac3c-26376e98ef5e}} ({{formula:8b560476-e87e-4ec0-ade4-608270e19e90}} mag in {{formula:d5d0d17c-1a3f-44e4-ad27-c6f7d43bae76}} ) when the CP10 sample is trimmed of all sources that they identify visually as having “definite” or “possible” point sources. In addition, color offsets are similar when no correction for point source contamination is used in our sample ({{formula:b9f33a38-0002-432a-a6b6-4b43183e3d2b}} mag in {{formula:8446faf5-4163-47a0-88d5-0540f16faaac}} ; see Figure REF and Table REF ). In CP10's Figure 11d it is clear that sources visually identified as possessing “definite” or “possible” point sources display anomalous color gradients (i.e., blue nuclear regions). These lines of evidence point to some amount of point source contamination in CP10's extended colors.", "{{cite:2df46d31605ffbf7595e51696069b6fb5ceb6f27}} find that X-ray selected AGN hosts at {{formula:a5512280-8fee-46ec-9331-c4d175783c25}} are frequently in dust-enshrouded galaxies. Approximately one-half of our sample occupy a region in the rest-frame {{formula:753ec637-7aac-4867-a82a-6d5b0f603ebb}} vs. {{formula:1c99d212-d27a-4a1f-a7d9-fb55d5375c85}} diagram corresponding to {{formula:6f03307c-89f7-4653-98db-51eeeb764e51}} , so our results are not inconsistent. Although the authors do not explicitly compare the locations of obscured and unobscured sources in this plot, it is clear from their Figure 2 that the mean locations of these populations are similar, as in our sample." ]

[]

[ "Most of operational grammars for mainstream software languages are handcrafted, many are not publicly disclosed, few are documented. The firstcase reported in detail in 1998 was PLEX (Programming Language for EXchanges), a proprietary DSL for real time embedded software systems byEricsson {{cite:c2a4fe46414b5553917946e829901b414311248d}}, a successful application of the same technology to COBOL followed {{cite:c978f54fa73ff636a56e2f6b701c9dc812828666}}. Grammar recovery technique isnot only needed for legacy languages, examples of more modern and presumably more accurately engineered grammars being nontrivially extractedinclude C# in {{cite:e7c215fe07f355ed54eded723a17a7a112011873}} and {{cite:874857437fee2018fd660b6e49243eda5ac74725}} and Java in {{cite:1e73129b00503cc8b34d757a6de1bd90166ec54f}} and {{cite:d164141bf125da79a68c993923e9f6cbc924f34e}}. The wholeprocess of MediaWiki grammar extraction is documented by this report, all corrections and refactorings are available online, as is the endresult (under CC-BY-SA license).", "Wiki Creole 1.0http://wikicreole.org is an attempt for engineering an idealwiki syntax and a formal grammar for it. While the goal of specifying the wiki syntax with a grammaris not foreign to us, but the benefits listed in {{cite:b3cfffdf1c311d005b7a3735177282e6eb40e331}} are highly questionable:", "1. Trivial parser construction.In the paper cited above it is claimed that applying a parser generator is trivial.However, the main prerequisite for it is successful grammar adaptation for theparticular parsing technology {{cite:7ee945ffeea1424a5c2df2ac38b6c02bcc5c8db2}}. A Wiki Creole grammar was specifically geared toward ANTLR,and it is a highly sophisticated task to migrate it anywhere if at some point ANTLR useis deemed to be undesirable.Hence, the result is not reproducible without considerable effort and expertise.2. Foundation for subsequent semantics specification.The grammar can certainly serve as a basis for specifying semantics.However, the choice of a suitable calculus for such semantics specification is of even more importance.Furthermore, syntax definition does not guarantee the absence of ambiguities in semantics,or even changes of semantics as a part of language evolution (cf., evolutionary changes of HTML elements).3. Improved communication between wikiware developers.The paper claimed that if wiki syntax is specified with a grammar, there can be no different interpretations of it.However, it is quite common to have different interpretations (dialects) of even mainstream programming languages,plus wiki technology in its current state heavily relies on fault tolerance (somewhat less so in the future when no bare textediting should be taking place).4. Same rendering behaviour that users rely on.Depending on the browser or the particular gadget that the end user deploys to access the wiki, rendering behaviourcan be vastly different, and this has nothing to do with the syntax specification.5. Simplified syntax extension.It is a very known fact in formal grammar theory {{cite:442baef75e7185d0c670fd3646cf855fe95a9477}} that grammar classes are not compositional: that is, the resultof combining two LL(*) grammars (which ANTLR uses) does not necessarily belong to the LL(*) class; we can onlyprove that it will still be context-free {{cite:f05a651d51add30ca27dc6bfd027608efdcb05e8}}. In other words, it is indeed easy to specify a syntax extension, but suchthe extended grammar sometimes will not be operational.Modular grammars can be deployed in frameworks which use different parsing technologies, such as in Meta-Environment {{cite:215657fa1705f891c6be89d6757d60771fe12508}}or in Rascal {{cite:3a862aded63e2b4aefb1ca6236b318d06e61b97d}} or in MPS {{cite:b2e90b46a400abb93637b45160a11bb37829e960}}, but not in ANTLR.6. Performance predictions.The paper claims that it is easier to predict performance of a parser made with “well-understood language theory” than witha parser based on regular expressions. However, there are implementation algorithms of regular expressions that demonstratequadratic behaviour {{cite:e58a55fcc20c84922c5ca6995257d5b1855a3141}}, and ANTLR uses the same technology for matching lookahead anyway, which immediately meansthat their performance is the same.7. Discovering ambiguities.It is true that ambiguity analysis is easier on a formal grammar than on the prose, but it is not achieved by “more rigorous specificationmechanism” and even the most advanced techniques of today do not always succeed {{cite:3a756a9276acde6157627beb4b01a2fae5385d04}}.8. Well-defined interchange format.A well designed interchange format between different types of wikiware is a separate effort that should be based on appropriategeneralisations of many previously existing wiki notations, not on one artificially created one, even if that one is better designed.", "In general, Wiki Creole initiative is relevant for us because it can serve as a common grammar denominator later to converge several wikigrammars {{cite:a33b7f049d2671ecefb02049fc3d0d2857828fce}}, {{cite:2d5918b056fe382e210f4e2d2d30fbce9c6b4f09}}, but is neither contributing nor conflicting directly with our grammar recovery project." ]

[]

[ "Discrete approximation methods are commonly used in continuous and hybrid systems analysis (see {{cite:298531a47fd30f2557f18e439094f4ee33d0f2ce}} for an overview regarding reachability) to handle the uncountability of the state space. Direct methods work on the original system and rely on a successor operation iteratively computing the reachable set whereas indirect methods abstract from the continuous model by a finite structure for which the analysis is simpler. Our method belongs to the latter class, sinceit uses numerical simulations and creates the abstraction automaton. Considering a fixed set of initial conditions, there is a certain overhead with generating states of the automaton in comparison with simple numerical simulations.However, the advantage of constructing the automaton is obtaining a global view of the dynamics. Moreover, in addition to rectangular abstraction, the automaton is augmented with weighted transitions which represent quantitative information describing volumes of subsets of initial conditions belonging to attraction basins of different parts of the phase space.", "An indirect method based on rectangular abstraction automaton making the finite quotient of the continuous state space has been employed, e.g., in {{cite:d2eb3b2216d0bd406539f917524683e6d7fa66e5}}, {{cite:9ffb747f166e375523913b073ef4c3ac057fcc4f}}, {{cite:c3f14e01847e27d548de59b422549c7840e61fbc}}. In general, these methods rely on results {{cite:5250223a05f0aac54a129edbfd621e7a6c221429}}, {{cite:f8cce4e977aadb58baa2fa5a759556a31bf0d378}} and are applicable to (piece-wise) affine or (piece-wise) multi-affine systems.Although not addressed formally in this paper, our technique can be considered as a refinement of {{cite:d2eb3b2216d0bd406539f917524683e6d7fa66e5}}. However, we focus on obtaining satisfactory approximate results eliminating the extent of spurious behaviour coming from conservativeness of rectangular abstraction.Our technique can be employed for the recognition of spurious behaviour of the rectangular abstraction transition system.", "The technique presented in {{cite:62d676136447c90088183e943e92e68ed02dcbd2}} employes timed automata for the finite quotient of a continuous system as an alternative to piece-wise linear approximations. Another indirect technique adapted to multi-affine biological models is {{cite:131da993155297d80bdb4ca5effebe897f481ee5}}. The approach also employes rectangular abstraction, but results in less conservative reachable sets by means of polyhedral operations. In {{cite:298531a47fd30f2557f18e439094f4ee33d0f2ce}}, {{cite:4ddfcd19e3589170d16fec77acc8da1d823de1f3}} there are techniques proposed for rectangular refinement that go towards reduction of over-conservativeness. These techniques work fine for linear systems while leaving the non-linear systems as a challenge.", "Direct methods are mostly based on hybridization realized by partitioning the system state space into domains where the local continuous behaviour is linearized {{cite:70f41c341dbd9c54b78fd8d74bc2463a22902fb7}}. This method, in an improved form, has been applied to non-linear biochemical dynamical systems {{cite:384d2f439aad64d20f37536798f9e15d28430d74}}. In general, direct methods give good results for low-dimensional systems and small initial sets. In comparison with indirect approaches, they are computationally harder. From this viewpoint, our approach lies between both extremes." ]

[]

[ "There exists cloud-based systems that enable data sharing frommultiple sources. SenseWeb {{cite:4c0b8a8384e0d0ac13987680d7cdf94b361985c6}},SensorBase {{cite:0747ce2048ea7ea547f58b8fa11281b082104016}} are examples of cloud services that letusers upload and share their sensor data. They support coarse-grainedaccess control model in which an user either makes its dataset public,shares it with a list of collaborators or keeps it private. Similarly,Google's Fusion Table {{cite:c57fe8864b1bbbe2bd65fbbcddb80c2b91ce4141}} allows user to upload genericdata and to perform simple analysis such as data visualization on thecloud. Recently, companies such as Okta {{cite:7618f34b2d0ae1df50e0b31e6523f0ed5daab8be}} have startedimplementing cloud-brokerage models that provide centralized servicefor management of enterprises' resources, including access control.However, these access control model is also coarse-grained, whichmeans it cannot deal with the access scenarios we consider in thispaper. In addition, data owners in these systems upload their datasetsonto a centralized cloud, whereas our work does not make suchassumption (we consider multi-cloud environment in which differentdata owner uses its own cloud provider).", "There are also numerous works focusing on access control and dataprivacy on the cloud. Airavat {{cite:21f34cd5ba6d8456366bf5dcc0dae33d50f72f86}}, for example, assumes thecloud is trusted in enforcing access control. It uses a simplemandatory access control system available in SELinux {{cite:10d72fdf68f256ed58e4b5a9b1b04eb12c80f3ec}},and provides a trusted environment for executingMapReduce {{cite:957916ffca98e42c62fd145205ed2d8eab2a5ab2}} jobs while guaranteeing differentialprivacy {{cite:3d3dbf1cf42dca6ef8b6db2fa0af166628ea2e98}}. Our work makes the same assumption aboutclouds' trustworthiness, but aims at improving the access controlaspect of the system, which is complementary to Airavat. Otherworks {{cite:e3ef7a0e4f5196a0bef89a646b7452e90a4cb868}}, {{cite:b34989430560e13d79345d89f61586f2b2d1f892}}, {{cite:e3aa542243c0cfd52cc8c70b20a1057cefd878f2}} assume the cloud is untrustedand employ cryptographic approach for access control. In {{cite:e3ef7a0e4f5196a0bef89a646b7452e90a4cb868}},data is encrypted with attribute-basedencryption {{cite:3f39303e37ae89ac9de4c71a2e74d473db4466d9}}, {{cite:5c2042c46798ff41b0219232e45cafe138b7ac94}} by a proxy using a proxyre-encryption technique. Embedded in the ciphertext are conditionsthat must be met when decrypting. Plustus andCloudProof {{cite:b34989430560e13d79345d89f61586f2b2d1f892}}, {{cite:e3aa542243c0cfd52cc8c70b20a1057cefd878f2}} use broadcastencryption {{cite:9ef210893df0e0e765395dd8ce206468dd501c16}} to protect the data, while keymanagement {{cite:b34989430560e13d79345d89f61586f2b2d1f892}} is done using key rolling and lazyrevocation techniques. These cryptographic approaches provide strongguarantees for data security, but they cannot express fine-grainedaccess control policies as described in our work. Thus the focus in these works is also complimentary to ours. In addition, keymanagement and revocation protocols are complex and incur muchoverhead in such an untrusted environment.", "Multiple policies matching in XACML is usually resolved by thetop-level policy combining algorithms. XACML supports only a limitednumber of combining algorithms. Ninghui et al. {{cite:8f16ff9affae4ae11a40b740bab68bd6a55392d6}} andRao et al. {{cite:8d958bc7af2e6a17e99ca37bc6791e43b647a6d3}} propose a formal language for expressing morefine-grained policy composition. The language can deal with evaluationerrors and combining of obligations. Mazolleni etal. {{cite:fa735fbc0c55f2fb42f10bd33ef1641632029f85}} propose a method for combining policies basedon their similarity and users' preferences.", "Time-series data — similar to those considered in our paper —could arrive at the system in continuous streams, for which relationaldatabases such as MySQL and Postgresql are not ideal.Aurora {{cite:b130dd7406f253f3c77b02dc10bcfde79cc55148}} is a popular data stream management system thataddresses limitations of relational databases when it comes to streamdata. Carminati et al. {{cite:e310c65bedd5ef20f9291e03aca5c66daf9d543f}}, {{cite:3d46e0f66fe7fc9671c6f95a95ac674896b31360}} are among thefirst to propose a model and implementation of access control for datastreams based on Aurora. The model supports four access scenarios:column-based, value-based, general window and sliding window. Ourframework supports all of these scenarios for on-demand queries overarchival databases. The extension to eXACML that deals with continuousqueries over stream databases is left for future work." ]

[]

[ "In this section we give an overview of research that contributed to the analysis and systematization of cross-context communication systems. Most previous research activities were focused on evaluating and comparing security properties of cross-context communication systems based on browser primitives not intended for cross-context communication. Other research was focused on designing new browser primitives and client-side libraries that overcome specific security deficiencies.", "One of the earliest critiques of the lack of secure browser primitives for cross-context communication was given by the proposal of the {{formula:d0b5676c-5d89-4a75-9370-fd05a4e29beb}} module{{formula:e45d7cc8-bb73-4777-beee-bc4965178dbc}} HTML tag and API {{cite:3a14363106931fa4866a64793733414800a5c4d2}}, {{cite:1f61895c5154ecd437182742d16cbfe53b738de2}}. Although it was never standardized or implemented, the proposal inspired research of the later standardized HTML5 postMessage API {{cite:8c726a8b705d17d2fc2746f0b7ef03207c76f4e7}}. A list of several similar early research proposals is given in {{cite:c832532ab6d9860a7b13f982b2c6d2598af608a5}}. In {{cite:3a14363106931fa4866a64793733414800a5c4d2}} two techniques for communication between window contexts are analyzed with respect to confidentiality and authenticity; fragment identifier messaging (FIM) and higher-level protocols based on FIM, and the postMessage API.", "In {{cite:c47221fd9716bddb40717638a586a347f52e049a}} the authors analyze whether existing and proposed browser mashup communication primitives enable communication between two principals, browser contexts for example, without ceding complete control to each other. Vulnerabilities of primitives are illustrated through several proof-of-concept attacks and recommendations for prevention are given. Furthermore, an evaluation of design choices for access control aspects of communication primitives is given; for example, using values versus objects for communication. In {{cite:a4ce570ea08bc93a461091014c4e911e13dbc7d9}} the authors demonstrate that the existing browser security model was not designed to support multi-context Web applications, and that as a consequence these are typically implemented insecurely. The paper also emphasizes the need for higher-level communication abstractions and presents a secure component model based on a publish-subscribe communication abstraction. In {{cite:4ea702df12654715810193d4e36dacb709927d9a}} a critique of cross-context communication based on browser cookies is given with respect to security. The paper also recognizes the security disadvantages of using server-side proxies for cross-context communication. Furthermore, a proposal for a secure publish-subscribe communication system is given. In {{cite:0f94b1ab8207e53837e2572885c050b60ed492e4}} the authors analyze incoherencies in browser access control policies. A special part of this analysis are browser resource types which may be shared among principals and their interaction, which is a subset of cross-context communication systems.", "In {{cite:dc3e3b09f788f08a8b4821f4027295c6d440671e}} the authors analyze inter-widget communication, a specific application of cross-context communication, with the purpose of maximizing usability of widget-based personal learning environments. The analysis is based on a framework for categorization of inter-widget communication systems, an approach similar to the one we present later in this paper. For example, the authors distinguish between same-browser and cross-browser communication, inter-widget and intra-widget communication, several types of event distribution, such as broadcast and direct subscription, and several types of security and semantic interoperability. However, the framework is strongly focused on end-user usability of inter-widget communication and does not take into account many technical dimensions of cross-context communication, such as discovery, cross-origin support and reliability.", "The most recent and broad analysis of browser primitives for Web application interactivity is given in {{cite:24770350e532c4619064d17a7ecdc903dfb8a8b0}}. The authors present results of a usage analysis of several new browser primitives implemented in browsers as a part of the HTML5 group of standards. The results show that the postMessage API, Web Storage API {{cite:aa7bc160421bcde2a55c79c32f3dd711c33a422a}} and the Web SQL Database APIs {{cite:c1cfc9988906897d0544a70d926cba786d50884b}} are being used insecurely. Moreover, the authors give insights into why these primitives can potentially be hard to use safely and propose the economy of liabilities principle in designing security primitives - a primitive must minimize the liability that the user undertakes to ensure application security. The authors also propose several enhancements to the postMessage API to shift the burden of verifying and ensuring security properties from the developer to the browser.", "Another relevant field of research is secure architectures for modern Web browsers. In this research field, the browser and management of Web applications are observed at a lower level with regard to operating system integration and inter-process communication. Still, this work is the foundation for implementing communication abstractions at the Web application level. The security architecture of the open-source Chromium browser is described in {{cite:5c46ffef37242a194ba184ddea32c17272afa043}}. In Chromium, the traditional monolithic architecture of browsers is replaced with an architecture based on two modules in separate protection domains: a browser kernel, interacting with the operating system, and a rendering engine, executing with restricted privileges in a sandbox. The authors describe the security advantages of the architecture and describe how other architectures make it difficult to implement cross-context communication primitives. The architecture of Gazelle, a secure Web browser constructed as a multi-principal operating system, is presented in {{cite:e72ebd87f726c588533270526f873224b057cc10}}. Gazelle's security model protects principals, contexts from different origins, by separating their resources into hardware-isolated protection domains. The authors describe benefits of aligning the browser architecture with the SOP policy and analyze cross-principal interaction possibilities.", "Lastly, operating systems (OS) {{cite:151f5cdb1c02ccf5bc7a5e53743411ab21a20ee2}} have been extensively researched through the last decades providing valuable experience for designing browsers as multi-context execution environments. Specifically, multi-process applications and inter-process communication mechanisms designed for OSes provide a starting point for cross-context browser communication research. However, these mechanisms were seldom taken into account in previous cross-context communication research.", "In conclusion, cross-context communication is still an unresearched field and trailing behind industry requirements. First, since SOP has been a major issue for the last several years, research has been focused mainly on security aspects of cross-context communication and disregarded other aspects. Second, since until recently only window contexts were in use, no prior research includes worker contexts in their analyses. Third, many existing systems for cross-context data exchange have similarly received little attention. Fourth, there is no systematic approach aimed at analyzing cross-context communication or a broad systematization of existing systems. Fifth, existing operating systems IPC mechanism research has not been integrated into cross-context communication research. Our systematization presented in the next section addresses some of the stated challenges." ]

[]

[ "The classic two-user interference channel consists of twotransmitter-receiver pairs communicating in the presence of interferencefrom each other. Although the capacity region of the two-user Gaussianinterference channel is still not known exactly, it can beapproximated to within one bit {{cite:2244edb2b8464005645e342e6d143d4141c7fc91}} using aHan-Kobayashi power splitting strategy {{cite:c6c76721d98e9fec708e534670fa411effe24404}}.", "The use of cooperative communication for interference mitigation hasreceived much attention recently. For example, {{cite:9343d0864ad067c9685a25ffece5830ba3a3c043}}, {{cite:7293b71e1c74d3d9a3f95c4fa23ed07ee4d9e00b}}, {{cite:a4ef032595e481c425e59ac106b0be4507e62fbc}}studied the Gaussian Z-interference channel witha unidirectional receiver cooperation link, and{{cite:da450d4f8a39a2f2b832d7ea484028a081fca1b6}}, {{cite:2ff16d661105cbf91be0e408620544405366dcfb}}, {{cite:174149244d4bce8ae8ad3642abf721ae4db416c4}}, {{cite:50f40bd961e46fb56e2879a064325ffe0cb46451}} studied theGaussian interference channel with bi-directional transmitter/receivercooperation links.In addition, the Gaussian interference channel with an additionalrelay node has also been studied extensively in the literature.Depending on the types of the linksbetween the relay and the transmitters/receivers, the relay-interferencechannel can be categorized as having in-band transmission/reception{{cite:cf8e7934b91b85ca60637091a3ba8bbdce5de28d}}, {{cite:e6d1278111b27788d50c8a3b528a9a8612911fb5}}, {{cite:ab1670501cc55ef1e2a88dd19117c30e27f5465f}}, {{cite:730307cdc3b3dc9363cce6c6ecd4d0da01642bd6}}, {{cite:b3a5e52f46170f9e06c0ddcdc6d9519e303cb34b}}, {{cite:3caf649e818e5190d44246bc2ead5eee87f96117}}, {{cite:db87f1305cb614c22f2b1193d4ebbebe9595744b}}, {{cite:cf8e7934b91b85ca60637091a3ba8bbdce5de28d}}, out-of-band transmission/reception{{cite:4873c1d7be73dbc9fb87fb7212999483c6d3b305}}, {{cite:61d4e98b50e6450379d3235ef49c79a469421c88}}, {{cite:cd0f1d4077195225ca559d87f787353fdf9572a2}}, out-of-bandtransmission and in-band reception {{cite:6f2448d47d6836c2febdea6bc6bb385cacbd5037}}, {{cite:7dc74224a36a4fc369c0b96c3793e0177804cbbc}}, {{cite:7aa2224c0247c04409763a896aa9f34e232f37b8}}, {{cite:609691516114bc9c2a21bf421190bf05a27bff8a}}, orin-band transmission and out-of-band reception {{cite:8462d284d0021ff969f23b6d7ea4dfec4d0640ab}},the last of which is directly related to the channel model in this paper.In the following, we review differenttransmission schemes and relaying strategies that have emerged foreach of these cases.", "For interference channels equipped with an in-band transmission andreception relay, the relay interacts with both transmitters andreceivers in the same frequency band. Relaying strategies that havebeen investigated in the literature include decode-and-forward,compress-and-forward, and amplify-and-forward. For example,{{cite:ab1670501cc55ef1e2a88dd19117c30e27f5465f}}, {{cite:730307cdc3b3dc9363cce6c6ecd4d0da01642bd6}} show that decoding-and-forwardingeither the intended signal or theinterfering signal to a receiver can both be beneficial. The former is termed as signal relaying, the latter interference forwarding. Decode-and-forward and half-duplexamplify-and-forward strategies are also studied in{{cite:b3a5e52f46170f9e06c0ddcdc6d9519e303cb34b}}, {{cite:db87f1305cb614c22f2b1193d4ebbebe9595744b}}. When combiningdecode-and-forward relaying strategy and the Han-Kobayashi ratesplitting input scheme, {{cite:cf8e7934b91b85ca60637091a3ba8bbdce5de28d}} gives an achievable rateregion that has a shape similar to the Chong-Motani-Garg (CMG) region for the interference channel {{cite:5d78d85b5c4118b57ae17556cbdaa56bbd5ecd89}}.The exact capacity for this type of relay-interference channel is in general open, but thereis a special potent-relay case {{cite:e6d1278111b27788d50c8a3b528a9a8612911fb5}} for whichthe sum capacity is known in some specific regimes.", "The difficulty in establishing the capacity of the interferencechannel with in-band transmission/reception relay is in part due tothe fact that the relay's received and transmit signals intertwinewith that of the underlying interference channel.To simplify the matter, the interference channel with an out-of-bandtransmission/reception relay has been studied in{{cite:4873c1d7be73dbc9fb87fb7212999483c6d3b305}}, {{cite:cd0f1d4077195225ca559d87f787353fdf9572a2}}, {{cite:61d4e98b50e6450379d3235ef49c79a469421c88}}. In this channel model, the relay essentiallyoperates on a separate set of parallel channels. Based on signal relaying andinterference forwarding strategies, {{cite:4873c1d7be73dbc9fb87fb7212999483c6d3b305}}identifies the condition under which the capacityregion can be achieved with separable or nonseparable coding betweenthe out-of-band relay and the underlying interference channel.Further, {{cite:61d4e98b50e6450379d3235ef49c79a469421c88}} studies this channel model in asymmetric setting and characterizes the sum capacity to within {{formula:27b62faf-9249-4850-9695-5964d59441d9}} bits. The transmission scheme of {{cite:61d4e98b50e6450379d3235ef49c79a469421c88}}involves further splitting ofthe common messages in the Han-Kobayashi scheme and a relay strategythat combines nested lattice coding and Gaussian codes. It is shownthat in the strong interference regime, the use of structured codes isoptimal.{{figure:6175b241-b6d0-4e6a-b124-13515abecf36}}", "Another variation of the relay-interference channel involves an out-of-band reception and in-band transmission relay. This channel is studied in {{cite:6f2448d47d6836c2febdea6bc6bb385cacbd5037}}, in which the transmitterfurther splits the transmit signal according to the Han-Kobayashi scheme; the relay decodesonly part of the message depending the capacity of thetransmitter-relay links; the rest of the codewords are transmitteddirectly from the sources to the destinations without the help of therelay. With this partial decode-and-forward relaying scheme, the sumcapacity is found under a so-called strong relay-interferencecondition.", "The interference channel with an in-band reception/out-of-bandtransmission relay has been briefly discussed in {{cite:8462d284d0021ff969f23b6d7ea4dfec4d0640ab}},and studied in {{cite:7dc74224a36a4fc369c0b96c3793e0177804cbbc}}, {{cite:7aa2224c0247c04409763a896aa9f34e232f37b8}} for a case where therelay-destination links are shared between the two receivers.Conventional decode-and-forward and compress-and-forward relaystrategies are not well matched for helping bothreceivers simultaneously with a common relayed message. Thus,{{cite:7dc74224a36a4fc369c0b96c3793e0177804cbbc}}, {{cite:7aa2224c0247c04409763a896aa9f34e232f37b8}} consider a generalizedhash-and-forward (GHF) strategy, which generalizes the conventionalcompress-and-forward scheme, and is shown to achieve the capacityregion of this channel model to within a constant number of bits in the regime where the shared relay-destination link rate issufficiently small. The channel model under consideration in thispaper further extends the shared relay-destinations link to be a degradedbroadcast channel. We focus on a different weak-relay regime. The main objective is similar: to efficientlyuse the relay bits to simultaneously benefit both users and to achieve capacity to within a constant gap.", "Finally, the GHF relay strategy used in this paper is essentially thesame as the noisy network coding {{cite:048f188678648b40b4aa0772678a4a3106046166}}, {{cite:7d046a4f9b52273b505428a48a086641141a11b2}}, {{cite:5fc46388c6615c0bed2ff3b462b4fbdaa89fc47d}} and thequantize-map-and-forward relay strategies {{cite:b2f9d8c95a90cc9c4a6111f11a8034e3a0a7323e}}. The result of this paper can be thought of as an effort in generalizing these relay strategies to a particular case ofthe multiple unicast setting, for which constant-gap result continues tohold for certain channel-parameter regimes. Relatedworks for the multiple unicast problem include {{cite:c56237a135e17b42d0c861121c35dd30dc6993ff}}, {{cite:08dda488aa4f4d0c459e380cd51a4d9a6e53b1d9}}, {{cite:87affb03efb44f38ac5cb69c6b9a8fe9d8a8d954}}." ]

[]

[ "Much of the work on sparse recovery has relied on the RestrictedIsometry Property {{cite:3ee6b7bbe0e376c6c5dd2a9813861bc78fe655d6}}. None of this work has been able toget better than 2-approximate recovery, so there are relatively fewpapers achieving {{formula:51f6177d-301c-47e7-b8b3-ebb07e01436f}} -approximate recovery. The existing oneswith {{formula:9c6e396d-e86e-4e4f-9bf4-b968cf406fea}} measurements are surveyed above (exceptfor {{cite:5b00bc5e0dfd9bf9abbe5e871ea1fa87dccbca28}}, which has worse dependence on {{formula:4b9bf526-6500-427b-abd8-84e524337522}} than {{cite:fa25f415194eacd016296d55c0d8591d809eca64}}for the same regime).", "A couple of previous works have studied the {{formula:d01217e6-2271-4abb-b80d-65538cd08f45}} problem, where every coordinate must be estimated with small error.This problem is harder than {{formula:5a670e51-5e61-4c00-8cb8-81d9b42d04e1}} sparse recovery withsparse output. For {{formula:c43f7055-4de6-48b2-bf03-a01cc000c58b}} , {{cite:ab9746134ac0bcd224d687895b427f3a8c9ad4f0}} showed that schemes usingGaussian matrices {{formula:54a6c68d-7ece-4841-9da6-69261d43dd0b}} require {{formula:4170f894-81e3-44f3-9e7f-db85c5ceb27e}} . For {{formula:a2782a87-e2e4-4403-bd5a-87f96febfbbf}} , {{cite:6d1861dbd7f7f151a3f9b72798f18022cb089cd3}} showed that any sketch requires{{formula:5e61dff6-50f7-48b1-aea8-aab31aa6024c}} bits (rather than measurements).", "Independently of this work and of each other, multipleauthors {{cite:9d380cd5807e077dfdf0b2b0f56003b30ba11351}}, {{cite:51f1ed79107ee014edd7c143f64f5acf7b72d07b}}, {{cite:81038c556be5d765e1389ad5cd4567a876ccac12}} have matched our{{formula:dcaaceb5-fedf-4ed4-ba12-c78714683fc5}} bound for {{formula:2eac63ee-994a-4a0e-8a79-c24bb87771b2}} inrelated settings. The details vary, but all proofs are broadlysimilar in structure to ours: they consider observing a large set of“well-separated” vectors under Gaussian noise. Fano's inequalitygives a lower bound on the mutual information between the observationand the signal; then, an upper bound on the mutual information isgiven by either the Shannon-Hartley theorem or a KL-divergenceargument. This technique does not seem useful for the other problemswe consider in this paper, such as lower bounds for {{formula:eab7ab55-ad4c-4895-9f2e-0e5b283d293a}} orthe sparse output setting." ]

[]

[ "The global slowdown of Wikipedia's growth rate (both in the number of editors and the number of edits per month) has been studied {{cite:77fc97c627bac9d7d3a1e30c71c85969519b86ef}}. It is found that medium-frequency editors now cover a lower percentage of the total population while high frequency editors continue to increase the number of their edits. Moreover, there are increased patterns of conflict and dominance (e.g., greater resistance to new edits in particular those from occasional editors), which may be the consequence of the increasingly limited opportunities in making novel contributions. These findings could guide us to generate other kinds of useful features to tackle the problem of edit number prediction. Furthermore, researchers have also investigated other activities of Wikipedia's editors, such as voting on the promotion of Wikipedia admins {{cite:58d675fa3912eb1d7f4c7cca8691c6c979a2fc33}}.", "In addition to Wikipedia, the temporal dynamics of online users' behaviour has been explored and exploited in web search {{cite:43e8e5e3eddd05b8b188aa0b5a9bb125cc73bbe9}}, {{cite:c51165f9a6b7c7404970d2ce9cf9d50af6cae2c0}}, {{cite:7d186076b0cd2177163f2a4018a257bca2f03722}}, {{cite:fec5631cbfdfe1b4fec8063881369e86d6fa723d}}, social tagging {{cite:e193d807298aa0095560169e03709395d521413a}}, {{cite:31d05c4e306197af9b3c567ec95829df8479fc77}}, blogging {{cite:e2a49ffc985326fe13003878a49a4ec0a4ab667a}}, twittering {{cite:191f3b90748e41558c937d64167dd79e1accbcb0}}, and collaborative filtering {{cite:e03befa8d1d04e6a6593b0009aac1bdd1fcca315}}. The power law {{cite:03a8f96a663de32e1198034adcbd9ab4436f2be9}} and the exponential decay {{cite:d537f04c8dca55867bd618d3d62ff699ac067dd1}} seem to be recurrent themes across application domains." ]

[]

[ "Recipe recommendation has been the subject of much prior work. Typically the goal has been to suggest recipes to users based on their past recipe ratings {{cite:fac3a934b9993aff6a49af1cf72a205af9e56cca}}{{cite:ba909f9d57d58b8060ff9d32bb9d724d0c5df625}} or browsing/cooking history {{cite:ff319eb9a09336a4a454887ce6964d31eb4ab7c3}}. The algorithms then find similar recipes based on overlapping ingredients, either treating each ingredient equally {{cite:4df387e82bd851cc9b754d3ce7692d8660703a77}} or by identifying key ingredients {{cite:d8be7eb4af62e1627f41a312d78805c2522bd99f}}. Instead of modeling recipes using ingredients, Wang et al. {{cite:966723a51f9e4ae901a480ac08fb2b95be07a5e9}} represent the recipes as graphs which are built on ingredients and cooking directions, and they demonstrate that graph representations can be used to easily aggregate Chinese dishes by the flow of cooking steps and the sequence of added ingredients. However, their approach only models the occurrence of ingredients or cooking methods, and doesn't take into account the relationships between ingredients. In contrast, in this paper we incorporate the likelihood of ingredients to co-occur, as well as the potential of one ingredient to act as a substitute for another.", "Another branch of research has focused on recommending recipes based on desired nutritional intake or promoting healthy food choices. Geleijnse et al. {{cite:53f472b207e93e0cc5e5e7b98b71aacc7d6a6c28}} designed a prototype of a personalized recipe advice system, which suggests recipes to users based on their past food selections and nutrition intake. In addition to nutrition information, Kamieth et al. {{cite:2c310ba131b4e09272a11d7c8aa809274c60d818}} built a personalized recipe recommendation system based on availability of ingredients and personal nutritional needs. Shidochi et al. {{cite:398993f02e46fa4ab35e1e18494774fc7bce1133}} proposed an algorithm to extract replaceable ingredients from recipes in order to satisfy users' various demands, such as calorie constraints and food availability. Their method identifies substitutable ingredients by matching the cooking actions that correspond to ingredient names.However, their assumption that substitutable ingredients are subject to the same processing methods is less direct and specific than extracting substitutions directly from user-contributed suggestions.", "Ahn et al. {{cite:df5fa35836266376de31cb4413a3b269e14d2d1b}} and Kinouchi et al {{cite:e04d6e538a988279b87097e20d4133151da24ed1}} examined networks involving ingredients derived from recipes, with the former modeling ingredients by their flavor bonds, and the latter examining the relationship between ingredients and recipes. In contrast, we derive direct ingredient-ingredient networks of both compliments and substitutes. We also step beyond characterizing these networks to demonstrating that they can be used to predict which recipes will be successful." ]

[]

[ "Plate{{cite:bca683fddcb74b1949a588a53f6953ea270c8cc9}} constructs a form of additive GP, but using only the first-order and {{formula:d283ce99-6398-493a-b6d4-69223006a16f}} th order terms. This model is motivated by the desire to trade off the interpretability of first-order models, with the flexibility of full-order models. Our experiments show that often, the intermediate degrees of interaction contribute most of the variance.", "A related functional ANOVA GP model{{cite:a5c635f5474db3d280c827e5218c5bd0bd4131f1}} decomposes the mean function into a weighted sum of GPs. However, the effect of a particular degree of interaction cannot be quantified by that approach. Also, computationally, the Gibbs sampling approach used in {{cite:a5c635f5474db3d280c827e5218c5bd0bd4131f1}} is disadvantageous.", "Christoudias et al.{{cite:1360dac379edff07faa3be7a10a0bb8573214031}} previously showed how mixtures of kernels can be learnt by gradient descent in the Gaussian process framework. They call this Bayesian localized multiple kernel learning.However, their approach learns a mixture over a small, fixed set of kernels, while our method learns a mixture over all possible products of those kernels." ]

[]

[ "Due to its powerful action-based reasoning ability, situation calculus is often chosen as the formalism to express other models and programming languages which are either too complex to understand and use, like Artificial Intelligence in games {{cite:21973491fdeb551eab89c387c88dc8209e1cfc30}} and Planning Domain Definition Language (PDDL) {{cite:0db7042768b8e2e760fbd883de1a8a7e2cb80375}}, or a little powerless to represent an entire complicated systems of different types, like Action Description Language (ADL) {{cite:72fee926fe942f81fcf9f6a77c1659c76b1b36a7}}. In the literature employing situation calculus as a formal method to express the semantics in PDDL {{cite:25144eb3f2fbfb63ea79f86d019d815b10edf0e7}} and ADL {{cite:367d98eaaab81ac86e7100e2a179722f446fa6cf}}, the authors have tried to introduce some typing mechanisms, which is only limited to add type element in syntax, and only applied to variables. Other significant terms, such as fluents and predicates, are still typeless. Moreover, in semantics and reasoning, typing mechanisms are hardly discussed in these papers, neither is type checking.", "Yilan Gu et al. {{cite:a35894f4cbfcf5d7efdf47bdcf3c7af2df22d807}} proposed a modified version of the situation calculus built using a two-variable fragment of the first-order logic extended with counting quantifiers. By introducing several additional groups of axiom to capture taxonomic reasoning and using similar regression operator in Raymond Reiter's work {{cite:69548e38f542da825d8e9e15ce4dccf48dc87d3f}}, the projection and executability problems are proved decidable although an initial knowledge base is incomplete and open. While their system concerns primarily on semantics of the new components proposed but rarely talks about typing on them, our well-typed version of situation calculus mentions typing mechanisms together with a modified situation calculus version in an all around way.", "There are also some attempts on modifying situation calculus only based on a lightweight version of the original one. Gerhard Lakemeyer et al. {{cite:6e870391f6065baf1568f0e5e65409491bcaf9e1}} proposed a new logic dialect of situation calculus with the situation terms suppressed, namely, {{figure:47d8c7ed-21fd-4096-970e-e3faf4fef4a6}}{{figure:e7417356-f471-4fb1-b4f1-9dba88c1dbe7}}{{figure:09eb6692-e417-4578-bef8-4d342ffec2ee}}" ]

[]

[ "Several papers addressed the privacy problems of smart metering in the recent past {{cite:1d2a22a99d517cbeb60600ef953d89b0d1251bbb}}, {{cite:a4bf67baa30a41e74e37481fb9fb5945f0db9df9}}, {{cite:b20d5f1746daa6efc0402f2b61d10d5f80bb1194}}, {{cite:e812b18616e942a1ce9fac5d2be40fc0370c65ed}}, {{cite:2a7ce089966f39b7da8d4fecf81e92d7851843be}}, {{cite:6b43678e29c2f8d29befe2fb76e60d4bec3cf4ca}}, {{cite:5a82a6643396171f290c7528445361dc313dee7e}}, {{cite:14c713ffad6eaa5862dad986f462f11475a080e8}}. However, only a few of them have proposed technical solutions to protect users' privacy.In {{cite:b20d5f1746daa6efc0402f2b61d10d5f80bb1194}}, {{cite:2a7ce089966f39b7da8d4fecf81e92d7851843be}}, the authors discuss the different security aspects of smart metering and the conflicting interests among stakeholders.The privacy of billing is considered in {{cite:5a82a6643396171f290c7528445361dc313dee7e}}, {{cite:a4bf67baa30a41e74e37481fb9fb5945f0db9df9}}. These techniques uses zero-knowledge proofs to ensure that the fee calculated by the user is correct without disclosing any consumption data.", "Seemingly, the privacy of monitoring the sum consumption of multiple users may be solved by simply anonymizing individual measurements like in {{cite:1d2a22a99d517cbeb60600ef953d89b0d1251bbb}} or using some mixnet. However, these “ad-hoc” techniques are dangerous and do not provide any real assurances of privacy. Several prominent examples in the history have shown that ad-hoc methods do not work {{cite:eda035d2e195d69f3a569ac62f5eaaea523e329c}}. Moreover,these techniques require an existing trusted third party who performs anonymization.The authors in {{cite:6b43678e29c2f8d29befe2fb76e60d4bec3cf4ca}} perturb the released aggregate with random noise and use a different model from ours to analyze the privacy of their scheme. However, they do not encrypt individual measurements which means that the added noise must be large enough to guarantee reasonable privacy. As individual noise shares sum up at the aggregation, the final noise makes the aggregate useless.In contrast to this, {{cite:14c713ffad6eaa5862dad986f462f11475a080e8}} uses homomorphic encryption to guarantee privacy for individual measurements. However, the aggregate is not perturbed which means that it is not differential private.", "The notion of differential privacy was first proposed in {{cite:7c6c0f7f1f7f0f67554f57573cc53a3fbf83a339}}.The main advantage of differential privacy over other privacy models is that it does not specify the prior knowledge of the adversary and provides rigorous privacy guarantee if each users' data is statistically independent {{cite:112cf52b7737179f505cee4ea88765ac186d314a}}.Initial works on differential privacy focused on the problem how a trusted curator (aggregator), whocollects all data from users, can differential privately release statistics. By contrast,our scheme ensures differential privacy even if the curator is untrusted.Although {{cite:219b7db98e2712283ecd6885b15c598ef1d7db04}} describes protocols for generating shares of random noise which is secure against malicious participants, it requires communication between users and it uses expensive secret sharing techniques resulting in high overhead in case of large number of users.Similarly, traditional Secure Multiparty Computation (SMC)techniques {{cite:b00558241780903433a362ba6606fd3fb3c4efa6}} {{cite:afe1e3f69f310911b28d37cd33a9b59a0345e655}} also require interactions between users. All these solutions are impractical for resource constrained smart meters where all the computation is done by the aggregator and users are not supposed to communicate with each other.", "Two closely related works to ours are {{cite:943e4e9013569d87de0b5d345e2370c21a2e28d8}} and {{cite:af8b12b696f19f9cf6d0e090e12f135b5d9d1255}}.In {{cite:943e4e9013569d87de0b5d345e2370c21a2e28d8}}, the authors propose a scheme to differential privately aggregate sums over multiple slots when the aggregator is untrusted. However, they use the threshold Paillier cryptosystem {{cite:71229f321114e85adc5d7e2fd19731268f45f36f}} for homomorphic encryption which is much more expensive compared to {{cite:d6a4ae0c371e36f5de4b09fb76fdaefe237b2289}} that we use.They also use different noise distribution technique which requires several rounds of message exchanges between the users and the aggregator.By contrast, our solution is much more efficient and simple: it requires only a single message exchange if there are no node failures, otherwise, we only need one extra round. In addition, our solution does not rely on expensive public key cryptography during aggregation.", "A recent paper {{cite:af8b12b696f19f9cf6d0e090e12f135b5d9d1255}} proposes another technique to privately aggregate time series data. This work differs from ours as follows:(1) they use a Diffie-Hellman-based encryption scheme, whereas our construction is based on a more efficient construction that only use modularadditions. This approach is better adapted to resource constrained devices like smart meters.(2) Although {{cite:af8b12b696f19f9cf6d0e090e12f135b5d9d1255}} does not require the establishment (and storage) of pairwise keys between nodes as opposed to our approach,it is unclear how {{cite:af8b12b696f19f9cf6d0e090e12f135b5d9d1255}} can be extended to tolerate node and communication failures. By contrast, our scheme is more robust, as the encryption key of non-responding nodes is known to other nodes in the network that can help to recover the aggregate.(3) Finally, {{cite:af8b12b696f19f9cf6d0e090e12f135b5d9d1255}} uses a different noise generation method from ours, but this technique only satisfies the relaxed {{formula:f99b2deb-7878-4853-9f72-9d635560ac8b}} -differential privacy definition. Indeed, in their scheme, each node adds noise probabilistically which means that none of the nodes add noise with some positive probability {{formula:516edfb9-ec16-4749-9d68-bda2da822e11}} . Although {{formula:a9036b7c-e637-40fe-a79d-76a3458264c4}} can be arbitrarily small, this also decreases the utility. By contrast, in our scheme, {{formula:6dfd1be1-60d3-43bb-bbbe-15b29ac7babb}} while ensuring nearly optimal utility." ]

[]

[ "Several path algorithms have been devised for special cases of the general regularization problem (REF ). For example, the homotopy method {{cite:ffa6bd5e4a94e6eddcb4dfaa3ceb53484316d8fb}} and the least angle regression (LARS) procedure {{cite:9b524a17eaaf8684513186dcebe78a703eee4022}} handle lasso penalized least squares problem. The solution path generated is piecewise linear and illustrates the tradeoffs between goodness of fit and sparsity. {{cite:446276cf028306681d2349b1554363a5857a0f46}} give sufficient conditions for a solution path to be piecewise linear and expand its applications to a wider range of loss and penalty functions. Recently {{cite:8da241c9f0ca6a124f6cc5dd4db5c96b6e02989d}} devise a dual path algorithm for generalized {{formula:49e5b87d-5ab9-424a-8faa-7c5af8003669}} penalized least squares problems, which is problem (REF ) with {{formula:f7770a53-4fab-41ab-ad93-8c5b4b842200}} quadratic but without the second inequality regularization term. {{cite:d53cd1544489f711d6ca05067becf59966932ad6}} consider (REF ) in full generality for quadratic {{formula:b3595a00-6e88-40f0-953e-73ce6f11242e}} . All these work concerns regularized linear regression for which the solution path is piecewise linear. Several attempts have been made to path following for regularized GLMs for which the solution path is no longer piecewise linear. {{cite:3c3e3a6b071aba48c6746fade66be3c433d35954}} propose a predictor-corrector approach to approximate the lasso path for GLMs. {{cite:33aebade294ce0531e5f89effb4a93e3d3aa728e}} presents an ordinary differential equation-based path algorithm which delivers the exact path for lasso penalized GLMs. {{cite:0e746d82576953bba98b51b710187322ff9410ab}} derives an approximate path algorithm for any convex loss regularized by a separable, but not necessarily convex penalty. Here a penalty function is called separable if its Hessian matrix is diagonal. The separability restriction on the penalty term excludes many important problems encountered in real applications.", "Our proposed approach generalizes previous work in several aspects. First, it works for any convex loss (or criterion) function. Second, it allows for any type of regularization in terms of linear functions of parameters, equality or inequality. Equality constrained regularizations include lasso, fused-lasso and generalized {{formula:7ab2daf5-18ab-4493-8c5e-3b8e5c753f64}} penalty for example. Inequality constrained regularizations are required in shape-restricted regression and nonparametric log-concave density estimation. Last but not least, it is an exact path algorithm." ]

[]

[ "For personalized PageRanks approximation, in addition to the workof {{cite:8c7cc8997c6279958020d36854728e9346aaad9e}}, {{cite:e0ab2ecd2960b883edda2b30ec52b8dc89148cf4}}, {{cite:d0c0cc60ca43fea6592fafbe358daf67d4147a08}}, {{cite:9af81fc51fa5d26574591abb95dd5bb92d84ad66}}, {{cite:42435806fd473fa2616cec69d4852dcd4b4f3c94}},Andersen et al {{cite:8c2c1f3156f65187c5d784e5402846c0de5efbec}}developed a 'backward' version of thelocal algorithm of {{cite:d0c0cc60ca43fea6592fafbe358daf67d4147a08}}.Their algorithm finds all nodes thatcontribute at least some fixed fraction {{formula:8bec2d1b-6f8c-4a3c-b2b9-03c35e4a799d}} to apage's PageRank in time {{formula:ecd65e1e-a555-43e6-bb4f-52cf2c80944a}} where {{formula:c8758b27-c1bf-4e45-8852-1f5f82ed0b27}} is the maximum out-degree in the network.This algorithm can be used to provide some reliable estimateto a node's PageRank.For example, for a given {{formula:da19b75b-f554-4be2-a030-a9fe9223b401}} , in time {{formula:5983f424-b2d7-4b4f-93e7-ac099faec86e}} it can bound the total contribution from the {{formula:d1da96d2-175f-452e-aa32-37391f4ad5a4}} highest contributorsto the node's PageRank.However, for networks with large {{formula:c04f1823-549f-493d-987b-2bcfba052169}} , its complexity may not besublinear.", "As suggested in {{cite:8c2c1f3156f65187c5d784e5402846c0de5efbec}}, one can view the entire set ofpersonalized PageRanks (defined by all vertices in a network) as an{{formula:b5cb58c2-ea73-4e1a-a9d7-d38253d11c55}} matrix,which is referred to as the PageRank matrix of the graph.In the PageRank matrix, each row represents the personalizedPageRanks from a particular vertex, and each column represents thecontributions to its PageRank from all vertices in the network.Note that the sum of each row is 1 and the sum of the{{formula:bb806b6f-0fdd-4bc0-84d6-4a490bd47425}} column is the PageRank of {{formula:efe4d6bd-2411-43c1-af61-430523131fb9}} .", "In light of this, the problem of Significant PageRanks can be viewed as a matrix sparsification ormatrix approximation problem. There has been a large body of work of finding a lowcomplexity approximation to a matrixthat preserves some of its properties.Perhaps the most relevant one to our goal is a low rank matrixapproximation under the {{formula:1af19390-b639-4c18-a823-6bbf21c0fbfb}} matrix norm.", "All current methods for finding such low rank approximations runs intime at least linear in the size of the input matrix.See {{cite:a24096c6e0b452687a8a71b46c05d9ed1ca030b7}} for a survey of recent results.", "Next, a linear time Monte Carlo based method to estimate PageRank of all nodes is devised in {{cite:fc8c5c8bae1b7f9a3cc83e0e21882e617432a47f}}.The method is based on running constant number of random walks from each of the nodes in the network.", "Last, in the context of sublinear time graph algorithms, our research isrelated to the work of {{cite:f1f27d9bc7129d59412dcce5de2557bdb7c6dfc0}},in which sublinear time algorithms are presented forestimating several quantities.Our implementation of the Jump and Crawl querymodel can be viewed as a stringent type of the adjacency-list graphmodel used in {{cite:f1f27d9bc7129d59412dcce5de2557bdb7c6dfc0}}." ]

[]

[ "The use of WS-BPEL has been extensively studied by using different types of formalismsuch as Petri nets, Finite State Machines and process algebras. Regardingthe use of WS-BPEL together with WS-RF there are few works, and they only show a description of this union,without a formalization of the model.In {{cite:bae01f40b0dec67a5444f30c98895e694c108182}} Slomiski uses BPEL4WS in Grid environments and discusses thebenefits and challenges of extensibility in the particular case of OGSI workflowscombined with WSRF-based Grids. Other two works centered around Grid environments are{{cite:c32619d89a83a9eb01c0ff1e3e6c6fbaf42a3939}} and {{cite:35ca36e12e7405b5410a85258480d9d5e4715879}}. The first justifiesthe use of BPEL extensibility to allow the combination of different GRIDs, whereasEzenwoye et al. {{cite:35ca36e12e7405b5410a85258480d9d5e4715879}} share their experience on BPEL to integrate,create and manage WS-Resources that implement the factory/instance pattern.", "On the Petri nets hand, Ouyang et al. {{cite:04f0632a960054ec66c79e47b7a20d611286b7df}} define the necessary elements for translating BPEL processes into Petri nets. Thus, theycover all the important aspects in the standard such as exception handling, dead path elimination and so on. The model they consider differs from ours in that we formalize the wholesystem as a composition of orchestrators with resources associated, whereas they describe the system as a general scope with nested sub-scopes leaving aside the possibility of administering resources. Furthermore, we have also formalized the event handling and notification mechanisms.Another extensive semantics for BPEL 2.0is presented in {{cite:96ecf2c6e09bcdb742d012771bf9ec51bdb1e81e}} by Dumas et al, which introduces two new interesting improvements. They define several patterns to simplify some huge nets and introduce the semantics for the WS-BPEL 2.0 new patterns. On the {{formula:df29c3dd-e38b-41d2-8538-22de3de84bff}} -calculus hand, Dragoni and Mazzara {{cite:bdb5ad02edee3ed90604d7dd71915de663020eb1}}propose a theoretical scheme focused on dependable composition for the WS-BPEL recoveryframework. In this approach, the recovery framework is simplifiedand analyzed via a conservative extension of {{formula:7567c255-f07b-41e1-a2df-67e3ea35511c}} -calculus. Theaim of this approach clearly differs from ours, but it helps us to havea bigger understanding of the WS-BPEL recovery framework. Other work focused on the BPELrecovery framework is {{cite:5879172d4af9586846e0454519e56f1c68b821c7}}. Although this is more interested in the compensation handler, they describe the corresponding rulesthat manage a web service composition. Our work is therefore quite complete as we define rules for nearly all possible activities. In addition, we also consider time constraints. Finally, we would like to highlight the works of Farahbod et al. {{cite:49de4bfa4a6abf575a055c95c9653b6778e52832}} and Busi et al. {{cite:abc9764b70da81557590dd5277d08cac9473e698}}. In the first one, the authors extract an abstract operational semantics for BPEL based on abstract state machines (ASM) defining the framework BPEL{{formula:03c933b2-7625-42d4-accb-23157b81a72b}} to manage the agents who perform the workflow activities. In this approach time constraints are considered, but they do not formalize the timed model. On the other hand, the goal of the latter one is fairly similar to ours. They also define a {{formula:eb589179-72b7-456d-bb0c-15b74eab42b1}} -calculus operational semantics for BPEL and describe a conformance notion. They present all the machinery to model web service compositions (choreographies and orchestrations). The main differences with our work are that we are more restrictive with respect to time constraints and we deal with distributed resources." ]

[]

[ "Signals help create well-structured discourse. Temporal signals canprovide context shifts and orderings {{cite:2f9c44f7c6598841f1b6b92c5a4cd1636b511a1c}}.These signal expressions therefore work as discourse segmentationmarkers {{cite:a62382afcde4f7380bcc2245a441cc236002dfa7}}. It has been shown that correctlyincluding such explicit markers make texts easier for human readers toprocess {{cite:71dbd352aae2428f7a232f7822225d3a9d533b85}}.", "Some prior work has approached linguistic characterisation of signals.Brée et al. bree1986temporal performed astudy of temporal conjunctions and prepositions and suggested rules fordiscriminating temporal from non-temporal uses of signal expressionsthat fall into these classes. However, this work is purely theoretical andnot a corpus-based study. Schlüter schluter2001temporal identifies signalexpressions used with the present perfect and compares their frequencyin British and US English. Vlach vlach1993temporalpresents a semantic framework that deals with duratives when used assignal modifiers (see Section REF ). Brée etal. bree1993towards later describe the ambiguity of ninetemporal prepositions in terms of their roles as temporal signals. Ourwork differs from the literature in that is it the first to be basedon gold standard annotations of temporal semantics and that itencompasses all temporal signal expressions, not just those of aparticular grammatical class.", "Intuitively, signal expressions contain temporal orderinginformation that human readers can access easily. Once temporalconjunctions are identified, existing semantic formalisms may beapplied to discourse semantics {{cite:b59fc9f426f873b185349ce64644fcd955433d35}}. It is howeverambiguous which temporal expression they attempt toconvey {{cite:382710dc67d6ad0f0e2ed8b026957df94aa50982}}. Our work quantifies this ambiguity fora subset of expressions.", "Previous work applying temporal signals has been related to thelabeling of temporal links {{cite:49396128fd98f118c0f6f56f21c3abb3766b47c9}} and questionanswering {{cite:d57d9fff084d076b52460aeb939b20e8af92bf77}}, {{cite:3320c205ec456b3dbf29cd32b93c6ce41b5877c0}}. Inparticular, Lapata and Lascarides lapata2006learningremove the temporal signal from sentences containing two temporallyconnected clauses and attempt to learn sentence-level temporalrelations using the orderings suggested by the removed signal astraining data. Directly applying signals to the temporal relationidentification task, Derczynski andGaizauskas derczynski2010using halved the error rate ofTLINK classification for TLINKs that have a signalby adding features describing signals. This raised classificationaccuracy from 62% to 82%." ]

[]

[ "When temporally ordering events, it is intuitively likely that signal information may be useful. The trend in previous automated TLINK classification work has not been to directly target signals as a primary source of ordering information, although other attributes of annotated TLINKs and EVENTs have been exploited as training features. For example, the best known automatic TimeML annotation tool (TARSQI {{cite:ad00d60a546b454a6a08d2861fc34695c374f653}}) performs no SIGNAL annotation. Lapata and Lascarides {{cite:9d579aeb4b229cfe5ed6659c855d5bcb50ff0bdc}} worked with signals, using a restricted reference list of signal tokens instead of drawing signal text from human-annotated data. This work was only on same-sentence temporal links. Their accuracy at temporal relation classification was 70.7%. Bethard and Martin {{cite:32de0c63e4c9c9b48e6d5dbec8a8952b21e2834d}} included some features that described signals, where the compl-word feature (the signal text) was the 8th strongest in their set of features for temporal relation classification. However, this work has a number of limitations. First, it only uses the signal word and a simple relation type suggestion as features. It is also restricted to verb-clause construction TLINKs. Finally, The classifier only has to choose from a set of three TLINK classes (before, overlap, after).", "Table REF shows results from replicating the previous experiment on event-event TLINKs. The baseline listed is the most-common-class in the training data. We achieved a similar score of 60.32% accuracy compared to 61.79% in the previous work. The differences may be attributed to the non-standard corpus that they use. The TLINK distribution over a merger of TimeBank v1.2 and the AQUAINT corpus differs from that listed in the paper." ]

[]

[ "Most previous studies of the NG model in semiotic/opinion dynamics have focused on populations of agents in which all pairwise interactions are allowed, i.e., the agents are placed on the vertices of a fully connected graph.In statistical mechanics, this topological structure is commonly referred to as “mean-field” topology {{cite:0e4e9a128b4afd8dffdd48750b21b0649fc449de}}, {{cite:0a804a4376a252f40994a112ba9dea7139150c44}}. Apart from mean-field case, the model has also been studiedon regular lattices {{cite:4b84479139a9ef80a56cf0d14203c7e6087dca91}}, {{cite:41d927b12578e1f007c8eb2832b74e095ede182f}}; small world networks {{cite:41d927b12578e1f007c8eb2832b74e095ede182f}}, {{cite:85cfb08b021beea33aa279236db3b80d5c7a0df6}}, {{cite:4a8ca4107edf6c19f1fab4b7b4f38744c1ae2c2e}}, {{cite:181201b7e1e4dd31c99af55702e7c9ac71392e2c}}; random geometric graphs {{cite:41d927b12578e1f007c8eb2832b74e095ede182f}}, {{cite:b8b109f3a82cdabd7f77f4b0f10a38c59781790b}}, {{cite:775f605de604d3e84eb3d99bc2bd6baceaf191e7}}; and static {{cite:b5e15a68e879eae5be2d828d66d68d1f4c13e57c}}, {{cite:7601d591da4eb3341d8da9a3c0c229113744011d}}, {{cite:f20532a99ba435f646299f2747b73ec8812592fc}}, dynamic {{cite:4e742e5ff7c5ccd0dd3879a9160b053b1665ef03}}, and empirical {{cite:48c2854cf864b69b7064fd6416561267c23e49ff}} complex networks.", "Lu et. al. {{cite:48c2854cf864b69b7064fd6416561267c23e49ff}} have studied the Naming Game dynamics on a high-school friendship network and have shown that the presence of community structures affect the behavior of the dynamics through the formation of long-living late-stage meta-stable clusters of opinions. Therefore,they propose injection of committed agents (agents who never change their opinion) into the population for fast agreement of the dynamics. Nardini et. al. {{cite:4e742e5ff7c5ccd0dd3879a9160b053b1665ef03}} studied the dynamics of NG on adaptive networks where the connections can be rewiredwith the evolution of the game.", "All these prior works have studied the NG dynamics on essentially static networks. Therefore, our study reported here is unique and different from the literature since we consider the evolution of the NG dynamics overtime-varying social structure." ]

[]

[ "Many different types of discrete abstractions have been studiedfor hybrid systems {{cite:2985235901e5b2065154f0d8918bca0873ee1a86}} includingpredicate abstraction {{cite:be12371404943a5d760544e8c70d85f42960a66a}} and abstractionsbased on invariants {{cite:55acf3c72da95b4ba6b84cde87c52f01ccad811a}}. The use ofcounter-example guided iterative abstraction-refinement has also beeninvestigated in the past (Cf. Alur etal. {{cite:7b4086e1803353b14626d43024e6903fdba0e344}} and Clarke etal. {{cite:45564a071eccd0c8e52372c0abca666e42d9edc3}}, for example). In thispaper, we consider continuous abstractions for continuous systemsspecified as ODEs, discrete systems and hybrid systems using a changeof bases transformation. As noted above, not all transformations canbe used for this purpose. Our abstractions for ODEs bear similaritiesto the notion of topological semi-conjugacy between flows of dynamicalsystems {{cite:db223e53c6dcde4de2245c3e6a063ce09fa905df}}.", "Previous work on invariant generation for hybrid system by the authorconstructs invariants by assuming a desired template form (ansatz)with unknown parameters and applying the “consecution” conditionssuch as strong consecution and constant scaleconsecution {{cite:d64d70a75dd69f8089929e728764c0a615684819}}.Matringe et al. present generalizations of these conditions usingmorphisms {{cite:21134ee1dd1a2d0d9420dc5e641ba2926928cdfa}}. Therein, they observethat strong and constant scale consecution conditions correspond to alinear abstraction of the original non-linear system of a restrictiveform. Specifically, the original system is abstracted by a system ofthe form {{formula:b88d20d7-54ed-4bfe-acd0-4d72433e2e27}} for strong consecution, and a system ofthe form {{formula:47534d80-9ae1-43f8-a112-12b2db15a750}} for constant-scaleconsecution. This paper builds upon this observation by Matringe etal. using fixed-point computation techniques to search for a generallinear abstraction that is related to the original system by a changeof basis transformation. Our work is also related to the technique ofdifferential invariants proposed by Platzer etal. {{cite:801097d2ba594ce384631fd05b5efe727f226651}}. At a high level Platzer etal. attempt to prove an invariant {{formula:0cd5821e-2af6-4c10-b233-0d55ba572863}} for a continuous system(often a subsystem of a larger hybrid system) using differentialinvariant rule wherein the state assertion {{formula:47fadaff-f6d8-4190-93fe-be74a46269ba}} isestablished. Likewise, to prove {{formula:004ff22f-fd04-495f-8604-2cba684999c7}} , it seeks to establish{{formula:c2a8b275-b8b1-4225-aca1-7a5d82016651}} . In this paper, we may view the same processthrough a CoB transformation {{formula:b94d5f0d-2f16-4151-b843-e06dfdd81059}} that allows us to writethe abstract dynamics as {{formula:57463254-81e5-4ce3-9ef2-69ac0c8ca4ac}} . Going further, we seekto compute {{formula:6e758a38-f262-4d26-9f64-80dda2872be9}} that maps the dynamics toan affine or a polynomial system. On the other hand, differentialinvariants allow us to reason about Boolean combinations of assertionsand embed into a rich dynamic-logic framework combining discrete andcontinuous actions on the state. The work here and its extension todifferential inequalities {{cite:b025fd00aa4d89ff2ca431c93f28c1bbd63c8c42}} canbe utilized in such a framework.", "Fixed point techniques for deriving invariants of differentialequations have been proposed by the author in previouspapers {{cite:6e55b9ade8cb459b880b915d8152e6190bcc2762}}, {{cite:7f1a6766f57c142aa5c5d5290cb27175f8909660}}These techniques have addressed the derivation of polyhedralinvariants for affinesystems {{cite:6e55b9ade8cb459b880b915d8152e6190bcc2762}} and algebraicinvariants for systems with polynomial right-handsides {{cite:7f1a6766f57c142aa5c5d5290cb27175f8909660}}. In this technique, weemploy the machinery of fixed-points. Our primary goal is not toderive invariants, per se, but to search for abstractions ofnon-linear systems into linear systems." ]

[]

[ "Currently, there are several developed and implemented network protocols in WSNs, however, most of which have been designed to send event-based data rather than answering to explicit queries from the sink. In this section we present three protocols: LEACH {{cite:026b6a7c730f2381568b86abe8e26183476a8a55}}, CTP {{cite:b3e2e47fa5d0c7230dfb0c54c27165a8ec1c167d}} and Direct Diffusion {{cite:1ef2c817b6e21ce0e7a6198c5a93d4688d889ba0}}, which have some similarities with the HBCP. Furthermore, it has been included some mechanisms that, although very different in nature, aim to achieve the same goal as the HRS and BLQE.", "LEACH is a low-energy, self-organizing, adaptive clustering routing protocol that uses randomization to distribute the energy consumption evenly among the sensors in the network. It belongs to the hierarchical family of network protocols, where the different sensors in the network are divided in clusters, and its main purpose is to increase the whole sensor network life-time through network traffic balancing. Sensors elect themselves as cluster-heads at a given time with a certain probability, and the other nodes select from the existing clusters which they want to belong to by choosing the cluster-heads that has the minimum delivery cost (based on the RSSI). Once the network is configured, a cluster-head creates a schedule for each node in the cluster, similarly to TDMA, so the non-cluster-heads can turn off the radio except when they have to transmit, which minimizes the energy dissipated per sensor. LEACH does not define how data has to be collected, however, in contrast to HBCP, nodes always have a routeto reach the sink. In addition, one of the drawbacks of LEACH is that the network cannot have more than two hops, therefore, it is not suitable for large multihop networks.", "The Collection Tree Protocol (CTP) is a collection protocol based on events designed to improve the data collection reliability while sending a reduced number of control packets (beacons). It has been implemented to be hardware and MAC protocol independent, and the protocol evaluation in different testbeds and platforms has shown a reliability results between 90-99.9%, while sending up to 73% fewer control packets than existing approaches {{cite:f6722ffea5fdc60338b428a84b7441565555652f}}. CTP is basically characterized by the way routes are computed, and how beaconing is adapted to do not overload the network. Routes are address-free and the nodes just know its next hop and the cost to reach the sink. The parent selection mechanism is done using the ETX (Expected Transmission) {{cite:7fa79fe60bcca09c6c450ffdeb157262d65e82c1}} value which is computed according to the route cost and the parent reliability. Firstly, the sink starts sending a beacon with an ETX equal to zero, the nodes that receive that beacon announce their cost to reach the sink sending anotherbeacon, and that isrepeated for the following levels on the network. In case that a node receives more than one packet it will always choose the one with the lowest ETX as it would be the best route to reach the sink. With this routing algorithm a node knows if there is a loop in the network just comparing the ETX values, because the parent ETX should always be lower than the one of its son. The routes are periodically transmitted, however depending on the network conditions, the routes maintenance frequency decreases, and only increases if it is needed (link failure). Differently from the HBCP, CTP is thought to collect data with high frequency. Moreover, as the data transmissions are based on events, there is no kind of coordination, so nodes should always be awake to be able to receive data to later forward it. Because of this, CTP is not energy-efficient for low periodicity data collections.", "Direct Diffusion is a data-centric routing protocol for on-demand collections, where the sink broadcast a query to the whole network which contains parameters related to the type of information it is looking for, such as name of objects, interval, geographical area, interval, etc. When nodes receive a query, they analyze it, and if their information matches with the sink interests they start sending data to the sink for a given period of time, which was fixed in the query. Despite the fact that HBCP and Direct Diffusion are on-demand protocols, their purposes are very different. Direct Diffusion is designed for event-based applications that require to receive data from a portion or the entire network for a given interval of time, while HBCP is designed to collect one reading from each sensor in the network. In addition, Direct Diffusion does not implement any mechanism to reduce the collisions due to hidden nodes, what makes it not suitable for collecting a large amount of data during a short period of time.", "There are some TDMA-like protocols (as the LMAC defined by Van Hoesel et. al) {{cite:be8d6ac77fcf28c8ad433a454972b223b944a93e}} in which each sensor node picks a random slot from the free slots (slots not used in a 2-hop distance) in order to avoid hidden terminal interference. In LMAC sensor nodes exchange information about the slots they see as free. By doing this, a sensor node can select a slot from the 2-hops away unused slots, i.e., a slot considered free by all of its neighbors. The slot selection starts by a message sent by the sink. After receiving it, sensor nodes select the slot to use for transmitting data. Data transmissions are preceded by control messages that serve to synchronize sensor nodes and to inform about the intended destination of the data packet, then non interested nodes can go to sleep. The LMAC concept of reusing slots is very close the HRS concept of allowing nodes in different sectors to transmit during a given period of time. Moreover, the reuse distance (slot/sector) used in the LMAC and the HRS isthe same. However, given that LMAC needs to synchronize every node in the network, the configuration process is more complex and requires more transmissions of control messages, and as a result would entail higher energy consumption.", "The link quality estimation has gained attention in the last years because of the impact that links have in the delivery rate and energy consumption metrics. The basic metrics used for estimating the quality of a link are three: RSSI, LQI and PRR. From these, the two most used are the RSSI and LQI because they have shown that for certain thresholds they are capable of providing 95% reliability {{cite:53dd00798886264d4c668b911d22b65229deefcc}}. However, differently from the RSSI, the LQI quality estimation is more accurate but needs at least 120 packets for a good estimation. Overall the different Link Quality Estimators (LQEs), the Fuzzy Link Quality Estimator (F-LQE){{cite:3418d3fc27b83590f909ff2fad2a0859b4e9f375}}, by Baccour et al., has shown to outperform most of the LQE -based estimators {{cite:7fa79fe60bcca09c6c450ffdeb157262d65e82c1}}{{cite:eec4da6b25f0c8572269395f711bf46062c02847}}{{cite:7daf40155a56fda2ea9980c12ded5c567d52157a}}{{cite:ada0108418d502773282539f7f09fa423813dead}}. F-LQE does not only take into account one link metric but four: the smoothed packet reception rate, the asymmetry level, the average signal to noise ratio and the linkstability. However, the penalty of obtaining such an accurate metric is time, which makes the F-LQE not suitable for volatile environments where a fast link estimation is required. In order to reduce the time required to estimate a link quality, Boano et. al. presented the Triangle metric {{cite:4f247d9d7ff3b35f875c217d79a822f076acc29b}}. This LQE metric exploits the correlation between the LQI, PRR and SNR metrics to obtain a link estimation in a fast way. Results show that a good link estimation can be achieved with just 10 packets, and that the accuracy of the link estimation increases with the number of packets sent. Nevertheless, besides reducing the number of packets to estimate the quality of a link, the total number of control packets grows exponentially with the number of nodes. With BLQE, as we have shown in Section , it is possible to obtain a good estimation of the quality of a link using only a reduced number of control packets (linearly dependent on the number of nodes). An approach that only uses theRSSI metric is the Kalman filter based link quality estimator (KLE) {{cite:640e09079dc9fd47d735cd163f03e1e87655cd66}}, which estimates the quality of a link with a single packet. With the RSSI of the received packet and the ground noise, the KLE does an approximation of the SNR and the PRR of the link. The main drawback of the KLE is that the function that maps the Received Signal Strength (RSS) with the SNR does not adapt to dynamic environments. BLQE also suffers in dynamic environments, however, the {{formula:8de14f80-59a7-4667-9038-0b779e8cda07}} parameter is easier to compute and it obtains the link quality in both directions." ]

[]

[ "Our work draws on the lower bounds fornon-adaptive sparse recovery, most directly {{cite:477225213a0d4096c3cae87a614831e09669d4a9}}.", "The main previous lower bound for adaptive sparse recovery gets {{formula:5861fb96-09fb-4665-a39c-038e5738938a}} for {{formula:02d05247-a5ff-4c40-8dae-6b70b37fc7fe}}  {{cite:7c4595c8170a0a2c7dc271d23d9e1a6313e783d4}}. They consider going down asimilar path to our {{formula:52a9a7f2-ba3d-4c0f-a577-bea0051bb3f7}} lower bound, but ultimatelyreject it as difficult to bound in the adaptive setting. Combiningtheir result with ours gives a {{formula:80f5271d-cc9f-4afe-a852-6091d97eaebc}} lower bound, compared with the {{formula:256539f8-606d-462f-8618-0143a2489cc4}} upper bound. The techniques in their paper do not imply any bounds forthe {{formula:79740143-8611-422d-8fb9-0b6fcd44a35d}} setting.", "For {{formula:f26f1a02-fa4e-4921-b40d-674974c8495e}} in the special case of adaptive Fourier measurements (wheremeasurement vectors are adaptively chosen from among {{formula:ec8afab2-fddf-4064-8d9a-1e3332e4eca3}} rows of theFourier matrix), {{cite:227c44e655498faca7a6fb3a593c8a8601c1524a}} shows {{formula:9a1f4cb0-1e6c-4875-bfcf-c7cf3cfa1ce7}} measurements are necessary. In this case the main difficulty withlower bounding adaptivity is avoided, because all measurement rows arechosen from a small set of vectors with bounded {{formula:228e1e14-d2d6-412c-80ed-b0e6d6edd319}} norm;however, some of the minor issues in using {{cite:477225213a0d4096c3cae87a614831e09669d4a9}} for an adaptivebound were dealt with there." ]

[]

[ "Tipping models first became popular by the works of {{cite:2b9c147baac117b549773972ce492102edc48dd3}} and {{cite:0d56e13d7fb46429e81dc9948012ee15eb41148d}} where it was presented primarily in a social context. Since then, several variants have been introduced in the literature including the non-deterministic version of {{cite:e2c04993a1b50815b032354600600f1c3282817c}} (described later in this section) and a generalized version of {{cite:37ae1ac699c1e82f5ea2b5cf0f5ad37aefe2f776}}. In this paper we focused on the deterministic version. In {{cite:fbea353b6b2f6aab3880b2e15a3702a35e82d05a}}, the authors look at deterministic tipping where each node is activated upon a percentage of neighbors being activated. Dryer and Roberts {{cite:e93a0910dd99e9944e694ce901508fe2c6179241}} introduce the MIN-SEED problem, study its complexity, and describe several of its properties w.r.t. certain special cases of graphs/networks. The hardness of approximation for this problem is described in {{cite:c0ad6d19f2a469b06ee057df7d4b82e07129a47c}}. The work of {{cite:bd352e1acd664b38f3f04eafaa78eeff32320db0}} presents an algorithm for target-set selection whose complexity is determined by the tree-width of the graph - though it provides no experiments or evidence that the algorithm can scale for large datasets. The recent work of {{cite:e9bdf844e197a20f39a79707ced1365dabda8d68}} prove a non-trivial upper bound on the smallest seed set.", "Our algorithm is based on the idea of shell-decomposition that currently is prevalent in physics literature. In this process, which was introduced in {{cite:0e162d30d8c218489775af3740eafd0bb11f7df1}}, vertices (and their adjacent edges) are iteratively pruned from the network until a network “core” is produced. In the most common case, for some value {{formula:7892a739-6d52-4f8f-bffd-60117bcc83d7}} , nodes whose degree is less than {{formula:10bf76da-46d4-4f67-b9c6-13c27fbe45c8}} are pruned (in order of degree) until no more nodes can be removed. This process was used to model the Internet in {{cite:0de93c5813acb9da0724bf9f73ffd30aab20a5ff}} and find key spreaders under the SIR epidemic model in {{cite:e27086ab5f97810be94b4e93d3ea98c31383d6b1}}. More recently, a “heterogeneous” version of decomposition was introduced in {{cite:9f6e9432e2418b55bdc40c7336d7d7780c0d6628}} - in which each node is pruned according to a certain parameter - and the process is studied in that work based on a probability distribution of nodes with certain values for this parameter." ]

[]

[ "The GPLVM is effective as a nonlinear latent variable model in a wide variety of applications  {{cite:177ceb2217be27e93fa825a81b45643520a4771b}}, {{cite:7529a5eddab8d848bf93a96fa3da97774e0a397d}}, {{cite:4f7506264ff3cf1c711631a50645ab4dd6ce84c5}}.The latent positions {{formula:45c991d1-f582-4971-b0be-c0ae54f1bd02}} in the GPLVM are typically obtained by maximum a posteriori estimation or variational Bayesian inference {{cite:0c8109eefef26191c72a54fda33f7d0a3b57880f}}, placing a single fixed spherical Gaussian prior on {{formula:2bef367b-c1b8-4c78-bec3-e3d1f937ca79}} .A prior which penalizes a high-dimensional latent space is introduced by  {{cite:91a65cbaaec618d156a732e3ad32ea1bc598e026}}, in which the latent variables and their intrinsic dimensionality are simultaneously optimized.The iWMM can also infer the intrinsic dimensionality of nonlinear manifolds:inferring the Gaussian covariance for each latent cluster allows the variance of irrelevant dimensions to become small.Because each latent cluster has a different set of parameters, the effective dimension of each cluster can vary, allowing manifolds of different dimension in the observed space. This ability is demonstrated in figure REF b.", "The iWMM can also be viewed as a generalization of the mixture of probabilistic principle component analyzers {{cite:27c22e101fcd6e4f8b10b011442e17feb052844a}}, or mixture of factor analyzers {{cite:d6ee4b66d9bb22f9ae19783fbb8b1cc4a05ba57b}}, where the linear mapping of the mixtures is generalized to a nonlinear mapping by Gaussian processes, and number of components is infinite.", "There exist non-probabilistic clustering methods which can find clusters with complex shapes, such as spectral clustering {{cite:9b150b2e5b2057a92dedc18892cba01d2b2decf5}} and nonlinear manifold clustering {{cite:e848769f2b64c21c5bda04ea139755eac121c40a}}, {{cite:e937b816e381bf30a208f9e27cb6f8a8740ee97d}}.Spectral clustering finds clusters by first forming a similarity graph, then finding a low-dimensional latent representation using the graph, and finally, clustering the latent coordinates via k-means.The performance of spectral clustering depends on parameters which are usually set manually, such as the number of clusters, the number of neighbors, and the variance parameter used for constructing the similarity graph.In contrast, the iWMM infers such parameters automatically. One of the main advantages of the iWMM over these methods is that there is no need to construct a similarity graph.", "The kernel Gaussian mixture model {{cite:811e1a8fffc480c2f60c47df7d9416714f535a2e}} can also find non-Gaussian shaped clusters.This model estimates a GMM in the implicit high-dimensional feature space defined by the kernel mapping of the observed space.However, the kernel GMM uses a fixed nonlinear mapping function, with no guarantee that the latent points will be well-modeled by a GMM.In contrast, the iWMM infers the mapping function such that the latent co-ordinates will be well-modeled by a mixture of Gaussians." ]

[]

[ "Learning-to-rank is an active topic in machine learning, although ranking and permutations have been studied widely in statistics. One of the earliest paper in machine learning is perhaps {{cite:494a3cb868486383c3cab291ab17bcc0d3484193}}. The seminal paper {{cite:bd2e117293bd6969855fcbab8349801156411413}} stimulates much subsequent research. Machine learning methods extended to ranking can be divided into:", "Pointwise approaches, that include methods such as ordinal regression {{cite:5b13f2f588dcefcdf23d6f5e20605e4da3e5f21f}}, {{cite:b92a562da4d6dfbb72cfac288dac97eef3784b24}}. Each query-document pair is assigned a ordinal label, e.g. from the set {{formula:76165f77-176c-4d3d-8bd2-308c281b0542}} . This simplifies the problem as we do not need to worry about the exponential number of permutations. The complexity is therfore linear in the number of query-document pairs. The drawback is that the ordering relation between documents is not explicitly modelled.", "Pairwise approaches, that span preference to binary classification {{cite:bd2e117293bd6969855fcbab8349801156411413}}, {{cite:1eb83512a4686f97f28023ea527467bd60575665}}, {{cite:1cc8a5f574a44c0cc1fe49824b60d32604c680ba}} methods, where the goal is to learn a classifier that can separate two documents (per query). This casts the ranking problem into a standard classification framework, wherein many algorithms are readily available.The complexity is quadratic in number of documents per query and linear in number of queries.", "Listwise approaches, modelling the distribution of permutations {{cite:6ff1a7e8b5fba98ba0e0d6a857fc2d58b4c79b7e}}. The ultimate goal is to model a full distribution of all permutations, and the prediction phase outputs the most probable permutation. In the statistics community, this problem has been long addressed {{cite:e3a11006589080edadeea38968eaa03a977f6106}}, from a different angle. The main difficulty is that the number of permutations is exponential and thus approximate inference is often used.", "However, in IR, often the evaluation criteria is different from those employed in learning. So there is a trend to optimize the (approximate or bound) IR metrics {{cite:05e68937887ffed6fcaa85213a382aca6d620e3c}}." ]

[]

[ "Numerous survey or overview papers on trust and reputation models havebeen published. Many works start with a basic classification and thenenumerate and describe existing models{{cite:94301dcb16bfc1a7d84294c42918dee660e10bc6}}, {{cite:401320908c1594b1688b4ec9285a01fefade387d}}, {{cite:8c38f4efa6a494588c7d10d99b6eaa17c1bc20f7}}, {{cite:574ea54d3c7c01257365d48020b8a9f69a056756}}, {{cite:beaba334775b09145bf2389795dd3b9ac9fccbbf}}. Opposed to that, we tried in our work to use a rigorous methodologyfor the organization of the state of the art: extract the corestructure of prominent trust and reputation models, in order to getclear and simple differentiation criteria for the models of theliterature.", "{{cite:a86599ff00f3dc96481cc832e177b4eac9b4a9fc}} proposed a generic trust model that integratesseveral existing trustmodels. They show how to map each of these models to their model. However,they focus on the fusion of evaluations, whereas we examined theoverall structure of trust and reputation modeling.", "{{cite:c327d5a44a929c4ba96608134a56414539b79832}} define a functional ontology ofreputation, that is used in differentsystems {{cite:cf421c4d33c56cd919e30d67b360be5a64513b58}}, {{cite:be06bd3eb526282076bfa4ee4f2051fb28f2c49e}} in order to help agentsusing different trust and reputation models inter-operate. Theseapproaches try to cope with a situation where different models exist,whereas we try to propose a unified meta-model of trust andreputation." ]

[]

[ "There have been several recent works on CS with quantization andsaturation. {{cite:40b88b962256b2bf03702b623db98a0499e79387}} propose the formulation(). {{cite:d211cf47ce403e8351a12b54849a7c6503d927d5}} replace the {{formula:710793fb-2a9c-4949-acc7-4f00b91630f8}} constraint (REF ) by an {{formula:f2efbdf6-c4a2-49e2-b8a5-61a31649ed19}} constraint ({{formula:b3c7934b-f891-4991-bfe4-fdc725cd57ef}} ) to handle the oversampling case, and show that values {{formula:e4ca3190-0457-4c6b-a855-505443e762a1}} greater than 2 lead to an improvement of factor {{formula:5148af3b-361d-4947-aecd-b2db06486367}} onthe bound of error in the recovered signal. The model of{{cite:fa9735f6244d525cac8b1d75f241abf8d5df8698}} allows Gaussian noise in the measurements beforequantization, and solves the resulting formulation with an{{formula:ac07d02e-3191-44c4-bb29-15a787f4776e}} -regularized maximum likelihood formulation. The average distortionintroduced by scalar, vector, and entropy coded quantization of CS isstudied by {{cite:8efba6d3e1acecc4b8d2aa725b44698ff26f9519}}.", "The extreme case of 1-bit CS (in which only the sign of theobservation is recorded) has been studied by {{cite:84285c281436aa15385d5e6301c2012aaaee7f53}} and{{cite:02c22d38721c3ba9ab7fe20aaafb8114ce56efe6}}. In the latter paper, the {{formula:8f82cd05-b029-4b12-9388-81397875becd}} norm objectiveis minimized on the unit ball, with a sign consistency constraint. Theformer paper proposes two algorithms that require at most {{formula:b7d1bc33-1e5c-4da5-8797-d2d16ab65f06}} measurements to recover the unknown support of the true signal (thoughthey cannot recover the magnitudes of the nonzeros reliably)." ]

[]

[ "After the seminal work by Kempe, et al. {{cite:555c3dcbc0df369a15317ceadf772323af58ab30}},influence maximization in social networks has recently attractedmuch attention in data mining and social network analysis researchcommunities{{cite:d0ed55115307f67408fcd2221b1777e096170c5c}}, {{cite:b91368109277eeff90ae8febc15ef795f9660db9}}, {{cite:4c4b84df55cf722469d708b7e522358925f8d7a9}}, {{cite:37cd244ace30a6ac1d26f8fd1e0c79700b79f5eb}}, {{cite:9adfa90ae37b028a6c08949ba07e66d196da08c8}}, {{cite:b2f874126919996488fbcf7efd0478550df30882}}, {{cite:1d34082d9a1425b8a26493f7e7f2852a2e348d5f}}.A crucial subroutine in influence maximization is the influencefunction evaluation to which the influenceability estimation problempresented in this paper is closely related.In the following, we first review some notable work on influencemaximization problem, and then discuss the existing work oninfluence function evaluation. In {{cite:b91368109277eeff90ae8febc15ef795f9660db9}}, the authorsstudy the influence maximization problem under the context of waterdistribution and blogosphere monitoring. They propose a so-calledCELF framework for optimizing the influence maximization algorithms.To further accelerate the influence maximization algorithms, Chen,et al. in {{cite:4c4b84df55cf722469d708b7e522358925f8d7a9}} propose a scalable algorithm bysampling {{formula:cdb88575-cbaa-4286-82e4-c3cb1f76dcd6}} possible graphs and estimating the influence spread ofall vertices on each possible graph at one time. Subsequently, thesame authors propose a series of scalable algorithms in{{cite:37cd244ace30a6ac1d26f8fd1e0c79700b79f5eb}} and {{cite:b2f874126919996488fbcf7efd0478550df30882}} for influencemaximization by developing the heuristic vertices-selectionstrategies on unsigned and signed networks, respectively. Recently,Goyal, et al. in{{cite:9adfa90ae37b028a6c08949ba07e66d196da08c8}}, {{cite:1d34082d9a1425b8a26493f7e7f2852a2e348d5f}} considerthe problem of learning the influence probabilities, and study theinfluence maximization from a data-driven perspective. Note that allthe mentioned methods focus on the influence maximization problem.For the influence function evaluation problem, Kempe, et al. firstlypose it as an open problem in {{cite:d0ed55115307f67408fcd2221b1777e096170c5c}}. Then, Chen,et al. in {{cite:37cd244ace30a6ac1d26f8fd1e0c79700b79f5eb}} show that the influence functionevaluation problem is #P complete. Given the hardness of theproblem, most of the existing work for this problem, such as{{cite:555c3dcbc0df369a15317ceadf772323af58ab30}}, {{cite:b91368109277eeff90ae8febc15ef795f9660db9}}, {{cite:4c4b84df55cf722469d708b7e522358925f8d7a9}}, {{cite:37cd244ace30a6ac1d26f8fd1e0c79700b79f5eb}},are based on the Naive Monte-Carlo (NMC) sampling. In this paper,we study the influenceability evaluation problem and develop moreaccurate RSS estimators for estimating the influenceability, andour algorithms can also be used for influence function evaluation.", "Our work is also related to the uncertain graph mining. Recently,uncertain graphs mining have been attracted increased interestbecause of the increasing applications in biological database{{cite:bb52ddc39f82a2895c0fe555015ebcc2c3404b9e}}, network routing{{cite:eed16c6b9dfc5aa10008037e6e7aaf1fb9d5ef84}}, and influence networks{{cite:1d34082d9a1425b8a26493f7e7f2852a2e348d5f}}. There are a large body of workshave been proposed in the literature. Notable work includes findingthe reliable subgraph in a large uncertain graph{{cite:9a1f51191bc178a83fd3f016f5d96b22c23b6c9b}}, {{cite:2ea14f28ccd56bae892962c72455a948ced4f5fc}}, frequent subgraph miningin uncertain graph database{{cite:c98ad123221bffe34246c05d4b72300d1990c0db}}, {{cite:8a50efaad2dc6460639588548fd5d17d7934b2e0}}, subgraph search inlarge uncertain graph {{cite:70fc7568d2e4b483bf8b897555fc887c04849acf}}, K-nearestneighbor search in uncertain graph {{cite:a3e959f3d1b1d7a3ca3fe3fa3f5f2446fd43deaa}}, anddistance constraint reachability computation in uncertain graph{{cite:2840a9d24edc66aa5ea68361c08b0fee8520bbb4}}. In general, all the mentioneduncertain graph mining problems are shown to be #P-complete, andthereby finding the exact solution is intractable in large uncertaingraphs. Consequently, most existing work, such as{{cite:a3e959f3d1b1d7a3ca3fe3fa3f5f2446fd43deaa}} and {{cite:2ea14f28ccd56bae892962c72455a948ced4f5fc}}, are basedon NMC sampling. Basically, the NMC sampling based methods lead toa large variance, thus reduce the performance of the algorithms.Recently, Jin, et al. in {{cite:2840a9d24edc66aa5ea68361c08b0fee8520bbb4}} propose arecursive stratified sampling method for distance-constraintreachability computation on uncertain graph, although they do notclaim their method is a stratified sampling. It is important to notethat their method is a very special case of our RSS-I algorithm. Intheir method, they select only one edge for stratification at atime, and then recursively perform this procedure. Unlike theiralgorithm, first, we develop a generalized algorithm (RSS-I) thatselects {{formula:0764b57e-47b3-4e94-9b2f-946e40fcb955}} edges for stratification. Second, unlike theirreachability problem, here we study the influenceability evaluationproblem using the RSS-I sampling. Moreover, in our work, we alsodevelop another RSS estimator, i.e. RSS-II estimator. Note that allof our RSS estimators can also be applied into thedistance-constraint reachability computation problem.", "In addition, our work is related to the network reliabilityestimation problem, where a network is modeled as an uncertain graphand the goal is to estimate some reliability metrics of the network{{cite:43d0e4510d6e2dc76bce89c85e9502a0d8702a9c}}, {{cite:0d9df2d35dbbc3d82cf70ce9db676df5b8bb805d}}. There are many work onthis topic in the last five decades. Surveys can be found in{{cite:f663a49e520f392d522544855282f747ae3e4cc5}}, {{cite:0d9df2d35dbbc3d82cf70ce9db676df5b8bb805d}}.", "Below, we review the Monte-Carlo algorithms for network reliabilityestimation. Kumamoto, et al. {{cite:f484c0c7089e35a577d43f6db159b2a250741a56}} propose anefficient Monte-Carlo algorithm by exploiting the bound of thereliability metric. Fishman {{cite:e69702a93c6a041e142adeb1008a9aa0ad4ae32c}} proposes a moregeneralized Monte-Carlo algorithm based on such bound techniques forreliability estimation. Subsequently, Fishman {{cite:43d0e4510d6e2dc76bce89c85e9502a0d8702a9c}}compares four Monte-Carlo algorithms for network reliabilityestimation problem. Cancela, et al. in {{cite:f71f69c1d674ec23cd48473ab21e7b7e65b411ef}}propose a recursive variance-reduction algorithm for networkreliability estimation. Note that all the mentioned Monte-Carloalgorithms are tailored for the network reliability estimationproblem, and the reliability measure is typically a Boolean metricthus they cannot be used in our problem." ]

[]

[ "The uplink network MIMO model considered in this paper,where the joint processing takes place in the central server,is related to but is different from the multicell joint processing modelwith finite-capacity uni- or bi-direction backhaul links deployedbetween the base-stations. In this latter setting, the uplink multicellmodel becomes an interference channel with partial receivercooperation. Although a complete characterization of capacity for sucha model is still an open problem, information theoretical studies have beencarried out for the case of two-user Gaussian interference channelwith receiver cooperation {{cite:749c41c748f0d14e7c9a25a7dec147ac5d235fb2}}, {{cite:2608f7ebf5eef8b1313b7a966cb391c36ecf403c}}, {{cite:925f04cb99b1f7b45fdc8e1b7b0e8a1ad2aaf220}}, {{cite:c254012453502df6236ea57b52d26c2c5376e700}}, {{cite:04e35a25624ffe4e89dce8480907f331e58b43be}}.Likewise, the analysis of uplink cellular network under this setup hasbeen carried out in {{cite:02e3a3a16dd771abeabc34b5bcb4319769ee9295}}, {{cite:516ff4e0d1dc9c544867e8650b49e43687c73f6e}}, wherestrategies such asdecode-and-forward and compress-and-forward are proposed,and analytical results under the Wyner model are obtained.We mention briefly here that the downlink counterpart of this modelhas been studied in {{cite:6460226798525e10c80d75094ebe7882e154a083}}.", "For the uplink multicell joint processing model considered in thispaper, where the limited-capacity backhaul links are deployed betweenthe base-stations and a central processor asdepicted in Fig. REF , although a complete characterization of itsinformation theoretical capacity region is also an open problem,much work has been done on finding its achievable rateregion and the gap to the outer bound {{cite:3e07cd0b83c6a5c99636de1cbb0b36b7329af52f}}, {{cite:778e849f69bcc37debf44adb72ddd80a696e34d8}}, {{cite:84f66eb25834ff432208a97495d2499194c3648b}}, {{cite:4741a7184b62a754600e999fa273bd049bd63d76}}, {{cite:38529403d2e36955a582b552facc6d35ec4905bd}}, {{cite:232b847181b0fe0ed69ea53cb7bf39663bb23731}}.The early work by Sanderovich et al. {{cite:3e07cd0b83c6a5c99636de1cbb0b36b7329af52f}} considersa joint decoding scheme in which the base-stations quantize thereceived signals and forward them to the centralized processor, whichthen performs joint decoding of both the source messages and quantizedcodewords. Instead of joint decoding, {{cite:38529403d2e36955a582b552facc6d35ec4905bd}}, {{cite:4741a7184b62a754600e999fa273bd049bd63d76}} describeschemes which successively decode the quantization codewordsfirst, then the source messages. A comparison of successive decodingvs. joint decoding is contained in {{cite:232b847181b0fe0ed69ea53cb7bf39663bb23731}}. Alternatively,base-stations can also decode part of the messages of users intheir own cell, then forward the decoded data along with the remainingpart to the centralized processor{{cite:3e07cd0b83c6a5c99636de1cbb0b36b7329af52f}}, {{cite:778e849f69bcc37debf44adb72ddd80a696e34d8}}, {{cite:534783692c022c1225a48d466dd4973c69438dd1}}.The counterpart of these results for thedownlink with finite backhaul has been described in {{cite:fbb8fc5c9fb15da3a73d54c2fcf4fdf94e96793c}}.A review of some of these results is available in{{cite:b5f48ab390b0d85ba40879ded8d9b1097d66c910}}.", "The uplink multicell joint processing model is actually a special case ofa general multiple multicast relay network studied in{{cite:6bdef12a2b4c6c0672238a6764ddedcd4e111913}}, {{cite:e53330509a3a0b414d6ccd32560db707ec38b3a2}}, forwhich a generalization of quantize-and-forword can be shown to achieveto within a constant gap to the information theoretical capacity ofthe overall network. This is proved in {{cite:6bdef12a2b4c6c0672238a6764ddedcd4e111913}}using a quantize-map-and-forward technique, and in{{cite:e53330509a3a0b414d6ccd32560db707ec38b3a2}} using noisy network coding, both ofwhich require the joint decoding of the source messages and thequantized codewords.Thus in retrospect, the joint processing scheme of{{cite:3e07cd0b83c6a5c99636de1cbb0b36b7329af52f}} already achieves the approximate capacity ofthe uplink finite-backhaul multicell model, provided that thequantization noise level is appropriately chosen and joint decoding is doneat the central processor.", "It should be emphasized that joint decoding is challenging toimplement. Not only is the network-wide CSIrequired at the central processor,the computational complexity of joint decoding typically scalesexponentially with the total number of nodes in the network.Moreover, even a mere evaluation of the achievable rate can becomputationally prohibitive. The achievable rate region of thejoint decoding scheme as presented in {{cite:3e07cd0b83c6a5c99636de1cbb0b36b7329af52f}}involves {{formula:21406a80-e9be-4254-8b5f-5af4b984d7f9}} rate constraints, each requiring a minimization of{{formula:1ba986e6-1ed6-4c3e-8ff2-c91fbbe05c5d}} terms, where {{formula:fa6d6fd4-22b6-4f85-8766-ba4ef5ab4004}} is the number of users in the uplink multicellmodel. To obtain analytical results, {{cite:3e07cd0b83c6a5c99636de1cbb0b36b7329af52f}} andlikewise {{cite:516ff4e0d1dc9c544867e8650b49e43687c73f6e}}, {{cite:633dd9b03717b51714b0ddc8ee2c01e5a179b111}} assume certainsymmetry and obtain rate expressions under a Wyner cellular model.But the rate expression for an arbitrary network remains difficult toevaluate. On top of this, the achievable rate region needs to beoptimized over all choices of {{formula:dd4534fe-5fd8-4faa-8b2f-7421e1a663b4}} quantization noise levels, which isa difficult task, although recent progress has been made in{{cite:4741a7184b62a754600e999fa273bd049bd63d76}}, {{cite:232b847181b0fe0ed69ea53cb7bf39663bb23731}}, {{cite:836b31666b34d5e4776c7a48cc2180fcf2103eed}}." ]

[]

[ "Some of the earliest integer compression techniques are Golomb coding {{cite:2add8220389a5e81acecb5eaeca509d3189fe605}}, Rice coding {{cite:1eeb9242ef195c12488ca607c2e8c0369e489eca}}, as well as Elias gamma and delta coding {{cite:60694ab53433c23561ffa0dc02eb45081f1bffa9}}. In recent years, several faster techniques have been added such as the Simple family, binary packing, and patched coding.We briefly review them.", "Because we work with unsigned integers, we make use of two representations: binary and unary.In both systems numbers are represented using only two digits: 0 and 1.The binary notation is a standard positional base-2 system(e.g., {{formula:e2cdbb2a-1d5c-47c4-9033-d1dfa27c8cd6}} , {{formula:31d0ea98-690b-4132-b748-2efac9ff2917}} , {{formula:7a7c7443-42d9-4c84-9164-8158043dfefb}} ).Given a positive integer {{formula:7af62dfc-b2a7-46b6-93ba-50edcfef2032}} , the binary notation requires {{formula:3c903b42-ab7c-4cd7-8376-df9042713416}}  bits.Computers commonly storeunsigned integers in the binary notation using a fixed number of bits by adding leading zeros: e.g., 3 is written as 00000011 using 8 bits.In unary notation, we represent a number {{formula:d42af55c-92ff-4d3b-93da-030ca8eef0e0}} as a sequence of {{formula:928980f3-d7d9-4af9-85b6-4d522c888a1d}}  digits 0 followed by the digit 1 (e.g., {{formula:9d49a99a-942a-4d29-8fd0-2b9159c554d8}} , {{formula:80456d1c-54fb-415a-af0a-9bc3f217066d}} , {{formula:ee867141-a1e8-4fa1-9ce9-29e653fe2ba8}} ) {{cite:68785b1249014cb98a1cff74815c031ef80676f0}}.If the number {{formula:3a9b5459-f152-42c1-9c39-6097a963be82}} can be zero, we can store {{formula:8f10ab5f-9ec7-49bf-99a0-f4087141c6c0}} instead." ]

[]

[ "There are several morphological analysis methods {{cite:2a99a0e3dcb017d858f71a86744930deea5455e3}}, {{cite:d4e5c31b1339511da58e61c6e69c24d866d82675}}.Examples of these algorithms are the Comparison of Graphs {{cite:f86e8ada2ed87f4fd5b73344516ef5af75101690}}, the use of {{formula:d85d59a5-8978-4418-a7d9-7d447f5a24cf}} -grams {{cite:7e7e4d4bb7cab12b918fa2c21783616d568930a2}}, {{cite:90b838c5c4fb3d0e3c5af510e94261ab909dabe8}},the search for analogies {{cite:9f60ca169a10ba88152a2e3ac8ee29201c4aaa99}}, the surface models based on rules {{cite:1950930bd920ebfeeca419727d9afe026860bad5}}, {{cite:9f93a58172108c8f8596406474c2e212e15f7cf5}},the probabilistic models {{cite:11fe830cad72ee502656ffc6f8081d9cced29be9}}, the segmentation by optimization {{cite:02e38bd3f365183230b5e3d1d4df4c52b90213b5}}, {{cite:3eea9bded4db55d9a88d52567506fbc6953f5ae7}},the unsupervised learning of morphological families by ascending hierarchical classification {{cite:6f85138a1b5663028ff7f5d751040b6716ff24b8}},the lemmatization using Levenshtein distances {{cite:e0d6ed84460b59d30c28a804ab1e3ea750185418}} or identifying suffixes through entropy {{cite:cc7ffed4c122355e87f6905a023ff6bb840d21a2}}.These methods are distinguished by the type of results obtained, by the identification of lemmas, stems or suffixes.FlemmFlemm is available in web site: http://www.univ-nancy2.fr/pers/namer/Telecharger_Flemm.htm {{cite:9cf881ebf86b86ac22103891433c6394fbfaea3c}}is an analyzer for French which requires a text previously labeled by WinBrillWinBrill is available in web site: http://www.atilf.fr/scripts/mep.exe?HTML=mep_winbrill.txt;OUVRIR_MENU=1or by TreeTaggerTreetagger is available in web site: http://www.ims.uni-stuttgart.de/projekte/corplex/TreeTagger/DecisionTreeTagger.html.Flemm produces, among other results, the lemma of each word of the input text.Treetagger {{cite:256862f220b66025c41b3e72586bd3eda77aa4df}} is a multilingual tool that allows to annotate texts with information of Parts-Of-Speech (POS)The types of words are, for example, nouns, verbs, infinitives and particles. and with information of lemmatization.TreeTagger uses supervised machine learning and probabilistic methods {{cite:c01b682c594c18646caac70e37bf08f8da21f03b}}, {{cite:c6322f63d13955df47c90654d7fae8b032ccc9fc}}.It can be adapted to other languages as long as the lexical resources and manually labeled corpora are available.FreeLing is another example of a popular multilingual lemmatizerFreeLing is available in web site: http://www.lsi.upc.edu/~nlp/freeling/.", "Stemming transforms the variants of words into truncated forms.Two popular stemming algorithms are the Porter stemming algorithm {{cite:9f93a58172108c8f8596406474c2e212e15f7cf5}} and the Paice algorithm {{cite:4309cc0361a52d6bf1199883cc59b4888231652f}}.The methods of stemming and lemmatization can be applied when the terms are morphologically similar.Otherwise when the similarity is semantic, lexical search methods must be used.To reduce semantic variation, some systems use long dictionaries.Another systems use thesauri to associate words to entirely different morphological forms {{cite:9c26fa7285641c1466cb3fab599f8f143c357d28}}.Both methods are complementary since the stemming verifies similarities in the spelling level to infer lexical proximity,while the lexical algorithms use terminographic data with links to synonyms. {{cite:38be15dece580b7a57d8bd43f8594dd60ffa98c0}}.{{cite:b3d35f53ccd3481e48a79cc641ff8a9f5bd32076}} presents an unsupervised genetic algorithm for stemming inflectional languages.{{cite:c059d558b17182c5f0a7cf3915cece7bdf13a4a1}} proposes using morphological merged families into a single term to reduce the linguistic variety of Spanish indexed texts.", "Lexematization {{cite:99275711c3b4ff475ef8bcfcda614e9eea9eea92}} seeks morphological rearrangement of words belonging to the same family using automatic acquisition of morphological knowledgedirectly from the texts.Although the constitution on morphological families may be interesting in itself, its main interest lies in the benefits it produces for use as normalization mechanism(instead or in addition to stemming or lemmatization) in specific application domains.Probably the most common application domain is indexing terms in systems of Information Retrieval (IR).In recent years there have been numerous articles analyzing in different languages the efficiency of stemming/lemmatization in IR.In addition, significant progress has been made in IR in European languages other than English.In particular, {{cite:05fd63cdf1fad80ef82e11f8f9b09b08d006d228}} have evaluated corpora of CLEF evaluation campaignsCross-Language Evaluation Forum, http://www.clef-campaign.org/ (eight European languages).Their results show that morphological normalization techniques increase the efficiency of the IR systems and it can be used independently of the language.Reduction algorithms using syntactic and morphosyntactic variations have shown a significant reduction of storage costs and management by storing lexemes rather than terms {{cite:44baea827be10b51bb4f120612ae50a2d3eca0f8}}.{{cite:5a843bc66b6f3bb75ea34c7b67de3c41a41d498f}} works on the impacts of compound words and standardization in IR, finding no significant performance differences between stemming and lemmatization.", "However, the reality is that the linguistic resources necessary to establish morphological relationships without pre-defined rules are not available for all languages andall domains, without mention the constant creation of neologisms {{cite:7221c405161772bf0a6b950438a0395c6daa9b3f}}.The proposed solution for the specific task of automatic summarization is the Ultra-stemming of letters.", "Research in ATS was introduced by H.P. Luhn in 1958 {{cite:9d2b53538270086dfb722435851c8478bd148e3e}}.In the strategy proposed by Luhn, the sentences are scored for their component word values as determined by tf*idf-like weights.Scored sentences are then ranked and selected from the top until some summary length threshold is reached.Finally, the summary is generated by assembling the selected sentences in original source order.Although fairly simple, this extractive methodology is still used in current approaches.Later on, {{cite:551493da074bff458f7a04ec6abc5782fbf5768c}} extended this work by adding simple heuristic features of sentences such as their position in the text or some key phrases indicating the importance of the sentences.As the range of possible features for source characterization widened, choosing appropriate features, feature weights and feature combinations have became a central issue.A natural way to tackle this problem is to consider sentence extraction as a classification task.To this end, several machine learning approaches that uses document-summary pairs have been proposed {{cite:2fc3366f831ff515d19e2e97bb50d7c9f41b703d}}, {{cite:587148dbb79de352d99ca5af0f368e7eced0b270}}." ]

[]

[ "There are several accounts of predicative analysis of recursion in the (ICC) literature.We mention only those related works which are directly comparable to our work. See {{cite:7163d9e4758e3f9a2c44a187dba6c895b8c43fdb}} for anoverview on ICC.The mental predecessor of {{formula:0dac0536-51d0-4224-96a1-36ceaaefb8c6}} is the path order for {{formula:b2005e5e-80c2-4c71-a2bd-f974ff193284}}as put forward in {{cite:79ec79a2915a2c8319af9cad8cca633e70c5755d}}.Our main motivation lies in the observation that this order is directly only applicable to a handful of simple TRSs.This order only gains power if addition transformations are performed.But unfortunately powerful transformations are difficult to find automatically.", "Notable the clearest connection of our work is to Marion's light multiset path order (LMPO for short) {{cite:cf402b97a518a98eb09756b142d6409a16158410}}.This path order forms a strict extension of the hereproposed order {{formula:63345257-436e-43f4-97a7-e9a0abde9f5e}} . Similar to {{formula:626da178-1a44-4aa6-abaa-049afd008425}} it characterises {{formula:83e2d2c7-55f1-4b5c-992a-b15771491f16}} .As exemplified below however,compatible TRSs do not admit polynomially bounded runtime complexity in general.This renders {{formula:a6f35219-bff5-4498-a77e-67cd1e062bf0}} non-usable in our complexity analyser {{formula:71911f78-8d6e-464f-a8b1-c3a198cf039b}} .The definition of {{formula:7217ed93-72f8-4553-8bdd-1e50a59dc9fc}} has been calibrated with some effort to prevent such behaviour.", "Example The TRS {{formula:5fef2ccd-0231-404c-8461-fee45e75dd7c}} is given by the following rules:{{formula:6c7b6da3-b8de-41c3-b1b4-dbfeac861073}} ", "For a precedence {{formula:0f0e6580-6b3b-40fa-99d7-a2c5a91a1b1e}} that fulfils {{formula:99dedd5f-f9f2-4cbd-9d84-ef3345a3e4a7}} and {{formula:4d740537-5da1-43e5-8bd6-bc8cddb5c044}} we obtainthat {{formula:c53c7e17-dbc3-417c-aa81-df3da8290d76}} is compatible with {{formula:399ac467-3d5a-450a-9d01-521c1466def6}} .However it is straightforward to verify that the family of terms{{formula:26bf5350-490a-4153-a708-1633548c8c8a}} admits (innermost) derivations whose length grows exponentially in {{formula:f59e7fc1-9fcf-47b8-8323-ef91419c6ae0}} .Still the underlying function can be proven polynomial, essentially relying on memoisation techniques, c.f. {{cite:cf402b97a518a98eb09756b142d6409a16158410}}.", "The result of our main theorem can also be obtainedif one considers polynomial interpretations, where the interpretations ofconstructor symbols is restricted. Such restricted polynomial interpretationsare called additive in {{cite:df65e6296abd40a6aea8d6b41f1e44a1ea7279b7}}. Note that additive polynomial interpretationsalso characterise the functions computable in polytime, cf. {{cite:df65e6296abd40a6aea8d6b41f1e44a1ea7279b7}}.Although incomparable to our technique,unarguably such semantic techniques admit a better intensionality,but are difficult to implement efficiently in an automated setting.In our complexity tool , we see as a fruitful and fast extension that handlessystems in a fraction of a second.", "We also want to mention recent approaches for the automated analysis ofresource usage in programs.Notably, Hoffmann et al. {{cite:5afd7de89917f35b276eae5bfa340dfef06f2902}} provide an automatic multivariate amortisedcost analysis exploiting typing, which extends earlier results on amortisedcost analysis.To indicate the applicability of our method we have employed a straightforward (and complexity preserving)transformation of the RAML programs consideredin {{cite:5afd7de89917f35b276eae5bfa340dfef06f2902}} into TRSs. Equipped with our complexityanalyser can handle all examples from {{cite:5afd7de89917f35b276eae5bfa340dfef06f2902}}.Finally Albert et al. {{cite:7abdd4e44df20c35e2babb2cc266ad542d17be7a}} present an automated complexity toolfor Java{{formula:b74553a1-7981-4965-a9de-14c2ec58ec7d}} Bytecode programs andGulwani et al. {{cite:b93b6863c1e81e2db150de80f258833a2cc80ff8}} as well as Zuleger et al. {{cite:5194e777460d508221205a72ad63297d979c7615}}provide an automated complexity tool for C programs." ]

[]

[ "Since structure sharing {{cite:3d7cf15a3a825f9140d67132c0ee750d6dffa943}} was discarded and the Warren Abstract Machine (WAM) {{cite:52ec44f06f09b313c0f428462918b269cc4c19b8}} triumphed as the implementation model of Prolog, there has been little attention paid to exploiting data sharing in Prolog implementations.A lot of work has been done on indexing Prolog terms, but indexing is a different kind of sharing since it does not consider reuse of terms from different sources. In his Diploma thesis {{cite:f7826afa5b45f27fbbaf3cc3ed60bee93064a67a}}, Ulrich Neumerkel gave several example Prolog programs that would consume an-order-of-magnitude less space with data sharing than without sharing. He proposed applying hash-consing and DFA-minimization to sharing terms including cyclic ones. The proposed approach would incur considerable overhead if every compound term is hash-consed when created, and hence it is infeasible to incorporate the approach into the WAM. Following Appel and Goncalves's hash-consing garbage collector for SML/NJ {{cite:ebf2fd8ecae906fd6a6bfc9755b98a5a2ac3261e}}, Nguyen and Demoen recently built a similar garbage collector for hProlog {{cite:9cd815a93285571613aa491cc4a62107f7722bc7}}. The garbage collector hash-conses compound terms on the heap in one phase and performs absorption in another phase such that for the replications of a compound term only one copy is kept and all the others are garbage collected. Their experiment basically confirms the disappointing result reported in Appel and Goncalves's paper: the overhead outweighs the gain except for special programs.", "Hash-consing can be applied to the built-in predicate findall/3, as suggested by O'Keefe {{cite:0bc4d822cd522a3693aa8a36d95ef8f86066ca9b}}, to avoid repeatedly copying the same term in different answers. Currently, B-Prolog is the only Prolog system that supports hash-consing for findall/3. It employs a hash table for ground terms in the findall area. The algorithm and memory manager developed for the table area is reused for the findall area. With hash-consing, the system copies a ground term only once when copying answers from the findall area to the heap. Input sharing is exploited in the same way as for tabled subgoals. For a findall call, the compiler converts it into a call to a temporary predicate such that each argument of the generator occupies one slot in the stack frame. At runtime, the system first copies the arguments of the generator from the stack/heap to the findall area before the generator is executed. When an argument of the generator is found to be a ground compound term, its frame slot is set to reference the copy in the findall area. In this way, the argument and its subterms can be reused by the answers and the descendant calls. Nguyen and Demoen's implementation of input sharing for findall/3 {{cite:9cd815a93285571613aa491cc4a62107f7722bc7}} distinguishes between old terms that are created before the generator and new terms that are generated by the generator, and have answers share the old terms. Their scheme can exploit sharing of not only ground arguments but also ground terms in non-ground arguments. Their scheme may not be suited for tabled data since, unlike data in the findall area which live and die with the generator, tabled data are permanent. Also, their implementation does not exploit output sharing.", "A trie has been a popular data structure for organizing tabled subgoals and answers {{cite:9858d9492f309c79d6ee50e1723ef5c4406c6ac2}}. It is adopted by all the tabled Prolog systems except B-Prolog. As far as lists are concerned, a trie facilitates sharing of the prefixes while hash-consing allows for sharing of the suffixes. So for the two lists [1,2] and [1,2,3], the former shares the same path as the latter in the trie, but they are treated as separate lists when hash-consed; for the two lists [2,3] and [1,2,3], however, a trie allows for no sharing while hash-consing allows for complete sharing.", "Another advantage of tries is that they can be used to perform both variant testing and subsumption testing, and thus can be used in both variant-based and subsumption-based tabling systems. Hash-consing, on the other hand, can be used to perform equivalence testing only and thus cannot directly be used for subsumption-based tabling.", "Terms stored in a trie have a different representation from terms on the heap. For example, in the YAP system, tries are represented as trie instructions {{cite:92265da3c77d2f4c0a7125f7b7aba1a2d7572c7e}}. For this reason, when an answer is returned, it must be copied from its trie in the table area to the heap even if it is ground. In our system, structured ground terms in the table area have exactly the same representation as on the heap, so when they occur in an answer they do not need to be copied when the answer is returned.", "In the original implementation of XSB and YAP, one trie is used for all tabled subgoals, and for each subgoal one trie is used for the answer table. To enhance sharing, Raimundo and Rocha propose using a global trie for all tabled structured terms {{cite:c5936b2ef3cadf290b4a05ce4425ba5281ac0bd9}}. Due to the necessity of copying answers from the table area to the heap, the time complexity remains the same even when the space complexity drops.", "To some extent, the idea of representing sentences as position indexed facts {{cite:711596054636f6a31d9542c727731f8c50720491}}, {{cite:888d6ab4568ad4afcb975c239fb78b6d08c0a621}} is similar to hash-consing in the sense that a hash-consed term always is associated with a hash code. The translation from a program that deals with sequences represented as lists into one that uses position representation is not trivial. When difference lists are involved, the translation is even more complicated. The program obtained after translation may lose sharing opportunities. Therefore, hash-consing is a more practical solution to sharing than program transformation.", "As far as we know, our implementation is the first attempt to apply hash-consing to tabling. Our implementation enhances hash-consing with input sharing and hash code memoization to speed-up computation of hash codes. The extra cell used to store the hash code of a compound term is overhead if the term is never shared. Nevertheless, while the increase of space is always a constant factor, the gain in speed can be linear in the size of the data." ]

[]

[ "Friends' role in user interactions has been studied in sociology {{cite:4539e0e7034d03e5548fb1e00596fd4d7ed82477}}, but observing it on a wide scale has not been possible until online social networks attracted millions of users and provided researchers with social network data. For online social networks, Ellison et al. {{cite:4ef1572978bfa2329e3d246d1a54ad161e7f3491}} defined friends as social capital in terms of an individual's ability to stay connected with members of a previously inhabited community. Differing from this work, we study how friends can help users to interact with new people on social networks. Although these interactions can increase users' contributions to the network {{cite:ad302b643be34bc4452cbc932731cafbe9ba5cfc}} and help the social network evolve by creation of new friendships {{cite:43ef1d1e5264fc0f209c42c2cbfd162dc6a76159}}, they can also impact the privacy of users by disclosing profile data. Squicciarini et al. {{cite:965e0ac1afb2424b08efc032953e5988af0a4a97}} have addressed concerns of data disclosure by defining access rules that are tailored by 1) the users' privacy preferences, 2) sensitivity of profile data and 3) the objective risk of disclosing this data to other users. Similarly, Terzi et al. {{cite:9676f007fe356cde40a90ad6828dac77f640a3fc}} has considered the sensitivity of data to compute a privacy score for users. Although these works regulate profile data disclosure during user interactions, they do not study the role of friends who connect users on the social network graph and facilitate interactions. Indeed, research works (see {{cite:f715b98226572b74a42754143f76d08c9f873144}} for a review) have been limited to finding the best privacy settings by observing the interaction intensity of user-friend pairs {{cite:bda1e5f12e9b42e2f9374ddec4e91bf99ef5dcb3}} or by asking the user to choose privacy settings {{cite:db3bb6b231878bb69854b26859b8518c3569b774}}. Without explicit user involvement, Leskovec et al. {{cite:943ea7d96c742f7d433672966232fbc616d73953}} have shown that the attitude of a user toward another can be estimated from evidence provided by their relationships with other members of the social network. Similar works try to find friendship levels of two social network users (see {{cite:bd3877930d6a5b94f46d5a30bd0686818884af05}} for a survey). Although these work can explain relations between social network users, they cannot show how existence of mutual friends can change these relations.", "Privacy risks that are associated by friends' actions in information disclosure has been studied in {{cite:e63eb7defde44da8206637c8d28ae58f692461f4}}, but the authors work with direct actions (e.g., re-sharing user's photos) of friends, rather than their friendship patterns.Recent privacy research focused on creating global models of risk or privacy rather than finding the best privacy settings, so that ideal privacy settings can be mined automatically and presented to the user more easily. In {{cite:b8f1aad305b06bdd93ce4d40664331cc7b158e9e}}, Akcora et al. prepared a risk model for social network users in order to regulate personal data disclosure. Similarly, Terzi et al. {{cite:9676f007fe356cde40a90ad6828dac77f640a3fc}} has modeled privacy by considering how sensitive personal data is disclosed in interactions. Although users assign global privacy or risk scores to other social network users, friend roles in information disclosure are ignored in these work.", "An advantage of global models is that once they are learned, privacy settings can be transfered and applied to other users. In such a shared privacy work, Bonneau et al. {{cite:494506cd215e463098cbff4922da489d3a522ac2}} use suites of privacy settings which are specified by friends or trusted experts. However, the authors do not use a global risk/privacy model, and users should know which suites to use without knowing the risk of social network users surrounding him/her." ]

[]

[ "Research in Automatic Text Summarization was introduced by H.P. Luhn in 1958 {{cite:56d99b998514e757155533a0fda81f226facc4ce}}.In the strategy proposed by Luhn, the sentences are scored for their component word values as determined by tf*idf-like weights.Scored sentences are then ranked and selected from the top until some summary length threshold is reached.Finally, the summary is generated by assembling the selected sentences in the original source order.Although fairly simple, this extractive methodology is still used in current approaches.Later on, {{cite:0d7430750f2df458a23ec8a8d1bdd82c1ba5a72d}} extended this work by adding simple heuristic features such as the position of sentences in the text or some key phrases indicatethe importance of the sentences.As the range of possible features for source characterization widened, choosing appropriate features, feature weights and feature combinations have become a central issue.", "A natural way to tackle this problem is to consider sentence extraction as a classification task.To this end, several machine learning approaches that uses document-summary pairs have been proposed {{cite:c983d1af3c443af3beb738aad88ac25e4b0cb7de}}, {{cite:01e68c1be7abd011d63c60d191fce5b306826634}},An hybrid method mixing statistical and linguistics algorithms is presented in {{cite:162d3fedc15acda21b789f63398e7f6602c51762}}.{{cite:afad821f128585b71524236e1dfa930e4ba320ba}} and {{cite:bff63d1ba0abcf517a4200797a624e35b7c572bf}} propose a good state-of-art of Automatic Text Summarization tasks and algorithms." ]

[]

[ "Problem REF is an example of a subspace search problem. In {{formula:6b6c373e-1797-45ea-889b-2c619e5fb2a2}} , for {{formula:77944e97-cc3b-4e83-ac79-7eedeefe5a36}} and {{formula:a2ffc298-d340-43f8-acc1-78737eeba198}} , efficient algorithms with sublinear query complexity in {{formula:08a59014-48d6-4c6f-a926-a0c25a46e504}} exist for the approximate versions {{cite:23505f40f93d6e854d7f31fe3447248fc629b8e7}}, {{cite:77314cbb2152c8d6e6151c8726fdeadbcfcf67f9}}. For {{formula:4abba47b-d7ca-4777-83bd-f6873bf16f67}} , recent attempts {{cite:31a988e8708d06445e6feeb83c292a3607c372f1}}, {{cite:eb57ec3800b570fac0ac59668a3594de361c77ad}} offered promising numerical examples, but not sublinear complexity guarantees. Results in theoretical computer science suggest that these limitations may be intrinsic to the problem {{cite:3042dc702404ab17a819acf222ea065820986560}}.", "These attempts exploit special properties of the {{formula:dcd4a5c2-44f0-4114-8411-d18bdc0e831d}} version of Problem REF , and do not apply to its {{formula:38be039e-cdd8-4815-94d1-f197100bfedb}} variant. However, the {{formula:83cf7f08-a03a-4122-a69a-b64142c09ce2}} variant retains the aforementioned difficulties, suggesting that an algorithm for {{formula:e3088025-70bf-4649-9f6c-d9a2c1a7b133}} near subspace search with sublinear dependence on {{formula:95df15bc-9927-48a7-a83c-3f3bff6cd036}} is unlikely as well.Although it could be possible if we are willing to accept time and space complexity exponential in {{formula:8eeddc65-d5c3-467b-9396-c3f113588ffd}} or {{formula:3e3b970d-0dcd-426a-af80-53ce389dfb07}} , ala {{cite:b293a2c049eea0e82b9d209963010c6c8e1e9bf9}}. This motivates us to focus on ameliorating the dependence on {{formula:8c53ff7c-c67f-44b7-8a0c-b0ce64f6464d}} . Our approach is very simple and very natural: Cauchy projections are chosen because the Cauchy is the unique 1-stable distribution, a property which has been widely exploited in previous algorithmic work {{cite:23505f40f93d6e854d7f31fe3447248fc629b8e7}}, {{cite:f6f87b70bb80376dc4883bea8e1695e1e76ec7dd}}, {{cite:b8022e826d6081d6e5927812c126bc78c83daa51}}.", "However, on a technical level, it is not obvious that Cauchy embedding should succeed for this problem. The Cauchy is a heavy tailed distribution, and because of this it does not yield embeddings that very tightly preserve distances between points, as in the Johnson-Lindenstrauss lemma. In fact, for {{formula:8c79e312-4f99-4e89-8b3a-304e5d77d1cf}} , there exist lower bounds showing that certain point sets in {{formula:bba5cfd6-6e47-41ec-b13c-cdedb20d7285}} cannot be embedded in significantly lower-dimensional spaces without incurring non-negligble distortion {{cite:6bfc462a96bb07272ebe8116986e59f714d398f0}}. For a single subspace, embedding results exist – most notably due to Soehler and Woodruff {{cite:b8022e826d6081d6e5927812c126bc78c83daa51}}, but the distortion incurred is so large as to render them inapplicable to Problem REF ." ]

[]

[ "In our recent work {{cite:d302ccce8af1c914c467e2ced31a67f824ab9d6b}}, we showed that for a general composite detector with K hexagonal detector modules, the probability amplitude {{formula:19ca26d1-e4e5-4036-9196-16233d547a3b}} for a monoenergetic {{formula:14a26b6f-528b-4f5a-bbfa-efa8d2708ad9}} -ray, could be expressed (see equation 3.2 of ref. {{cite:d302ccce8af1c914c467e2ced31a67f824ab9d6b}}) as{{formula:ee87b4bd-e957-43ce-b36f-7c8e960a27c4}} ", "where the coefficient {{formula:1a681873-1e4b-43aa-b842-207b19455b0c}} is the ratio of covered lateral surface area to total lateral surface area of the composite detector, and {{formula:e7a6826f-47d6-4691-874b-25f5e7447047}} are probabilities describing the interaction of a monoenergetic {{formula:63e0cae3-2b21-4b9e-b144-4ac8855acc3f}} -ray with a composite detector. Here, the first and second terms correspond to double and triple hit events which contribute to FEP in addback mode. If we denote {{formula:ab37c332-d8a2-4a98-afe1-b4542916510e}} by {{formula:98d2a127-fd2b-4f9d-836b-84018591737a}} and {{formula:7c4d7fff-6cf8-4491-a0bb-2c4445d485ec}} by {{formula:643f10b3-b7ec-471f-b8fc-c692b748dd3f}} , then we have{{formula:fc8bcdb1-7ba3-466b-a1ba-242608588440}} ", "Note that the knowledge of {{formula:f8c56534-ee7d-4472-8d57-578dcbd109bf}} and {{formula:e1118b42-0cc7-4a35-bffd-d3dbf8f8812b}} are sufficient to know {{formula:da7095c4-47ce-42f6-997c-166a18273338}} since {{formula:36f9bc68-5ada-4871-98e4-b36f8395ee56}} . In order to find expressions for {{formula:c0a39a66-44f8-475b-8aa0-f91fcab7a2b3}} and {{formula:080e5327-9c12-44cf-afcb-7a8f56df6fea}} , we can consider the detailed calculations for the clover detector as shown in section 2.3 of ref. {{cite:f187938a95688da4de0a2c99bf70bac3e6a37c74}}. Similarly, for the present case, it could be shown that{{formula:6022a416-f430-454b-860d-27aeec1a38aa}} ", "For simplicity, we make two approximations: {{formula:04fbd4bc-9a29-4573-8856-21e029b36e7d}} and {{formula:a9c5dc08-9290-4e3e-83b5-8dc12699efae}} . Thus, from the present formalism, it is possible get an estimate of the peak-to-total ratio in single detector mode given by{{formula:76a973f2-3127-4870-9b2a-8ebe789cb154}} ", "An interesting application: From the above formulation, it is possible to know the value of {{formula:a77ef7ea-3c7c-426e-9728-c8f69ea0d952}} of a composite detector if we know the same for a different composite detector at a given energy. Let us consider the cluster detector and another composite detector, such that{{formula:361461d5-f917-4ef4-80d6-da96059dceef}} ", "and{{formula:1ac313bf-d2a5-4a3f-939b-748c5cc68871}} ", "If we consider the quantity{{formula:82f89d18-7391-43cc-a40d-9a6fcc443abc}} ", "it is observed that {{formula:d61b0d6a-e6e5-420d-bcdc-e10d36bf38a7}} . Thus, using {{formula:a7e536f1-bfd6-48e9-8434-704475c581ea}} , we can find the peak-to-total ratio of a general composite detector, if we know the value of one of them (cluster detector in the present case) at an energy." ]

[]

[ "Various aspects of the resiliency of terrorist networks have been previously explored in the literature. For instance, {{cite:bb7a0bf692e78fe7b350c18c1446c92b4b7584c9}} studies the ability such network to facilitate communication while maintaining secrecy while {{cite:f1bccf98af3c1ded15aa18a2a722dc9b20c6eb0e}} studies how such networks are resilient to cascades. However, to our knowledge, the network-wide degree centrality in such networks - and how to increase this property - has not been previously studied.", "There has been much work dealing with the removal of nodes from a network to maximize fragmentation {{cite:2a76eef4037e6fd29d2e68de70c1612ba161e58a}}, {{cite:1af21cd8b24c99f96d9819353438e210cca8f06a}}, {{cite:a7345f9b924906746d9f4a3f4e69080e86fb61cd}} where the nodes removed are mean to either increase fragmentation of the network or reduce the size of the largest connected component. While this work has many applications, it is important to note that there are special considerations of terrorist and insurgent networks that we must account for in a targeting strategy. For instance, if conducting a counter-intelligence operation while targeting, as in the case of {{cite:f66790b5c763c441f4e2ded861101785dc61563d}}, it may be desirable to preserve some amount of connectivity in the network. Additionally, fragmentation of a network may result in the splintering of an organization into smaller, but more radical and deadly organizations. This happens because in some cases, it may be desirable to keep certain terrorist or insurgent leaders in place to restrain certain, more radical elements of their organization. Such splinter was observed for the insurgent organization Jaysh al-Mahdi in Iraq {{cite:65ddc90426268854c9e7b93dead6b3927758a42e}}. Further, these techniques do not specifically address the issue of emerging leaders. Hence, if they were to be used for counter-terrorism or counter-insurgency, they would likely still benefit from a shaping operation to reduce organization's ability to regenerate leadership.", "There has been some previous work on identifying emerging leaders in terrorist networks. Although such an approach could be useful in identifying certain leaders, it does not account the organizations ability as a whole to regenerate leadership. In {{cite:de1253496fff158832ef8f0b64954067d5c363bf}}, the topic of cognitive demand is studied. The cognitive load of an individual deals with their ability to handle multiple demands on their time and work on complex tasks. Typically, this can be obtained by studying networks where the nodes may represent more than individual people - but tasks, events, and responsibilities. However, it may often be the case that this type of information is often limited or non-existent in many situations. Additionally, as discussed throughout this paper, the targeting of individual nodes may often not be possible for various reasons. Hence, our framework, that focuses on the network's ability to regenerate leadership as opposed to finding individual emerging leaders may be more useful as we can restrict the available nodes in our search using the “no strike list.” By removing these nodes from targeting consideration - but by still considering their structural role - our framework allows a security force to reduce the regenerative ability of a terror network by “working around” individuals that may not be targeted.", "In more recent work {{cite:a1de9fc44deaa484d022b1f3eff8019309687fc3}} looks at the problem of removing leadership nodes from a terrorist or criminal network in a manner that accounts for new links created in the aftermath of an operation. Additionally, {{cite:7f500b9c0de4f5e1632e179cd45a29665987bfd5}} look at identifying leaders in covert terrorist network who attempt to minimize their communication due to the clandestine nature of their operations. They do this by introducing a new centrality measure called “covertness centrality.” Both of these approaches are complementary to ours as they focus on the leadership of the terrorist or insurgent group - as this approach focuses on the networks ability to re-generate leadership. A more complete integration of this approach leadership targeting method such as these (i.e. using a network-wide version of covertness centrality) is an obvious direction for future work." ]

[]

[ "To our knowledge, this particular problem setting has not been directly addressed in the literature.Temporal data is most commonly treated using a model of sources which update continuously, or synchronously at an underlying temporal sampling rate.Pertinent formulations for our purposes include the infinite factorial hidden Markov model (infinite FHMM) of {{cite:ab670bf7f6234ff81c59ec63d68b16d97471ab28}},or the probability hypothesis density filter (PHD filter) {{cite:862f34f07b66b655a989dc9431be704be1b7399d}},both of which infer an unknown number of independent Markov sources.FHMMs assume that the underlying sources are not intermittent during their lifetime,and also that they persist throughout the whole observation period.Pragmatically, intermittent emissions may be handled by incorporating silence states,though the duration of such states cannot take an arbitrary distribution.The PHD filter allows for stochastic missed detections but not for structured intermittency.", "Among techniques which do not assume a synchronous update,graph clustering approaches such as normalised cuts have similarities to our approach {{cite:89fb21bf112cc90b5a02f8dbf41f0121deebea8e}}.In particular, {{cite:c4a267a5c38fd198ed7d900769a8ce787644e0d1}} apply normalized cuts in order to cluster temporally-ordered data.However, the normalised cuts method is applied to undirected graphs,and {{cite:c4a267a5c38fd198ed7d900769a8ce787644e0d1}} use perceptually-motivated similarity criteria rather than directed Markov dependencies as considered herein.Further, the normalized cuts method does not include a representation of clutter noise,and so {{cite:c4a267a5c38fd198ed7d900769a8ce787644e0d1}} perform signal/noise cluster selection as a separate postprocessing step.In the present work we include an explicit noise model.", "Our problem setting also exhibits similarities with that of structure discovery in Bayesian networks {{cite:3fa4361a7dc3a4b8b41eea1bff213acdceb99bdb}}.However, in that context the dependency structure is inferred from correlations present in multiple observations from each vertex in the structure.In the present case we have only one observation per vertex,plus the partial ordering implied by temporality.", "In the followingwe develop a model in which an unknown number of point-process sources are assumed to be active as well as Poisson clutter,and describe how to perform a maximum likelihood inference which clusters the signal into individual identified tracks plus clutter noise.We then demonstrate the performance of the approach in synthetic experiments,and in an experiment analysing birdsong audio." ]

[]

[ "There is sparse literature on the design principles of social tagging tools {{cite:775912c42fddf7b4ac4d5c144b6a4928637a06cc}}, {{cite:44a0bd13e4066c4be1e4bc1ba5202582380df5de}}, {{cite:b0fa94bb1b543713224051912c3faab83d3dfde3}}. Some work has been done on building hierarchy structure from the social tagging to uncover the hidden child-parent semantics {{cite:1201888670064eb8fc32678ab19e1f68a843278e}}, {{cite:71ccc873b276e48d4364a1800ff264b0be72246d}}.Visualization, as a powerful type of social bookmarking tool, has been studied and utilized in several existing designs.Visualization of hyperlinks between Web pages was adopted to enhance adaptive navigation {{cite:808456208c0452e712e596ac672fbec2a901e9e8}}.Cluster Map, a social bookmark visualization tool, highlighted the relationships among users and bookmarks to identify tag and community structures {{cite:c30c6fe6f0fc319da6525011280f153d26896f50}}. Unlike ClusterMap, GalViz is designed to emphasize the semantic relationships between tags and resources to help users manage existing resources and discover new ones.Besides, the graphical interface was shown to be useful for distributed collaborations and interactions on social bookmarking {{cite:a25ee9eaef307b9ec74c67a35808f34c81be2511}}.Graphical visualization of concept networks was integrated into several Web applications as an innovative interactive user interface {{cite:f89ecdc894a5890ff3457493129a2181e1ea8aa4}}, {{cite:6b248cf248dbd9888d3d9dcf0905b73ce81b9e53}}. Most of such existing applications display context of items of the same type, but GalViz is able to provide the visualization of heterogenous networks among two different objects, tags and resources.", "GiveALink.org, as a research-oriented social tagging platform, broadly examines several aspects of social tagging to foster the construction and applications of socially driven semantic annotation networks. Previous research includes the design of effective similarity relationships {{cite:5c980e660e1aa7d8ee3c7503d8eb0ddd9f6618d3}}, social spam detection {{cite:eefba676565491ede5c24cd38630dafd161d8c33}}, and social tagging games as incentive for collecting high-quality annotations {{cite:03224f067131c77d543a2b3d9f20db2e2fba4924}}. Former work in GiveALink on exploratory navigation interfaces {{cite:788dbc0b184a26b9ec4c60b9a670a34be02fc89e}} and bookmark management {{cite:35c26dd55e3605365810b8c2c4c8b2113ac99c96}} have significant influence on the new design presented in the paper." ]

[]

[ "A lot of research works have been dedicated to the implementation of physical layersecurity by adopting artificial noise generation for cooperative jamming. These works can be roughly classified into twocategories depending on weather the information of eavesdroppers channels and locations is known or not.", "For the case that the information of eavesdroppers channels and locations is available, manymethods can be employed to improve physical layers security byoptimizing the artificial noise generation and power control. Incase that the global channel state information is available, toachieve the goal of maximizing the secrecy rates while minimizing thetotal transmit power, a few cooperative transmission schemes have beenproposed in{{cite:95cfb2756d75f9f9a289ce9919c95c235c43b311}}{{cite:4454c27bfae78133999ecfe35e9b0c211a8e752f}}{{cite:3a3b397ddc224938ea07081df2dfa8c01d08ac70}},and for two-hop wireless networks the optimal transmissionstrategies were presented in{{cite:556ebb178887ccabbc621ded45cab4ec9a3dbfb3}}{{cite:e1d897d5c39ce3b492a7a8fcd816de1a8d5ffc17}}. With respect to smallnetworks, cooperative jamming with multiple relays and multipleeavesdroppers and knowledge of channels and locations was consideredin {{cite:795d83ef7582dc5e6cbe1a605c96fb90e318d2ab}}{{cite:27e139363b63643972f0822ad4fb56e4080552aa}}. Even if only localchannel information rather than global channel state information is known, it was proved that the near-optimal secrecy rate can achieved by cooperative jamming schemes{{cite:f95000b1215012ea3e526e84f829b5d6b0e13a57}}{{cite:1214c6e7d21dea89c033e3918b1c1e33ac88b18d}}. Except channelinformation, the relative locations were also considered for optimizingcooperative jamming and power allocation to disrupt an eavesdropperwith known location {{cite:57527954280b12505d94d788590f07cb8dcdce14}}{{cite:f0c4bce2f4728d9aea1ffa397987d961b20e7e60}}.In addition, L. Lai et al. established the utility of usercooperation in facilitating secure wireless communications andproposed cooperation strategies in the additive White Gaussian Noise(AWGN) channel {{cite:2f9fc0cd5369dd1c28b5bd6e04c07d4f470d1d59}}, R. Negi et al. showed howartificially generated noise can be added to the information bearingsignal to achieve secrecy in the multiple and single antennascenario under the constraint on total power transmitted by allnodes {{cite:227eed6a90a4eeaba2a6085b634e51dcfbbde946}}. The physical layer security issues in atwo-way untrusted relay system was also investigated with friendlyjammers in {{cite:da2d7a98882222a6724978ee84fdab9f8dcaca41}}{{cite:b40f6df5e120ffcdfd13d6ec6bbc7cb058ce6442}}. Thecooperative communications in mobile ad hoc networks was discussed in{{cite:49046df69d515c727acd0b9ec29d6b39a1012aa5}}. Effective criteria for relay and jamming nodeselection were developed to ensure nonzero secrecy rate in case ofgiven sufficient relays in {{cite:91c7ef998bba0ddd7062fe7ae6ac65a32072ffa6}}.", "For the case that the information of eavesdropper channelsand locations is unknown, the works in{{cite:3f3d3bfa767a2de44b9f171e7347862a4dee314a}}{{cite:7578ff7d0308e088ce0e2fdaadf1bca2c32fc783}} considered thesecrecy for two-hop wireless networks, the works in{{cite:5916c96635f9fe7a08aa5aab4115c33ccb9148b6}}{{cite:9f6f3a4262b7974fbbf731142cdf2a8cb57c64f1}}{{cite:c5b94e933e07c4460cdbb9ca9228587b8e8ad460}}considered the secrecy for large wireless networks, and the furtherwork in {{cite:82c8ee637bb5b7a4466881d07fe19e7f132f3478}} considered the energy efficiencycooperative jamming strategies. These works considered howcooperative jamming by friendly nodes can impact the security of thenetwork and compared it with a straightforward approach based onmulti-user diversity. They also proposed some protocols to embedcooperative jamming techniques for protecting single links into alarge multi-hop network and explored network scaling results onthe number of eavesdroppers one network can tolerate. A.Sheikholeslami et al. explored the interference from multiple cooperative sessionsto confuse the eavesdroppers in a large wireless network {{cite:6d411da2ce61b0a46357e79e8618da9e32083d53}}. The cooperative relay scheme for the broadcastchannel was further investigated in {{cite:9cbbd9dcd6b6a8dc912a81278ceab9a9aa59b9e1}}{{cite:2442fe4f72f230b8607a8336519bed80753ddcbc}}." ]

[]

[ "In addition to the IBM models and the HMM alignment model previouslymentioned, one can find in the literature different approaches to theproblem of word-position alignment. In this section we focus on thoseapproaches that make use of SBI in some way; for a complete review ofthe state of the art in word alignment the reader is referred to{{cite:fa1aa77d2a629719099c2240b39a14992a5f9b5a}}.", "{{cite:5fa29cb273f31d70ab238505a8d1944ba9d5811e}} introduces the use of a bilingual dictionary as a SBIto obtain an initial alignment between seed words in a parallelcorpus. These seed words are chosen so that they cannot have multipletranslations (in both languages) and are frequent enough to becomeuseful references in both texts of the parallel corpus. These initialalignments are then used to align the other words appearing aroundthem in the parallel texts using an heuristic method similar to theone introduced by {{cite:3f12e2b9f1ca36e5267dbc77833515c509b3c2b9}}.", "{{cite:de68f9a1d9732a6e048dec9de4a3ad0a807bc356}} propose the use of a log-linear(maximum-entropy style) model {{cite:5de4ac1e316926e5bc4d15a4f58879a025d2bed7}} to combine the IBMmodel 3 alignment model with information coming from part-of-speechtaggers and bilingual dictionaries; the work was later extended toinclude new features and a new training procedure{{cite:d441a500120f7ea6e434dc720ab943d43c480b61}}. The main differences between their workand the one presented here are: (i) we do not rely on any previouslycomputed alignment model; (ii) we use any possible SBI which mayrelate multi-word segments, and (iii) they model the word-positionalignment task as a structured prediction problem{{cite:fa1aa77d2a629719099c2240b39a14992a5f9b5a}} that generates the whole alignmentstructure, whereas we model each association of positionsindependently. We will further discuss this last difference in thenext section." ]

[]

[ "We are interested in the problem of finding a {{formula:96eb5ec7-c0f0-4b0c-801e-14b02ad99164}} -approximate minimum vertex cover in a graph; i.e., one whose cardinality is at most {{formula:fd24aa2a-0b94-4fba-a605-5cd5f6d37f72}}  times that of the minimum vertex cover. For {{formula:6f005ca9-65ba-4ff2-9b53-a3ee4cc47d55}} the problem is easily solved in linear time {{cite:141b9de1bbcb86d3bf1935ebd4d2bc8fd4776307}}; for {{formula:ff1b5e77-66cd-41ea-afeb-6a4c967f4fd5}} the problem is known to be {{formula:1ea58197-19a1-42af-bf63-08c8c3225f3c}} -hard {{cite:b491ed87ee110e09fec4693a64e3264dbf3f68ca}}. But for {{formula:aff25117-c39a-4637-8676-4530f079cd5b}} , say, we know neither polynomial-time solvability, nor {{formula:fcb0f938-d15b-4d52-8a73-fc64d1027b38}} -hardness. Based on the nearly 30-year lack of progress on the algorithms side, it is reasonable to suspect that there is no efficient {{formula:cb232db5-60af-4b63-989d-19e0e62c202e}} -approximation algorithm. Indeed, this is known to be true assuming the Unique-Games Conjecture {{cite:b05ea90ec078952bd0baa3b52362c693d91fe01a}}. However there is reasonable doubt about the Unique-Games Conjecture {{cite:505719187de79d75ddf437d3e0ff668a86c8cf10}} and one may seek alternative evidence of hardness. One fairly strong form of evidence is showing that various linear programming-based and/or semidefinite programming-based generic polynomial-time optimization algorithms fail to even {{formula:a5ae9692-83b3-49ba-b412-665671c73b04}} -certify minimum vertex covers. Here we are using the following definition:We say that an algorithm {{formula:ac915737-2789-4e79-9614-fb9c5701c7a6}} -certifies the minimum vertex cover in a graph if, on input a graph {{formula:f78e7588-7e01-419b-8398-c3673ebec0d6}} , it outputs a number {{formula:c28edc9e-9f57-40a5-b585-31516ac280f2}} guaranteed to satisfy {{formula:2bc90591-a174-4fbf-b7d4-30b10d2d1ba0}} . (This is a strictly easier problem than actually finding {{formula:52296f48-5524-411b-b55f-23628a8c7caf}} -approximate minimum vertex covers.)There is a long line of work of this type {{cite:a004c2652f3e9b19e00f86326cea367f0926c97d}}, {{cite:011d52d9863cb887a8a23d59803431141402ed45}}, {{cite:305c86650dc9255eed1460695dda3d7e2afc2ad2}}, {{cite:7596df84d0a1d2b22e8f405f8555cedeb9d5b170}}, {{cite:b4df11ce4f2dd266955a4ddba10f112dda6447c3}}, {{cite:6a07a47aaf4a1e8c65895c692c5eb86297ef01e9}}, {{cite:89a778b80828eede8c063a3c2d390ffde51e1618}}, {{cite:5519da346be530e936c42183bac009aad14aef0d}}, {{cite:5fdaa24dad520f3a9fa77e91387ab8d0927afff0}}, {{cite:0f8210655df26fd545266f242f678e890bab1a20}}, {{cite:6b5409e989be6c6995e646ad8c93c381735a72b3}}, {{cite:389d201d71a5931b733475121fa3361a7540e51f}}, {{cite:7d9591a2a770eb1d750f34488cbfd141c06c68ad}}, {{cite:a4fc3839e5f05e272c0c8544eae70c3798658149}}, {{cite:b3433486ed458f67d262a94f9748ed55a4dd6dd7}}, {{cite:10cbeeaab0db6c5f9a228dfc8d3d552a2392272a}}; see Georgiou's thesis {{cite:44c965a4d2ab4bcd86e2a3927bc096c97803f41a}} for a recent survey. Briefly, it is known that the following algorithms fail to provide {{formula:2300561b-d902-415d-a9fd-33928078336f}} -certifications for the minimum vertex cover: {{formula:68478335-1ee4-4c00-a186-f2d35f54a924}} levels of the Sherali–Adams LP hierarchy {{cite:389d201d71a5931b733475121fa3361a7540e51f}} (where {{formula:1bc08655-4503-46df-80ba-44a63374f20e}} is the number of vertices); {{formula:6766f6cb-e273-472a-9b26-1ff99db106d8}} levels of the Lovász–Schrijver{{formula:7e2f34fe-8dc7-440f-8d74-02aa3e80aab2}} SDP hierarchy; and, 6 levels of the Sherali–Adams{{formula:119bc85f-f6e9-4b15-ae31-4ec417ccde61}} SDP hierarchy {{cite:10cbeeaab0db6c5f9a228dfc8d3d552a2392272a}}. Further, {{cite:10cbeeaab0db6c5f9a228dfc8d3d552a2392272a}} conjectures (based on numerical evidence) that their 6-level result can be extended to any constant number of levels. Since the Sherali–Adams{{formula:d7e480ba-17bc-4ae8-b78d-5bf4056b0edd}} hierarchy is stronger than the Sherali–Adams and Lovász–Schrijver{{formula:c58bc2a4-99dc-41a1-b13f-1f64e6095373}} hierarchies, this conjecture would subsume the other two mentioned results, at least with regards to ruling out polynomial-time (constant-level) algorithms.", "The Frankl–Rödl graphs, defined as follows, are used to show failure of the Sherali–Adams{{formula:8c99d7c5-a003-41eb-8de9-cba1ac59b4c4}} hierarchy to {{formula:3afa103f-c241-40b6-8a28-1aed6a9aa620}} -certify the minimum vertex cover. As far as we are aware, these are the only known “hard instances” even for the basic SDP relaxation (for factor {{formula:400fdc97-684c-4c8a-a247-58a0f4c3b35c}} ).", "Let {{formula:e4f416df-c16a-4a4a-a981-b14612d668ed}} and let {{formula:7cd13a13-1899-444e-8f68-ca8089204907}} be such that {{formula:55dc3019-3705-4eeb-83f4-09a10eca2b5a}} is an even integer. The Frankl–Rödl graph {{formula:7e636a31-0bd5-4cea-a02a-5087cd75617e}} is the undirected graph with vertex set {{formula:9709d98c-6059-424a-a5ca-d7d733337f25}} and edge set {{formula:e0cd84b1-b080-4b7c-af3c-fbac2892ca82}} , where {{formula:1b9daa3a-b2aa-43c6-982f-61370c32222c}} denotes Hamming distance.", "The following theorem is essentially due to Frankl and Rödl {{cite:7846c2b7d33130a10b38ab1d9badf306c1b7655b}}, though a few details of the proof are only worked out in {{cite:a4fc3839e5f05e272c0c8544eae70c3798658149}}. There exists {{formula:a775378a-bdeb-41bc-971c-f437806e9f7f}} such that for all {{formula:79ef818c-5c43-47f8-96b5-417586280487}} it holds that {{formula:17bdfb04-e3fb-4f12-8bfc-25e1b7600172}} , where {{formula:2d776607-74a4-44b5-9390-ab510e4efc34}} is the number of vertices in the graph. In particular, {{formula:41405c2f-78f8-45d9-95ab-d8bb6faf8fe7}} for {{formula:84317461-538b-4a2a-a261-178e84c8c234}} .", "On the other hand, Benabbas, Chan, Georgiou, and Magen {{cite:10cbeeaab0db6c5f9a228dfc8d3d552a2392272a}} showed that for all {{formula:5417650c-1023-4854-93a2-b6d87a568991}} there exists {{formula:e1418ea8-fb19-4877-bc77-204dea9ce817}} such that the level-6 Sherali–Adams{{formula:7807d900-e12e-4ac2-bfda-a1e12d20fd8d}} algorithm can only certify that {{formula:8d463f09-466f-4291-97db-b54315352dce}} (for {{formula:1c418b1c-73c5-49d2-b996-35d15970233b}} sufficiently large). They further conjectured that this remains true for level-{{formula:e048d13b-8ec7-427f-9a8e-a44f1110e7a2}} Sherali–Adams{{formula:72e8165f-d059-4ab1-a439-8070a9052989}} assuming {{formula:a7aab9a0-8e88-46cc-94b5-bebe8324629f}} is small enough. Specifically, they suggest fixing {{formula:a1bba116-ea52-45e5-a894-4b6ebc7457d7}} ." ]

[]

[ "Although category presentations – directed mutli-graphs with path equivalence constraints – are a very common notation for schemas {{cite:08fdcecbceb1fdf7ae0608e93c8192d6bd368b90}}, there has been very little work to treat such schemas categorically {{cite:c3a101208fc464ce5f29573c073d0dd6bf1115fd}}. Instead, most work treats such schemas as abbreviations for relational schemas. For example, in Clio {{cite:8d4fad6c80eeb69cac5364c8ebf8b6be918c7071}}, users draw lines connecting related elements between two schemas-as-graphs and Clio generates a relational query (semantically similar to our {{formula:c9edc08a-c67c-4f35-8b5f-fdf95a7903b2}} ) that implements the user's intended data transformation. Behind the scenes, the user's correspondence is translated into a formula in the relational language of second-order tuple generating dependencies, from which a query is generated {{cite:25a3bf87d802f1cc4e94ccd8f65591f048be4423}}.", "In many ways, our work is an extension and improvement of Rosebrugh et al's initial work on understanding category presentations as database schemas {{cite:c3a101208fc464ce5f29573c073d0dd6bf1115fd}}. In that work, the authors identify the {{formula:6421e56b-1634-40d3-b45c-a58c48a3adfe}} and {{formula:15b46f77-e962-4652-809c-a0c8fae7ee10}} data migration functors, but they do not identify {{formula:3308f059-fb0a-4bb4-8b40-213ea019e6b3}} as their adjoint. Moreover, they remark that they were unable to implement {{formula:7a2b6748-b0f4-4d10-a7ea-cce2138f629a}} and {{formula:8030653e-4456-46a9-9c1a-6d366397b742}} using relational algebra, they do not formalize a query language or investigate the behavior of the {{formula:e6a3be8b-0507-4377-9b80-7df343828214}} and {{formula:f52043f4-0464-4384-b2ce-2b9acdc97bce}} with respect to composition, and they do not consider “attributes”. Our mathematical development diverges from their subsequent work on “sketches” {{cite:a86dc4632888fe8e53bfe72c41b74641efb3ade6}}.", "Category-theoretic techniques were instrumental in the development of nested relational data model {{cite:cce4e0b76a4352845657f465c853a4df9be10e47}}. However, at this time we do not believe the functorial data model and the nested relational model to be closely connected. The functorial data model is more closely related to work on categorical logic and type theory, where operations similar to {{formula:a14c982b-af57-4791-aab0-14162ec1103d}} , and {{formula:2dd6a8a5-af70-40d1-9829-62eda588b613}} often appear under the slogan that “quantification is adjoint to substitution” {{cite:936f2397df87a87a360e827a0019f95fbdc0a2c2}}." ]

[]

[ "An epidemic working through a population, cascading electrical power failures, product adoption, and the spread of a mutant gene are all examples of diffusion processes that can happen in multi-agent systems structured as complex networks. These network processes have been studied in a variety of disciplines, including computer science {{cite:c1fccad04989ae2156493718174a78bc0e49e994}}, biology {{cite:e19a5430bc280a234010fb5f1be884c0b3ab545a}}, sociology {{cite:dbc32d6bad2926e1748effe732f3efcf1bd67d3a}}, economics {{cite:0c9b913459d3e15844949cbd8a3ba269f22b6482}}, and physics {{cite:4fb08764f0a18e636475f077442c992b7f717373}}. Much existing work in this area is based on pre-existing models in sociology and economics – in particular the work of {{cite:dbc32d6bad2926e1748effe732f3efcf1bd67d3a}}, {{cite:0c9b913459d3e15844949cbd8a3ba269f22b6482}}. However, recent examinations of social networks – both analysis of large data sets and experimental – have indicated that there may be additional factors to consider that are not taken into account by these models. These include the attributes of nodes and edges, competing diffusion processes, and time. In this paper, we outline seven design criteria (Section REF ) for such a framework and introduce {{formula:06a7669a-9b00-4d8a-b294-ea00e5051997}} (Section ), which is to the best of our knowledge the first logical language for modeling diffusion in complex networks that meets these criteria. {{formula:5a14930e-2cd8-4fb2-ad3e-9e10ab650898}} is a rule-based framework (inspired by logic programming) that can richly express how agents adopt or fail to adopt certain behaviors, and how these behaviors cascade through a network. We also introduce fixed-point based algorithms that allow for the calculation of the result of the diffusion process in Section . Note that these algorithms are proven not only to be correct, but also to run in polynomial time. Hence, our approach can not only better express many aspects of cascades in complex networks, but it can do so in a reasonable amount of time. We conclude by discussing applications of {{formula:2a46812e-6e52-4d4e-be94-4903d6e6f7f7}} in Section .", "Proofs of all results stated in this paper can be found in the appendix.", "The above criteria can be summarized as the desire to design the most expressive language for network cascades possible while still allowing computation of the outcome of a diffusion process to be completed in a tractable amount of time. As a comparison, let us briefly describe some relevant related work. Perhaps the best known general model for representing diffusion in complex networks is the independent cascade/linear threshold (IC/LT) model of {{cite:c1fccad04989ae2156493718174a78bc0e49e994}}. However, although this framework was shown to be capable of expressing a wide variety of sociological models, it assumes the Markov property and does not allow for the representation of multiple attributes on vertices and edges. A more recent framework, social network optimization problems (SNOPs) {{cite:6f65ea641af8f02bc8b8ee07d466d4ab972de9a1}} uses logic programming to allow for the representation of attributes, but this framework does not allow for competing processes or non-monotonic cascades. A related logic programming framework, competitive diffusion (CD) {{cite:116613e22e543711dcc10409f28c91dd50154d29}} allows for competitive diffusion and non-monotonic processes but does not explicitly represent time and also makes Markovian assumptions. Further, we also note that the semantics of CD yields a “most probable interpretation” that is not a unique solution. Hence, a given model in that framework can lead to multiple and possibly contradictory, outcomes to a cascade (this problem is avoided in {{formula:50963dd4-56bf-4cd0-ac8c-4d74f2459cfb}} ). Another popular class of models is Evolutionary Graph Theory (EGT) {{cite:e19a5430bc280a234010fb5f1be884c0b3ab545a}}, which is highly related to the voter model (VM) {{cite:4fb08764f0a18e636475f077442c992b7f717373}}. Although this framework allows for competing processes and non-monotonic diffusion, it also makes Markovian assumptions while not explicitly representing time. Further, determining the outcome of a cascade in those models is NP-hard, while determining the outcome in {{formula:825448e5-6a05-40b3-a7b3-650fa01608e2}} can be accomplished in polynomial time. Table REF lists how these models compare to {{formula:ff2fd3c1-847e-48b8-9327-beba837e6b8c}} when considering our design criteria.{{table:8145e7d2-12de-40bd-b57f-5ca3eb6d10f2}}" ]

[]

[ "Natural conjugate gradients for variational inference are investigatedin {{cite:59f59858746edf88dfa6b95aa93fa3b0b49a91ef}}, but the analysis is mostly devoted tothe case where the variational approximation is Gaussian, and theresulting gradient equation involves an integral which is notnecessarily tractable.", "The use of variational inference in probabilistic programming isexplored in {{cite:2cfa0f4e79b3201c91351af84a572410a1f7de61}}. The authors similarly notethat it is easy to sample from the variational program. However, theyonly use this observation to estimate the free energy of thevariational program, but they do not estimate the gradient of thatfree energy. While they do highlight the need for optimizing theparameters of the variational program, they do not offer a generalalgorithm for doing so, instead suggesting rejection sampling orimportance sampling.", "Use of stochastic approximations for variational inference is alsoused by Carbonetto {{cite:604ca3ef653484bb60e5bb9b275e9d7766bbfa0f}}. Their approach isvery different from ours: they use Sequential Monte Carlo to refinegradient estimates, and require that the family of variationaldistributions contains the target distribution. While their approachis fairly general, it cannot be automatically generated forarbitrarily complex probabilistic models.", "Finally, stochastic gradient methods are also used in online variationalinference algorithms, in particular in the work of Blei et al. in stochastic variational inference (for instance, online LDA {{cite:c0c080c988dcc389ac2f2d0e549953fad50d8772}}, online HDP {{cite:e85764c67318c003b080e57a84b2042b27503c3e}}, and more generally under conjugacy assumptions {{cite:53023ac197d3ebd5a09b8fd1436787fd6be3822e}}), as a way to refine estimates of latent variable distributions without processing all the observations. However, this approach requires a manual derivation of the variational equation for coordinate descent, which is only possible under conjugacy assumptions which will in general not hold for arbitrary probabilistic programs." ]

[]

[ "A lot of research works have been dedicated to load-balancetransmission scheme for balanced energy consumption among systemnodes to prolong the network lifetime in wireless networks. A fewdynamic load balancing strategies and schemes were proposed in{{cite:f15180a683e6ca5f638457f111203b1e20fc057a}}{{cite:b1ec3e2e88dcaae6d6d2aeb4922a4cb69cfb7cf8}} for distributedsystems. For wireless mesh network, a multi-hop transmission schemeis proposed in {{cite:a1f4ca6691f809fe526fd04559cfefb4ac2b043e}}, in which information relay isselected based on the current load of the relay nodes. For wirelessaccess networks, a distributed routing algorithm that performsdynamic load-balance by constructs a load-balanced backbone tree{{cite:a89e13d52cf2c7d297bc6b02018b1fe1f972253e}}. J. Gao et al. extended the shortest pathrouting to support load-balance {{cite:0d3113bb7e7c3ad46ddb6194a10f20125adf0861}}. In particular,for energy constrained wireless sensor networks, load-balance issignificant important, and a lot of transmission schemes wereproposed for load-balance among relays and prolonging the networklifetime{{cite:79ebc0aee75053825a03053e1e7a09a2489cf7b5}}{{cite:20c44c7922dfa78a5716467fda6c12ec757a5291}}{{cite:e10199d214136249019833a2bdb0a8fd1bed686f}}.Lifetime optimization and security of multi-hop wireless networkswas further considered and the secure transmission scheme withload-balance is proposed in{{cite:082e6a702f1016e864a7ed8bc620209ef4b28bf8}}{{cite:5adc9e952beab0bf06399b1fed42415b60335242}}.", "Recently, attention is turning to achieve physical layer secrecy andsecure transmission scheme via cooperative relays is considered inlarge wireless networks. Some transmission protocols are proposed toselect the optimal relay in terms of the maximum secrecy capacity orminimum transmit power. In case that eavesdropper channels orlocations is known, node cooperation is used to improve theperformance of secure wireless communications and a few cooperativetransmission protocols were proposed to jam eavesdroppers{{cite:947ef70c671636d407c407b3d928af3d6e44fe8c}}{{cite:3071c5fbc80513bf7ce648fd4ab4f520996c566f}}. In case thateavesdropper channels or locations is unknown, D. Goeckel et al.proposed a transmission protocol based on optimal relay selection{{cite:afd4b41cfc2056430e1ae4a60eae565a32a13c4a}}{{cite:d0e96756823e112fed5af306ba3fdbce58c444fe}}. For bothone-dimensional and two-dimensional networks, a secure transmissionprotocol is proposed in {{cite:eb14a9f654d362cddf577904cfc1bb6a526753a5}}. Z. Ding et al.considered the opportunistic use of relays and proposed two secrecytransmission protocols {{cite:13b01dbee652f32d982e07c0d024d5ad5d5c8b6f}}. The \"two-way secrecyscheme\" was studied in {{cite:d1a70151c8abc126a88dc703f3858789c5584297}} {{cite:1414a526055f464d4927b7c4ff5de3b7dd08f733}}and M. Dehghan et al. explored the energy efficiency of cooperativejamming scheme {{cite:a65679bd8812ee904ba2038db2c3f94fe65d7f04}}. A. Sheikholeslami et al.proposed a protocol, where the signal of a given transmitter isprotected by the aggregate interference produced by the othertransmitters {{cite:375285b5049c82bdce639c947ea8bdb21af544b4}}. A secure transmissionprotocol are presented in case where the eavesdroppers collude{{cite:78ace17a2bb86091a11a9a3b792a9e438cec3517}}. J. Li et al. proposed two securetransmission protocols to confound the eavesdroppers{{cite:162126ed998743881a19c4185c80ebea5dc48550}}. The above works mainly focus on the maximumthe secrecy capacity, in which the system nodes with best linkcondition is always selected as information relay. Such, theseprotocols have less load-balance capacity. In order to address thisproblem, Y. Shen et al. further proposed a protocol with randomrelay selection in {{cite:71506f7a7fcf2d9d2a5dfa282af97346c67c6a0c}}{{cite:eed0c74b2cd74f59b04a325fecaac0242a9358b9}}.This protocol can provide good load-balance capacity and balancedenergy consumption among the relays, whereas it has low transmissionefficiency." ]

[]

[ "Evolutionary graph theory was originally introduced in {{cite:f0f450880574ab0f597de57f7d590299b71eb672}}. Previously, we have compiled a comprehensive review {{cite:e517c71c41e7c480a33107ffe6a1bc24d786e777}} for a general overview of the work in this exciting new area.", "While most work dealing with evolutionary graphs rely on Monte Carlo simulation, there are some good analytical approximations for the undirected/unweighted cased based on the degree of the vertices in question. Antal et al. {{cite:c6d545019c05ff68a7549b1c65cb5a419e4ceacb}} use the mean-field approach to create these approximations for the undirected/unweighted case. Broom et al. {{cite:a1280357ad18d0a2091a9c5c5277763a1c3ce663}} derive an exact analytical result for the undirected/unweighted case in neutral drift, which agrees with the results of Antal et al. They also show that fixation probability is additive in that case (a result which we extend in this paper using a different proof technique). However, the results of {{cite:e1cad9e990b46a69dd388a6591ec9c5585c74bc5}} demonstrate that mean-field approximations break down in the case of weighted, directed graphs. {{cite:273ac3ffbe40f711ce0794dd8170ae91f679f5eb}} also studied weighted, directed graphs, but does so by using Monte Carlo simulation. {{cite:6cb7d542dd297ab15fc0daad2ebd0e533d28f8ca}} derive exact computation of fixation probability through means of linear programming. However, that approach requires an exponential number of both constraints and variables and is intractable. The recent work of {{cite:db9e4081e49221fe17cc8d813c497f48391c1e76}} introduces a parameter called graph determinacy which measures the degree to which fixation or extinction is determined while starting from a randomly choses initial configuration. This property is then used to analyze some special cases of evolutionary graphs under birth-death updating. There has been some work on algorithms for fixation probability calculation that rely on a randomized approach {{cite:5929ce3fde51975d89114000ceffca0af0baf3fd}}, {{cite:3e1a0bd24d1e3c2fe23fefa79fce7f7a5acf2cc5}}. {{cite:5929ce3fde51975d89114000ceffca0af0baf3fd}} present a heuristic technique for speeding up Monte Carlo simulations by early termination while {{cite:3e1a0bd24d1e3c2fe23fefa79fce7f7a5acf2cc5}} present utilize simulation runs in a fully-polynomial randomized approximation scheme. However, our framework differs in that it does not rely on simulation at all and provides a deterministic result. Further, our non-randomized approach also allows for additional insights into the evolutionary process - such as monitoring the expected number of mutants as a function of time. Recently, {{cite:24b7cb3b6c522542d06155bd9a34578607863be1}} study the related problem of determining the probability of fixation given a single, randomly placed mutant in the graph where the vertices are “islands” and there are many individuals residing on each island in a well-mixed population. They use quasi-fixed points of ODE's to obtain an approximation of the fixation probability and performed experiments with a maximum of 5 islands (vertices) containing 50 individuals each. This continuous approximation provides the best results when the number of individuals in each island is much larger than the number of islands. As the problem of this paper can be thought of as a special case where each island has just one individual, it seems unlikely that the approximation of Houchmandzadeh and Vallade's approach will hold here.", "Some of the results in this paper were previously presented in conferences by the authors {{cite:096e1c5b43b6c76689799a940702999e215438f1}}, {{cite:48ebe86353e2c9d40d7faaef76316b1248b67e91}}. The analysis and experiments concerning the expected number of mutants at a given time, the extension of the framework for other update rules (beyond birth-death), the use of the framework for the case of {{formula:7b77bda0-3ece-4e8b-8fb0-c1c742e1e98f}} , and the neurology applications are all new results appearing for the first time in this paper." ]

[]

[ "Algorithms for distributed sparse recovery (with no central processing unit) in sensor networks have been proposed in the literature in the last fewyears. We distinguish two classes:", "algorithms based on the decentralization of subgradient methods for convex optimization (DSM, {{cite:02a1a1b7276a11f8db62df6ce2b46484848c2272}});distributed implementation of the alternate method of multipliers (ADMM, {{cite:d2fe3ce84976e146c306c872cfe9f43fbb6c8b7f}}, {{cite:ff463d5ab3043f6f52aa756b81a7e7503f8c6aaf}}, {{cite:4a65a3a510ad41f7ae82529e98ff2de319988972}});" ]

[]

[ "While information-centric networks and software defined networking has been explored as separate areas of research in the past, recent trend in adoption of OpenFlow has sparked research on combining the two. The benefits are very obvious: while SDN brings low maintenance and reduced network complexity, distributed caching brings low latency. Some recent findings in this vein includes {{cite:1866d7351a159c4ccea63a317aa9eaa56178cd16}} which proposes a application driven routing model on top of OpenFlow. Another notable work is {{cite:bc8b889322fa935801126e1a65ef44c3900dd328}} which proposes a system to dynamically redirect requests from a client based on content. Note that our approach is different from this, we have proposed a distributed caching system based on content based switching, thus extending {{cite:bc8b889322fa935801126e1a65ef44c3900dd328}}.", "Software defined networking (SDN) decouples the control plane and the forwarding plane of a network. The forwarding plane then exports some simple APIs to the control plane, which then utilizes these APIs to provide desired network policy. The decoupling allows for the control plane to be written in software, and thus be programmable.", "Current approaches for SDN, such as OpenFlow {{cite:452335f4015ac3b941af93a253fd5af46097e1f1}}, focus on controlling switching element, and adopt a definition of the forwarding plane which takes traffic flows as the unit to which a policy is applied.", "However, this approach suffers from two limitations: 1) The network might include other non-forwarding elements; it is the common view of most future Internet architectures {{cite:41ea647aa23ec6cb870fff872d275e321b4c3057}}, {{cite:a7f3d53231d7c34d77008934c205d6ea3d80edbe}}, {{cite:4a4fa86ddcc6e00eca66d299b58133e955312531}} and many commercial products {{cite:4486995770a34fca2a3a360b0459b2b0257a5d3d}} combine a switching element with additional capability, such as storage and processing; these capabilities need to be advertised to the control plane so that the programmable policy takes them into account. 2) The network might need to consider different policy for different objects at a finer granularity. Currently, OpenFlow ignores the specific content and provides a forwarding behavior that is per-flow based, where the flow is defined as a combination of the source and destination MAC and IP addresses and protocol ports plus some other fields such as MPLS tags, VLAN tags. Many architectures attempt to place content as the focus of the architecture {{cite:a4e82619e534a01ce6b33996671b7f42468bcfd5}}, {{cite:50e77684456f83806d6e1d0a9ce0f0b9935580ad}}, {{cite:ae6795f8aecd99af8802da4e1fa74255a9b1db0a}}, {{cite:1f47bc5fed769bb746fbd54cb9efa7dc8383b447}} and this is not supported by the current SDN proposals.", "SDN has been considering L4-L7 extensions (say by Citrix or Qosmos), but those often take the form of DPI modules which do not integrate the routing decisions based upon the observed content. To operate on the granularity of content, many new architectures {{cite:41ea647aa23ec6cb870fff872d275e321b4c3057}}, {{cite:a7f3d53231d7c34d77008934c205d6ea3d80edbe}}, {{cite:4a4fa86ddcc6e00eca66d299b58133e955312531}}, {{cite:50e77684456f83806d6e1d0a9ce0f0b9935580ad}}, {{cite:bca3e3e76a8ca83e52a038b80eaa5bb8bae00841}} have been proposed, which make identifying content and forwarding based upon data possible. See {{cite:89bae29d7268f34820f1b7f864e6b9047bd6246d}} for a short survey on ICNs. However, these architectures typically require to replace the whole network while our architectural model is based upon inserting a content-management layer in the SDN controller.", "Other proposals to extend the SDN framework include integration with the application layer transport optimization {{cite:4542e1f4bd40d01c31db84458326d063ba2aca39}}. In an earlier paper {{cite:06b7d3cd871fdc4eddade9d7fa7119d717921db2}}, we have discussed extending the definition of network elements to include caches. However, this still requires a mapping layer between content and flow which we introduce here. {{cite:e6dfc05859757012511cacddd049ec9065326172}} proposed to use HTTP as the basis for an ICN, and our work finds inspiration in this idea, as we focus on HTTP content, in our implementation in particular. The extensions to HTTP from {{cite:e6dfc05859757012511cacddd049ec9065326172}} could be used in ContentFlow, but would require modifications of the end points." ]

[]

[ "ALTD {{cite:c89e3d13f0e7e6d17776e45d28493c0a12da6254}} is an effort to track software and library usage at HPCsites. It takes a proactive approach by intercepting and recording everyinvocation of the linker and the job scheduler. Our work is complementaryin that it performs post-mortem analysis and works on systems without ALTD.", "The work by Rosenblum et al {{cite:325b282b43df12f4016816a243428aaa1b8aadf7}} is the first attempt toinfer the compiler provenance. They used sophisticated machine learning bymodeling and classifying the code byte stream as a linear chain ConditionalRandom Field. As in most supervised learning systems, a lengthy trainingphase is required.The resulting system can then infer the source compiler with a probability.Their approach has several drawbacks, which our method addresses: Theyfocus solely on executable code and ignores other parts of ELF files, thepreprocessing/training phase, albeit one-time, is slow and complex, themodel parameters cannot be updated incrementally with ease when a newcompiler is added, and it is unclear if their model can discern the nuancesamong different versions of the same compiler.", "Kim's approach {{cite:7fad5281a8c5971320aec0c08a6127417b54d4f7}} is closest to ours in spirit, but it misses thekey feature in our implementation: the relocation table. It produces a signature bycopying the first 25 bytes of a library function code verbatim. Withsuch a short signature and lack of relocation information, his tool hasvery limited success in identifying library code snippets." ]

[]

[ "B. Sotomayor et al. {{cite:d6b992ef4368fd231539295fa7eb52d148a74bd8}} proposed a lease-based model and First-Come-First-Serve (FCFS) and backfilling algorithms to schedule best-effort, immediate and advanced reservation jobs. The FCFS and backfilling algorithms consider only performance metric (e.g. waiting time, slowdown). To maximize performance, these scheduling algorithms tend to choose free load servers (i.e. highest-ranking scores) when allocates a new lease. Therefore, a lease with single VM can be allocated on big, multi-core physical machine. This way could be waste energy, both of the FCFS and backfilling does not consider on the energy efficiency.", "S. Albers et al. {{cite:2f8cfa553526ddea6487dcf22ad044f3ae210469}} reviewed some energy efficient algorithms which are used to minimize flow time by changing processor speed adapt to job size. G. Laszewski et al. {{cite:698fd003e310adbe9f286dfb4156c220c5bdfc48}} proposed scheduling heuristics and to present application experience for reducing power consumption of parallel tasks in a cluster with the Dynamic Voltage Frequency Scaling (DVFS) technique. We did not use the DVFS technique to reduce energy consumption on datacenter.", "Some studies {{cite:c9c630bdcd540cdeabb32ec9a4834acad3d24cad}}{{cite:129157caa7b4edc8f7a6e069a612843540f42ceb}}{{cite:8d77df60e8f6398aa2c9edabbd628eeff817219c}} proposed algorithms to solve the virtual machine allocation in private cloud to minimize energy consumption. A. Beloglazov et al. {{cite:129157caa7b4edc8f7a6e069a612843540f42ceb}}{{cite:8d77df60e8f6398aa2c9edabbd628eeff817219c}} presented a best-fit decreasing heuristic on VM allocation, named MBFD, and VM migration policies under adaptive thresholds. The MBFD tends to allocate a VM to such as active physical machine that would take the minimum increase of power consumption (i.e. the MBFD prefers a physical machine with minimum power increasing). However, the MBFD cannot find an optimal allocation for all VMs. In our simulation, for example, the GAPA can find a better VM allocation (lesser energy consumption) than the minimum increase of power consumption (best-fit decrease) heuristic in our experiments. In this example, our GAPA will choose one Dell server to allocate these 16 VMs. As a result, our GAPA consumes less total energy than the best-fit heuristic does.", "Another study on allocation of VMs {{cite:c9c630bdcd540cdeabb32ec9a4834acad3d24cad}} developed a score-based allocation method to calculate scores matrix of allocations of {{formula:13dc8dc9-9580-43ad-b16e-3e489203e9a2}} VMs to {{formula:e6e80513-c943-4abf-8c54-f4914909f78e}} physical machines. A score is sum of many factors such as power consumption, hardware and software fulfillment, resource requirement. These studies are only suitable for service allocation, in which each VM will execute a long running, persistent application. We consider each user job has a limited duration time. In addition to, our GAPA can find an optimal schedule for the static VM allocation problem on single objective is minimum energy consumption.", "In a recently work, J. Kolodziej et al. {{cite:8fdfd6ab8e084a6fd9e9144866f661eba3d87778}} presents evolutionary algorithms for energy management. None of these solutions solves same our SVMAP problem." ]

[]

[ "Cluster analysis is a basic problem in various disciplines {{cite:acc9cb6e9962875d3472f77958c0617c772e31e7}}, such as pattern recognition, data mining and computer vision, and a huge number of such methods have been proposed. It is beyond the scope of this paper to list all of them, therefore, we focus on methods closely related to ours.", "The discrete replicator dynamic has been used for cluster extraction for a very long time, probably dating back to the well-known work of Motzkin and Straus {{cite:1e5f0804c5ab4ad604812cfc4103f8fe64e7126a}}, which relates the global maximum of (REF ) with the maximum clique of graph. The maximum clique problem {{cite:5f1dfbfdef0db68f73fcd7aa0935eddbb792b963}} is a very fundamental problem in computer science, but it is NP-complete {{cite:1e5f0804c5ab4ad604812cfc4103f8fe64e7126a}}. According to the Motzkin-Straus theorem, we can get the maximum clique by solving (REF ). In {{cite:88c96333d8a91325b4cb2d14e7d823407d66123b}}, the discrete replicator dynamic has been used to extract cluster on weighted graph. The obtained cluster, called dominant set, is a generalization of the concept of maximal clique. To efficiently enumerate dense clusters on graph, a fast algorithm has been proposed in {{cite:78aa9e030da7f915448b299eb254834dc86718db}}, which evolves to a dense subgraph by iteratively shrinking and expansion. In the shrink stage of this method, the discrete replicator dynamic is used to extract dense clusters. The clustering methods based on replicator dynamic have been successfully applied to many tasks, such as image segmentation {{cite:88c96333d8a91325b4cb2d14e7d823407d66123b}}, point set and image matching {{cite:78aa9e030da7f915448b299eb254834dc86718db}}, {{cite:7f2affe2dbfcbfa5234c7199795f5b6b28bfb8b5}}, {{cite:89ce7ee68c19ac55c53f4c575f688f5c16b6aec1}}, {{cite:3b28d98424e3cef5a9a38c6d6d8fc4566617d25b}}, {{cite:be7ca600c435ddd76d0f0cb0d49d444382a984f5}} and stereo correspondence {{cite:4922fdb60024cc92b7221b989520e59d4bb01f89}}. All these methods rely on discrete replicator dynamic to get the global optimum of (REF ); however, discrete replicator dynamic (REF ) is very sensitive to the degree distribution of graph {{formula:bee6357d-5c59-4041-8bdd-380c4761fb39}} , and usually does not evolve to the global maximum of (REF ). Since our approach has a much higher probability to evolve to the global maximum of (REF ) than discrete replicator dynamic, it is more suitable for these applications.", "Methods which try to extract dense clusters of certain sizes are also related to the proposed approach. For example, the methods to solve the densest {{formula:0fc78ea3-03cd-4e69-ba86-b20c812f2002}} -subgraph problem (D{{formula:e4528305-cf47-462c-b6ff-b25cd0896824}} S) {{cite:604f08258a5a65780193f3a4877554796a3f5bbd}}, {{cite:3cc1f1b99ec2e3c3072d61168de4519f1d760d7b}}. In hypergraph clustering, Liu et al. {{cite:8c54366a028606444d6586778d3f47a5e80bb98b}} proposed a method to control the least size of extracted clusters, and showed state-of-the-art results. Utilizing the fact that dense clusters are very robust, they also generalized {{formula:997d09c6-2ae6-432a-8b3c-e24100a60eee}} NN and proposed an alternative, {{formula:44fd6f0a-1d76-45c2-bca3-a0d40ef2b612}} DN {{cite:bc777ebf62b5e9e669699d8f4131057885715094}}. {{formula:55a56420-f8e8-4b60-9ce7-8937788d30f3}} DN is in fact a dense cluster of size {{formula:61d4acd8-35a7-4ed7-a73d-502ff54e14e9}} which has a strong connection with an object. Obviously, our approach can be used to solve these problems. In fact, our approach is more robust. In our approach, the parameter {{formula:69228637-c655-4b6c-a877-1eaf83d08931}} controls the least size of obtained clusters. Moreover, our approach can simultaneously and robustly find dense clusters of different sizes in one run of evolution process.", "Since dense subgraphs correspond to high-density regions in the data, the evolution process of path following replicator dynamic can be considered as ashrink process of high-density regions. The task of estimating high-density regions from data samples is a fundamental problem in a number of works, such as outlier detection and cluster analysis {{cite:a4f1cdf8b2f9bc91166ea35d734c1447bee8a187}}, {{cite:7138bfde4f52877f70df86a6dac54482230fe22a}}. The advantage of our method for this task is that our method can gradually reveal the landscape of multiple high-density regions of various shape at different scales. High density regions usually represent modes of data, and in this sense, our method is also closely related to mode-finding methods, such as mean shift {{cite:b237c1d8f1a36101a4b7171f71f44689eb74b738}}." ]

[]

[ "Though spatial reasoning has been studied extensively in AI {{cite:a768ebcfcda5aeb5dba9f117040d1af899f6dce0}}, {{cite:29e8384930982bc26257833c70a0fa020e97cdf1}}, {{cite:e9a9ca370ffa907b5522215ae066084c17878041}}, {{cite:fa91a26b3ce078b3fe7b6761bbffe409fc138162}}, many of the paradigms that have emerged for such reasoning are qualitative in nature. Such qualitative spatial reasoning efforts include the influential region connection calculus for qualitative reasoning about space. There has also been work on quantitative methods for reasoning about space {{cite:d298cf2119dc1176630a8f1bbb2ad97ad32757cb}} which contains articles on spatial reasoning in the presence of uncertainty using both logical and fuzzy methods. Spatial reasoning with quantitative information has been studied extensively in image processing {{cite:97bf8be1e2db76e1052f236e86f168c2659f2683}}, {{cite:62ff911a64941adf2959c04d32fa036779b7e369}}.", "However, unlike this vast body of work, this paper focuses on a different problem. Suppose we are dealing with a map {{formula:d1293637-458a-44dd-b44e-81b1ed1a5566}} , a cost function {{formula:d246afb9-a637-4264-bf2c-2ee07e4f8103}} , a set {{formula:09f2366a-3720-4173-9255-2d9f2b097aef}} of possible actions, a bound on the cost {{formula:a98b48d6-fead-4f7c-aa7e-dc21a0cccc8e}} , and a bound on the number of actions we can take, what set of actions should be taken so as to optimize a given objective function. Two versions of this problem are studied in this paper - GBGOP and BMGOP which differ in what they optimize. Both problems are proved to be NP-hard (NP-complete under realistic assumptions) and we further prove that the number of solutions to GBGOP is #P-complete. We also find limits on approximating an optimal solution to BMGOP and GBGOP (in PTIME) under accepted theoretical assumptions. We develop integer programming formulations of both problems and then present a way of simplifying the IP for GBGOP. We further present the BMGOP-Compute algorithm for BMGOP and show that it is polynomial and has a guaranteed approximation ratio (though not high enough to contract the NP-hardness result)." ]

[]

[ "While there is a wide literature that proposes frameworks to interact with commercial controllers like the WiiMote (eg: {{cite:4ba57f419630d4e4373ae14701477062c3a3be6a}}) and general purpose sensor nodes (eg. {{cite:e279f145892fd8f73ac7d0ea05d6bcd6d3d60096}}) there are only few works proposing frameworks composed by both hardware and software libraries for prototyping orientation and motion sensing. In {{cite:d9a39f9fa12951effcc5b3f5a3381df802821ba4}} the authors propose a framework composed by a wireless sensor board and gesture recognition algorithms. While their work on the algorithms is still interesting, their sensor hardware is now considerably outdated and the sensors used in FreeIMU are considerably more accurate, faster and easier to interface.", "DUL Radio{{cite:f44a5381af59742248777f38542edfc04e1410e8}} is a small sensor board containing an accelerometer, a microcontroller and wireless communication which has been designed for ease-of-use. While this device may be interesting for inexperienced developers, it may be also constraining compared to Arduino, as also noted by the authors. Our framework also aims to ease sensor based interaction prototyping, however we concentrated on not loosing the flexibility and extensibility of the Arduino platform while providing wider sensor input.", "Concerning commercial projects similar to our, when we started working on FreeIMU back in 2010, there wasn't any similar board using all digital sensors. That's actually why the development of FreeIMU started. At the moment however, some prototype friendly boards containing 9 DOM MARG sensors are available. Sparkfun 9 DOF Sensor Stick https://www.sparkfun.com/products/10724, Razor IMUhttps://www.sparkfun.com/products/10736 or 3D Robotics ArduIMU+ v3http://store.diydrones.com/ArduIMU_V3_p/kt-arduimu-30.htm are example of similar commercial products which also are Open Hardware and some are equipped with minor software libraries.", "Vectornav VN-100 http://www.vectornav.com/products/vn100-rug and X-IO X-IMU http://www.x-io.co.uk/node/9 are instead remarkable examples of closed source commerial IMUs since they provide ready to use proprietary orientation sensing algorithms.", "From an hardware point of view, the FreeIMU hardware is superior to both the Sparkfun 9 DOF Sensor Stick and Razor IMU since the sensors used in the FreeIMU v0.4.3 board are more precise and less noisyThe ADXL345 accelerometer used in the Sparkfun 9 DOF Stick and Razor IMU only gives 256 digits/g of resolution ({{formula:f949b4c5-ef06-4d6d-8863-857e83241a04}} scale) while the MPU6050 of the FreeIMU v0.4.3 gives 16,384 digits/g ({{formula:d3e8660d-1afb-4058-9553-ac9a64370284}} scale) thus being much more precise. Same considerations are applicable to the gyroscope.. The FreeIMU also features a barometer allowing to track the altitude of the prototyped device. The Razor IMU is also limited by being clocked at 8MHz while the FreeIMU is compatible with 16MHz Arduino boards. The ArduIMU+ v3 uses sensors similar to those of the FreeIMU but doesn't provide a barometer. Arguably the Razor IMU and ArduIMU+ v3 may be considered easier to use since they provide sensors and CPU on the same board while the FreeIMU needs to be connected to an host Arduino board: we believe however that this is not a limitation when prototyping HCI devices since in the prototype there will probably be other sensors or input devices requiring additional connections, thus the 4 connections required by the FreeIMU board shouldn't be a problem.", "From a software point of view, the FreeIMU library provides a much richer and polished API compared to those offered by 3rd parties on the Razor IMU or ArduIMU+ v3. The FreeIMU library is also the only one offering altitude estimation and a calibration GUI. While we don't currently have tests comparing the FreeIMU library orientation algorithm performance with the software for the Razor IMU or ArduIMU+ v3, according to some users “the FreeIMU implementation smoother, less noisy, and more responsive than the DCM based ArduIMU code”http://diydrones.com/forum/topics/freeimu-firmware-on-arduimu: this is the reason why some many Razor IMU or ArduIMU+ v3 users ended up using or porting the FreeIMU library to their hardware instead of using their board software.", "Compared to closed source IMU platforms, such as the Vectornav VN-100 and X-IO X-IMU, while we don't currently have test comparing the orientation sensing of our platform with those, we believe that the FreeIMU project can still give much greater possibilities then using a commercial platform. By being completely open, both from an hardware and software point of view, it can be freely implemented in 3rd parties derivative projects or products, something not possible when using closed source IMU platforms.", "The FreeIMU is also the only IMU project we know supported by a very active community of users, organized in the FreeIMU community websitehttp://freeimu.varesano.net/, sharing projects, code improvements and helping each other." ]

[]

[ "The notion of reciprocation is well studied in sociology, and, as defined by Gouldner {{cite:d9379c36c1650a1d8a47fb8b901472bbefbbfc2e}}, it is the norm of reciprocation that people should help those who help them. Indirectly, if you have helped someone then they have an obligation to help you back. Following this notion, several studies analyze the relationship between reciprocation and the human behavior in various social networks, such as child care {{cite:fd3a0831c0e15b41ecc1d06d4548d3a6e1e4ab0c}}, social altruism {{cite:fdbf61454880fe9d489f3eb8a61ef1ce02c5d1a3}}, and online communication {{cite:6fe12322f212a326f7344ecd666f9b7d8aa19983}}.Leider et al. {{cite:fdbf61454880fe9d489f3eb8a61ef1ce02c5d1a3}} use reciprocation as a means to study human behavior in social altruism in terms of favors and gifting. They conclude, by several online field experiments, that there is a 52% increase in directed altruism towards friends compared to random strangers. This number further increases by another 24% when it leads to further prospects of receiving return favors or gifts. Similar experiments were conducted by interviewing families from different working groups to understand the child care favors in social networks. Hansen {{cite:fd3a0831c0e15b41ecc1d06d4548d3a6e1e4ab0c}} reveals that social location and kinship improves reciprocation of interaction in these networks. In another recent work {{cite:6fe12322f212a326f7344ecd666f9b7d8aa19983}}, the Twitter reply network is analyzed to understand reciprocation in relation to user happiness, where the measure of happiness is computed using sentiment analysis of text content exchanged in Twitter messages. This work concludes that reciprocal behavior in the Twitter reply network shows high assortativity in terms of measured happiness of Twitter users. In all these works, reciprocation is studied on hand picked user samples or surveys {{cite:fd3a0831c0e15b41ecc1d06d4548d3a6e1e4ab0c}}, {{cite:fdbf61454880fe9d489f3eb8a61ef1ce02c5d1a3}} or used to understand specific human behavioral aspect such as happiness {{cite:6fe12322f212a326f7344ecd666f9b7d8aa19983}} or altruism {{cite:fdbf61454880fe9d489f3eb8a61ef1ce02c5d1a3}}.", "There has been recent developments on developing graph models using reciprocation as a model parameter. For instance, Zlatic and Stefancic {{cite:d060ded7c5d6206762c9c523425f881345804bd1}} develops a growing model for Wikipedia using three key parameters: reciprocal edges, degree, and size of modeled network. They empirically evaluate that the generated network closely matches the in- and out-degree distribution of the actual Wikipedia. There are also several papers that discuss the relationship between reciprocation and other network parameters such as degree distribution. Durak et al. {{cite:754c31c32ca11df6db429100292c0eb442ace24a}} propose a null model for directed graphs that combines the reciprocal and the one-way edges to generate directed graphs that match the in-, out-, and reciprocal-degree distributions. Ahmad et al. {{cite:ce2d6bc7021808682b2f11348b65edf3067d7d63}} use theories of social exchange as a basis for building a generative model GTPA for modeling temporally evolving directed networks.", "There are other recent works, such as Gralaschelli and Loffredo {{cite:72bfa3c14d7892abf153b94d79ce59e2782941dc}} that use statistical measures to conclude that the reciprocation of a network is never at random and it is either reciprocal or anti-reciprocal. They define a correlation coefficient in terms of the average and the actual reciprocation. When the correlation coefficient is positive and nonzero, it is correlated, and when negative, it is anti-correlated. Similarly, Zamora-Lopez et al. {{cite:5a7a6fbe9380af6970227a8f58d0b0a4a06af118}} propose a method to compute the expected reciprocation of the network as a function of in- and out-degree distributions and show empirically the expected reciprocation closely matches the actual in several real-life networks. reciprocation in weighted directed networks, especially in large scale mobile communication networks, is discussed in {{cite:d4b97161cd5e48d2c94cea77fce96266b9612a75}}, {{cite:da557816aa21c678f05a2fb96cdf974cc9c3c2ae}}, . Akoglu et al. {{cite:d4b97161cd5e48d2c94cea77fce96266b9612a75}} propose a novel triple power law (3PL) distribution that fits the reciprocation behavior in several data sets and show interesting properties of this distribution such as parsimony. The paper also concludes that reciprocation is higher among users with more common neighbors and larger degree correlation. Szell and Thurner {{cite:c2e2b0999b5a85e6a76c9c7b1a2ecb7f8e149aa8}} study the strength of positive and negative ties in friend and enemy networks and propose approximate social laws relating betweenness centrality and communication strength to the number of overlapping links. In another work Szell et al. {{cite:c531f982c8118370fe46728dda0fc64f082a1b95}} show that negative interactions have a lower reciprocation compared to the positive interactions.", "Reciprocation has also been used for link prediction {{cite:2a53247426ffdb57c003263ffd0ff04de99b1047}}, {{cite:b4ddfaee76eaee96d48b2597ea128a0496286e06}}, {{cite:7cc6a7381e36954b198d43843ea9be413b2635cf}}. Cheng et al. {{cite:2a53247426ffdb57c003263ffd0ff04de99b1047}} study several attributes for predicting reciprocation in the Twitter network, such as relative degree, absolute degree, and other link prediction features. They use regression and decision tree classifiers to predict the expected reciprocation and conclude that features that measure the relative status of two nodes are the best predictors of reciprocation. Zlatic and Stefancic {{cite:b4ddfaee76eaee96d48b2597ea128a0496286e06}} study how reciprocal edges influence other properties of complex networks, such as degree distribution and correlations. They present a statistical inference technique to estimate the number of reciprocal edges.", "In summary, all these related works address the following class of problems: modeling a generative process incorporating reciprocal edges, understanding the effect of reciprocation on a particular human behavior, or using reciprocation-based features in the task of link prediction. None of these papers attempt to study multiple relational networks such as trade, chat, etc. to understand the notion of reciprocation in trust relationship in heterogeneous networks, especially on large-scale real-life networks such as MMOGs. Also, these models do not use an evolving network scenario, where understanding the behavior of reciprocation on the time dimension is extremely important." ]

[]

[ "Index tuning. There has been a long line of research studies on the problem of tuningthe index configuration of a single DBMS (e.g., {{cite:9f9641aad1d8851c1236d4b0977b30984c2f49bd}}, {{cite:3fca6ff82318306624999451fddbb309f6dccb8a}}, {{cite:800d5aa2a106f25fe4caa3f1df4938153e4e6085}}).These methodsanalyze a representative workload and recommend an index configurationthat optimizes the evaluation of the workload according to the optimizer'sestimates.A recent study {{cite:3fca6ff82318306624999451fddbb309f6dccb8a}} has introducedthe COPHY index advisor thatoutperforms state-of-the-art commercial and research techniquesby up to an order of magnitudein terms of both solution quality and total execution time.Both RITA and COPHY leverage the same underlying principle of linear composability, which we will define and discuss extensively in Section REF ,in order to cast the index-tuning problem as a compact,efficiently-solvable Binary Integer Program (BIP).However, COPHY targets the conventional index-tuning problemwhere the goal is to compute a single index configuration fora single-node system.This problem scenario is much simpler than what we consider in our work, where there are several nodes in the system, each can carry a different index configuration, queries have to be distributed in a balanced fashion and the system must recover gracefully from failures. Leveraging the principle of linear composability in this generalized problem scenario is one of the key contributions of our work.", "Physical data organization on replicas. Previous works also considered the idea of diverging the physical organization of replicated data.The technique of Fractured Mirrors {{cite:45f531c876978b27f6beb48e012a027912ca8879}} buildsa mirrored database that stores its base data ina different physical organizations on disk(specifically, in a row-based and a column-based organization).Similarly, Distorted Mirrors {{cite:04179ac558375ed860276419544e6c0a17b94f56}} presents logicallybut not physically identical mirror disks for replicated data.However, they do not consider how to tune the indexes for each mirror.", "There are recent works {{cite:03aede024792291df2b6d360d74ae0d9173e313e}}, {{cite:d2044f912047c09f89fd4ac5f02c60d619c83f2c}}that explore different physical designs for different replicas inHadoop context.Specifically,TROJAN HDFS {{cite:03aede024792291df2b6d360d74ae0d9173e313e}} organizeseach physical replica of an HDFS blockin a different data layout, whereeach data layout corresponds to a different vertical partitioning.Likewise, HAIL {{cite:d2044f912047c09f89fd4ac5f02c60d619c83f2c}}organizes each physical replica of an HDFS blockin a different sorted order, which essentially amounts to exactly one clustered index per replica.Our work differs from these works in several aspects.First, these worksdo not consider the problem of spreading the load evenly among specialized replicasthat we are considering.Second,performance is very sensitive to failures, because the tuning options considered bythese papers lead to replicas that are highly specialized for subsets of the workload.When a replica fails, the corresponding queries will be rerouted to replicas withlittle provision to handle the rerouted workload,and hence performance may suffer.In contrast, we focus on divergent designsthat directly take into account the possibility of replicas failing,thus offering more stable performance when one or more replicas become unavailable.Third, {{cite:d2044f912047c09f89fd4ac5f02c60d619c83f2c}}creates one index per replica which restricts the extentto which we can tune each replica to the workload.Our methods do not have any such built-in limitations and areonly restricted by configurable constraints on the materialized indexes(e.g., total space consumption, or total maintenance cost).Fourth, our work can returna set of possible designs that representtrade-off points within a multi-dimensional space,e.g., between workload evaluation cost and design-materialization cost.These works do not support this functionality.", "The original study on divergent-design tuning {{cite:ef19a023c5b898db2df3b9de0b102107b0e47ab6}}introduced the {{formula:c7b5d585-ee5c-4cd2-bbbc-fc27e0074b21}} advisorwhich is the direct competitor to our proposed RITA advisor.However, as we discussed in Section ,{{formula:c12329e3-7ecc-4760-bb02-cb8944ebe333}} is fundamentally limited by the functionality of theunderlying single-system advisor,and cannot supportmany essentialtuning functionalitiesas RITA." ]

[]

[ "Straight-line programs, or equivalently algebraic circuits,are important both as a computational model and as a data structurefor polynomial computation. Their rich history includes bothalgorithmic advances and practical implementations{{cite:4447270ca84f860fb250c2a70e904ca81f4817fd}}, {{cite:e179abf44fa52a8a3a40d11e5e182028e5dea6ab}}, {{cite:9781b178c617a72b9a23a1287fee3c5a7d6636fe}}.", "One can naively interpolate a polynomial {{formula:f72cfa51-3043-4c47-bc8b-7afd739acfa2}} given by a straight-lineprogramusing a dense method, with {{formula:ec4a40d0-b843-4330-9e3b-5e59a81d36ea}} probes of degree 1. Prony's{{cite:b110dbf08ff3059b0fd5f7cb885322dde448b2ae}} interpolation algorithm — see {{cite:fd0bebf6505b50617515aee884b86d0c5e86849d}}, {{cite:9f79ffb0b428ad73075e6df501503aae263d0970}}, {{cite:e38204c1592cb52b093b095beeb7e2c23f5bc17c}} — is a sparse interpolation method that usesevaluations at only {{formula:b3e422c9-70cf-4e03-9bfa-87acfba2e632}} powers of a root of unity whose order isgreater than {{formula:f666ae02-7d6f-413e-a218-ccc82a2ab1b2}} . However, in the straight-line program model for ageneral ring, this would require evaluating at a symbolic {{formula:932e79ab-5728-4bbe-b446-96f708f83656}} th rootof unity, which would use at least {{formula:acbdb395-bf7a-4d65-9e33-87a7940c6bff}} ring operations anddefeat the benefit of sparsity. Problems with Prony's algorithm arealso seen in the classical model in that the underlying base ring {{formula:658de7fa-9463-4e6e-a0aa-dddbfc7e3073}} must also support an efficient discrete logarithm algorithm on entriesof high multiplicative order (which, for example, is not feasibleover large finite fields).", "We mention two algorithms specifically intended for straight-lineprograms." ]

[]

[ "The first approximation for conductance was obtained by discreteanalogues of the Cheeger inequality shownby Alon-Milman {{cite:139efb2afc637a21486cbe6906763277e15f697e}} and Alon {{cite:3504fd6e17b30ea4fa21af2ea32af7ce615ae0a3}}.Specifically, Cheeger's inequality relates the conductance {{formula:6be118a1-799d-4ef7-9b25-11bcc7b4f596}} to the second eigenvalue of the adjacency matrixof the graph – an efficiently computable quantity. This yields an approximation algorithm for {{formula:c831d685-eda8-497e-8d16-484f35caf0bc}} , one that is used heavily in practice for graph partitioning.However, the approximation for {{formula:f90e7da1-593b-46a7-8b0f-e65c2887ff4f}} obtained via Cheeger'sinequality is poor in termsof a approximation ratio, especially when the value of {{formula:409ca38e-3272-4c7f-aa04-3607aee84616}} is small.An {{formula:d04e32a8-d3ee-45c2-833e-3c1e644c8a98}} approximation algorithm for {{formula:aefa00ab-1e14-4542-8d8c-db2df582b955}} was obtained byLeighton and Rao {{cite:db3f48391892490f1944138258086e440c444029}}. Later work by Linial et al. {{cite:1d1fe94ae4ade31704741a1a84631d2a4231a706}} and Aumann and Rabani {{cite:30ece8fcb9e3f6f6a09f7ae3a3b69c7594009e39}} established a strong connection between theSparsest Cut problem and the theory of metric spaces, in turnspurring a large and rich body of literature. The current bestalgorithm for the problem is an {{formula:013322b7-e141-472d-b685-bad4bb8a16ea}} approximation fordue to Arora et al. {{cite:936cc62d4720f3b3536888986fe321c0e7017ee3}} usingsemidefinite programming techniques.", "Ambühl, Mastrolilli and Svensson {{cite:ce5d2a48547aefa2aed8b22da969d07d086ba081}} showed that {{formula:0a70e2b8-c059-4c34-87c2-4844cbb440c6}} and {{formula:4620b090-db30-4975-a3a2-3ffdcfe64893}} have no PTAS assuming that SAT does not havesub-exponential time algorithms.The current best approximation factor for {{formula:9d9a71d2-89b6-4c5f-814d-19fcbe2a17ab}} is {{formula:cdf570b6-b685-4e3d-a8c0-4697932b3245}} obtained using a convex relaxation {{cite:b7eb4422caa925b6721f1520257a3df13ebea435}}.Beyond this, the situation is much less clear for the approximability of vertex expansion. Applying Cheeger's method leads to a boundof {{formula:7ee0afe6-1ebd-4439-905b-be9cdeda54cb}} {{cite:3504fd6e17b30ea4fa21af2ea32af7ce615ae0a3}} where {{formula:05cbd425-14d8-43db-a806-9005c3d274c5}} is the maximum degree of the input graph." ]

[]

[ "A class of studies has focussed on enumerating the characteristicsof wired networks {{cite:9651bf7b1fe8f1de2fb14c28b62ccd45c4d8f401}}, {{cite:f7bb540636b0c1b466ecc480046f6e2e0331f41d}} that need to betaken into consideration while generating topologies from randomgraphs. Consequently, a parallel area of research is focussed onefficient generation {{cite:0e2ccb44e37e82e029464d75209d8263316b3df0}} and improvement of thefeatures of random graphs to model real wirednetworks {{cite:d9859fa6591b400eb373b8faa52f0976ea1911a2}}.", "With concerns to wireless networks,  {{cite:dcea5ffa768fa7b8a0373a21ca5ff75ac5d84f91}}investigates the impact of spatial distribution of nodes on theminimum node degree, and the k-connectivity in random networkgraphs. We take a completely opposite view of the problem indetermining if a randomly generated adjacency could be used forfaithfully representing a realistic wireless topology." ]

[]

[ "Current approaches to protect privacy in recommendation systems mostly address two different privacy concerns: protecting users' privacy from curious peers or malicious users {{cite:471002021a0f3e42410fa0dffa1d14a8966fc679}}, {{cite:b36870d1dd261669128a7ab0ee857afd2c82003e}}, and against unreliable service providers {{cite:9c3b349c9a662ecb250879c6857783ac69e0e7ab}}, {{cite:c35890d052aea532e054cd28f1ce6f2032c9da42}}, {{cite:03d52741d9e6bcfb67b6bb9376d037036f802668}}. In order to make the outcome of recommendation insensitive to single input so as to protect users private preference data from other users, privacy preserving algorithms from the differential privacy literature are modified to provide privacy guarantees. McSherry et al. {{cite:b36870d1dd261669128a7ab0ee857afd2c82003e}} adapted the leading approaches in the Netflix Prize competition to provide differential privacy and recommendations on movies. Machanavajjhala et al. {{cite:471002021a0f3e42410fa0dffa1d14a8966fc679}} studied recommendations based on a user's social network with differential privacy constraints. On the other side, in order to prevent a single party, e.g. the service provider, from gaining access to every user's data, cryptographic solutions are proposed in {{cite:9c3b349c9a662ecb250879c6857783ac69e0e7ab}}, {{cite:c35890d052aea532e054cd28f1ce6f2032c9da42}}, however, cryptography could be computationally expensive, especially for end-users. Nandi et al. {{cite:03d52741d9e6bcfb67b6bb9376d037036f802668}} proposed to preserve preference privacy from a single party by middleware, where computation and recommendation are performed locally.", "The focus of our work is to protect users from unreliable service providers, and to mitigate users' fear of potential intrusions of privacy by keeping a certain amount of anonymity. The curse of dimensionality and computational limitations of personal devices make deployment of {{cite:03d52741d9e6bcfb67b6bb9376d037036f802668}} difficult. The idea of using groups as a natural protective mechanism is inspired by the French philosopher Gilbert Simondon {{cite:815da082c98dbcc2c4c457d119cf187ff0f61fc0}}. An intriguing and interesting aspect of Simondon's theory of systems and technical objects is the idea of adopting an “associated milieu\" into the operation of the system. This associated milieu can be natural resources. For example, Simondon spoke of the Guimbal turbine (named after the engineer who invented it), which, to solve the problem of loss of energy and overheating, used oil to lubricate the engine and at the same time isolate it from water; it can then also integrate a river as the cooling agent of a turbine {{cite:815da082c98dbcc2c4c457d119cf187ff0f61fc0}}. The river here is the associated milieu for the technical system; it is part of the system rather than simply the environment. Groups for us serve a similar function as an associated milieu, that contribute to the preservation of individual privacy, while still supporting the functioning of the social network.", "In this paper, we propose a framework for using groups as a natural middleware to recommend products to users. Our framework can be combined with other differentially private recommendation solutions such as {{cite:b36870d1dd261669128a7ab0ee857afd2c82003e}}. More specifically, we design a simple distributed protocol to preserve users' privacy through a peer-to-peer preference exchange process. The effective size of the anonymity set asymptotically approaches the size of the group as time approaches infinity. After group opinion is aggregated, we construct a recommendation graph and use a random walk based method to make recommendations. The stable distribution resulting from a random walk on the graph is interpreted as a ranking of nodes for the purpose of prediction and recommendation. Personalized recommendation is only performed locally so that no private information is revealed to the service provider. We evaluate the performance of the proposed algorithm using the MovieLens and Epinions {{cite:94af87a0caef49cc434a42d50c1fbd606249118b}} dataset, and we compare the results with recommendation algorithms designed for individual users." ]

[]

[ "A great many papers in the compressive sensing literature discuss how to implement practical compressed sensing systems (e.g., see {{cite:3c39776d5461bd6a6b2564db53e9f79a6571f36b}}, {{cite:7f401a6dd97aa8d3cc2c6eec22b36b416c506810}}, {{cite:6a6d3eef7bbebdd2f44bda12566f71d0cdff9379}}) but few show CS imaging to be better than conventional imaging when all aspects of the two approaches are weighed and compared.Most practical CS hardware architectures can be found described on Igor Carron's website at https://sites.google.com/site/igorcarron2/compressedsensinghardwareand the latest information within the CS field may be discovered in his daily blog at http://nuit-blanche.blogspot.com/.", "As discussed above, two credible applications for CS are in its use for MRI{{cite:7bf50de3d8f33f2fd249c5345fbff58a49359f70}} and radio interferometry{{cite:91e732dd05fed1d215d838c96cf2dcda14ced3eb}}.We used these wins as prototypes for establishing the guidelines for successful application of CS.There are a number of papers in the literature applying CS to detection of stars, such as the paper by Gupta, et. al.{{cite:738c0f8ec67a585989bb529b0a20f1aae0641df7}}.Missile detection considered later in this paper bears similarities to star point source detections algorithms.This also bears some resemblance to the application of compressive sampling to radio interferometry{{cite:91e732dd05fed1d215d838c96cf2dcda14ced3eb}} where the use of partial Fourier reconstruction is described.", "The JASON report on \"Compressive sensing for DoD sensor systems\"{{cite:9f3e4b0325e800d6d7eb137caa2eb8332d30e804}} is closely related to the topic of this article as the report addresses the question where CS can be beneficially utilized for DoD radar applications.Some of the guidelines for successful application of CS can be found in this report but are not clearly delineated as in this paper.Ultimately the report finds that CS should be of interest to DoD but more research is required, such as this paper, to determine useful applications.", "The paper “Compressive sampling vs. conventional imaging” by Haupt and Nowark{{cite:b2f094d9873ded0494301e54b0ad99c945f44d16}} addresses the same question that is the focus of this paper.However, that paper does not provide general guidelines nor an example of a new CS application as is done in this paper.Several recent papers, such as one by Adcock, et. al.{{cite:14a7db16f19220cc61c75e5bde570c04e18cc926}}, point out the gap between CS theory and use in real world applications and show how to bridge it.", "There are in the CS literature papers related to using compressive sampling for target detection and tracking.The paper by Kashter, Levi, and Stern{{cite:80ee05689873f2fffffc2a6a75bf8d44d18ba490}} describe application of CS to motion detection.However, their method relies on frame differencing to achieve sparsity; that is, only the moving objects appear in the frame differenced image.Similarly the paper by Poon, et. al.{{cite:f95074a2cc0de99d599dc9717cf0bff9c987544b}} describes a realizable hardware application that they applied to a simplified target tracking problem.Again their method depends on frame differencing, hence the objects moving to different pixels from frame to frame.In the missile detection problem, the sub-pixel missile signal remains in the same pixel for tens of frames and frame differencing eliminates all possibility of detecting the already dim target.A paper by Duarte, et. al.{{cite:ed6e19a6cd177742f4752e000ef7c427623add8f}} describes the signal detection problem using CS principles without reconstructing the image.The missile detection application described here similarly follows this concept; that is, the goal is to detect the missile regardless of the quality of the image reconstruction." ]

[]

[ "The use of modularity maximization for community finding was first introduced in {{cite:d1f6b267cca3690742d8df871527617fec394a9e}} which also described how to find partitions that could nearly maximize this quantity. An exact method for addressing this optimization problem was introduced in {{cite:68a8b43dfffa1afc973ce64ae5d3c4ddb3407a7b}}. However, this method was based on integer programming and for many problem instances may take an exponential amount of time to complete. However, we note that an easy modification of that program can be used to address the problem of this paper as the quantity {{formula:63c1f2b4-4cc6-4876-b516-632151722e13}} can be solved in a pre-processing step and treated as a constant in the integer program formulation. Note that the time to complete such a step would be easily dominated by the overall runtime to even approximate a solution in such a method. In the same paper, modularity maximization was also shown to be NP-hard, which precludes an efficient approaches under current theoretical assumptions. In {{cite:60cdded6b1ba0027dab10b8e7d98f982550ec53d}} the Louvain algorithm is introduced which is shown to provide partitions that nearly maximize modularity and can scale to very large networks. The modification of the Louvain algorithm is what we leveraged in this paper.", "Modularity was extended to consider geosptial relationships using a distance-decay model in {{cite:3c89dc238ed376d89772059190571264de678632}} with the introduction of distance modularity which we use in this paper. Their approach modifies the null model to increase the expected number of edges between close nodes, it will tend to find communities that are more geographically disperse - hence meeting the requirement of our presented application. Our work extends on their theory - providing an algorithm to find an approximately optimal partitions wrt distance modularity, experimental results, and describes practical considerations - none of which were included in {{cite:3c89dc238ed376d89772059190571264de678632}} which only introduces the the concept of distance modularity and describes the mathematical properties of their alternative null model.", "The recent work of {{cite:4ca921be69a6a53b417bded4322cd23b4a3634ad}} introduces “spatial modularity” that also uses a distance-decay function in the null model - though somewhat different to that introduced in {{cite:3c89dc238ed376d89772059190571264de678632}}. They study the difference among partitions created by attempting to optimize both standard modularity and their alternate definition on a series of small simulated networks whose edges are formed based on varying degrees of correlation between space and node similarity (determined by randomly assigned attributes). The results of that paper have also inspired this work as they indicate that by considering geospatial relationships in the null model often yields different community structure than with the original definition of modularity introduced by {{cite:d1f6b267cca3690742d8df871527617fec394a9e}}. However, unlike this paper, the work of {{cite:4ca921be69a6a53b417bded4322cd23b4a3634ad}} only studies simulated networks (this paper only looks at real-world networks). The networks of this paper are an order of magnitude larger as {{cite:4ca921be69a6a53b417bded4322cd23b4a3634ad}} only considers networks of 100 nodes. Further, {{cite:4ca921be69a6a53b417bded4322cd23b4a3634ad}} does not describe any practical concerns in their approach that must be considered when creating a real-world system.", "Another important result on community finding in geosptial networks was that of {{cite:58d884ec1a0907f8f53f2920fa65fb7e493a4d23}} where the authors also modify modularity. However, in that work, the authors use a null model that is based on an empirically observed probability distribution of edge existence based on distance. Their optimization approach was tested on a network of Belgian communes of phone users and was shown to accurately identify linguistic communities. However, unlike this paper and the work of {{cite:3c89dc238ed376d89772059190571264de678632}} and {{cite:4ca921be69a6a53b417bded4322cd23b4a3634ad}}, as their null model is based on an empirically determined probability distribution, it will not necessarily ensure geographically-disperse communities - which is our target application. Further, the work of {{cite:58d884ec1a0907f8f53f2920fa65fb7e493a4d23}} does not describe practical considerations and their experimental evaluation is restricted to the Belgian phone network data consisting of 571 nodes.", "In addition to the aforementioned approaches, community detection in networks has also been explored in other manners that could potentially be proved applicable to geospatial applications - though to our knowledge no such application has been presented in the literature. For instance, the work of {{cite:ba39ea4edeba06763c1bb9fa6b529f7cc8cdb71c}} identifies communities based on both network topology and content analysis. Further, there are methods for community detection other than modularity maximization on networks (that do not consider spatial interactions). Leveraging one of these other approaches is an important direction for future work. See {{cite:8c64df85fe7ec25b387af757a7b93d7f09003e79}} for a comprehensive survey.", "There has been other recent work where geospatial networks have been explored with respect to problems other than community finding. The work of {{cite:7d68d30275826359c5c994e51e2640ee96748844}} discusses link-prediction and shows that by considering geography that results for this problem can be improved. The work of {{cite:725721d3c2d39187db79c8edb910080c997e8aec}} looks at identifying the location of users on Twitter using network topology. Further, there also have been empirical studies on social networks with a spatial component such as {{cite:cdac5e376c18d9ee17f8519b461a0a76dd11fd97}}. Along such lines, the mobility of users in a location-based social network is explored in {{cite:adc0355461a6085ea7fad11151227cc2d11160f7}}, {{cite:800e549a6897262974b5eb6332f5debca648bc41}}. More domain-specific empirical studies related to this work are also prevalent in the literature. Pertinent to our application include studies on terrorist networks {{cite:41e5c55c6dafef3009ec009f3b38d57f006dadab}} and criminal co-offender networks {{cite:4babd28152976ce89060fa2faf968381f82395ad}}." ]

[]

[ "Several algorithms for symbolic model checking of PLCS have been presented{{cite:2aa12f36d0d6cab1791c620d426d2fbee83a1f63}}, {{cite:0073960d2b55b92e9da199bc8bcbaa8e03f44a25}}.Markov decision processes (i.e., {{formula:52b586ab-e3d0-4814-8058-32269937ae95}} -player games)on infinitegraphs induced by PLCS were studied in {{cite:8e10cc760c603ecadfdc1d010344a9a5696927b6}},which showsthat {{formula:063d4234-b644-4a5e-8227-17967bb1c3c6}} -player games with almost-sureBüchi objectives are pure memoryless determined and decidable.This result was later generalized to {{formula:cafc5c0f-c388-4010-bc68-79f7bcd4aa38}} -player games{{cite:2fa976abc646877d0bfdbc63958f25a4be612500}},and further extendedto generalized Büchi objectives {{cite:3ddc21bfb5864394da0a5ea8f8ba8fcf1ed390db}}.On the other hand, {{formula:9141b219-96ed-45d7-bea6-9ef6068f8ad8}} -player games onPLCS with positive probabilityBüchi objectives (i.e., almost-sure co-Büchi objectives from the(here passive) opponent's point of view)can require infinite memory to winand are also undecidable {{cite:8e10cc760c603ecadfdc1d010344a9a5696927b6}}.(Undecidability and infinite memory requirementare separateresults, since decidability doesnot imply the existence offinite-memory strategies in infinite-state games).If players are restricted to finite-memory strategies,the {{formula:270fc0ff-2dc1-4914-8b0c-4fbff916eeb3}} -player game with positive probabilityparity objectives(even the more general Streett objectives)becomes decidable {{cite:8e10cc760c603ecadfdc1d010344a9a5696927b6}}.Note that the finite-memory case and the infinite-memory one are a priori incomparable problems,and neither subsumes the other.Cf. Section .", "Non-stochastic (2-player) parity games on infinite graphs were studiedin {{cite:ef70c26d818c5d2cead320d530b7e624e1e4c528}}, where it is shown that such games aredetermined, and that both players possess winning memoryless strategies intheir respective winning sets. Furthermore,a scheme for computing the winningsets and winning strategies is given.Stochastic games ({{formula:745104fd-e40e-4c1c-aca8-3285201104e9}} -player games) with parity conditions onfinite graphs are known to be memoryless determined andeffectively solvable{{cite:3d086a37c0fb4fbca86dd8526a619c9b65fd03c9}}, {{cite:63009f2226954626a8844f23569e679501eea75b}}, {{cite:fa8ca09040758909b5c8f26494439e3a2f45f301}}." ]

[]

[ "MathML {{cite:ecc989ad135301a2ee65f2a545ba4db2f9dd6585}} is the best-known open markup format for representing mathematical formulas.It is recommended by the W3C Math Working Group as a standard to represent mathematical expressions.MathML is an application of XML for describing mathematical notations and encoding mathematical content within a text format.MathML has two types of encoding: Content MathML addresses the meaning of formulas; and Presentation MathML addresses the display of formulas.We use MathML Presentation Markup to display mathematical expressions and MathML Content Markup to convey mathematical meaning.", "Most major computer algebra systems, such as Mathematica {{cite:4b16f1c4ae3ffd21858e922a5ac0478e238458cb}} and Maple {{cite:24ca29651c4633b4f806641a8654ed6fbea6cc1b}}, are capable of importing and exporting MathML of both formats.These importing and exporting functions enable the conversion from Presentation to Content MathML.Importing, of course, depends on the interpretation of each computer algebra systems engine.", "There is a project called SnuggleTeX {{cite:e23bcb92b5720071d04d866a50162a6ddbf712ca}}, which addresses the semantic interpretation of mathematical expressions.The project provides a direct way to generate Content MathML from Presentation MathML based on manually encoded rules.The current version at the time of writing this paper supports operators that are the same as ASCIIMathML {{cite:97a441e1a66b55867ecd4ee56efb5236c8cbc2f0}}.For example, it uses the ASCII string “\\{{formula:b163c4ef-3609-4248-b471-9af82a2937d1}} ” instead of the symbol “{{formula:0eeb4fdc-23eb-4b81-87d0-07719679ab15}} ”.One major drawback of this approach is that it always makes the same interpretation for the same Presentation MathML element.", "A recent study by Nghiem et al. {{cite:a07b2b4d902e87cbc650003ed12f57bab0e2196d}} also addressed the semantic interpretation of mathematical expressions.This study applied a method based on statistical machine translation to extract translation rules automatically.This approach contrasted with previous research, which tended to rely on manually encoded rules.This study also introduced segmentation rules used to segment mathematical expressions.Combining segmentation rules and translation rules strengthened the translation system and the best system achieved 20.89% error rate.The shortcoming of this approach is that it did not make use of text information surrounding mathematical expressions.", "Wolska et al. {{cite:376537a4644982fe0c2632aac0e61b6ac50947b2}}, {{cite:71dcf25fe1977e62a6940a62885b8c3e617704ae}} presented a knowledge-poor method of finding a denotation of simple symbolic expressions in mathematical discourse.The system used statistical co-occurrence measures to classify a simple symbolic expression into one of seven predefined concepts.They showed that the lexical information from the linguistic context immediately surrounding the expression improved the results.The lexical information from the larger document context also contributed to the best interpretation results.This approach had been evaluated on a gold standard manually annotated by experts, achieving 66% precision." ]

[]

[ "Two main approaches have been used in evaluating graph drawings: 1) approaches based on computational measures such as commonly used aesthetics and specifically developed measures; and 2) approaches based on empirical measures such as expert opinion, user preference and task performance. In this section, selected studies are given as examples.", "Didimo et al. {{cite:1919659edcdbae65a486da3d88923df146474a33}} conducted a study that evaluated different graph drawing algorithms. The quality of the resultant drawings was compared based on the extent to which they conformed to each of a set of aesthetics. The aesthetics used for the comparison purpose included number of crossings, crossing angle resolution, geodesic edge tendency and vertex angles. The results indicated that some algorithms had a better trade-off between these criteria than other algorithms.", "Brandes and Pich {{cite:1259a41d13f57dd04dde23bffd019af4433601cb}} compared distance-based graph drawing algorithms. In this study, the drawing quality was measured as how well the Euclidean distance between any two nodes represented their graph-theoretic distance. The results suggested that minimization of weighted stress yielded better layout than force-directed placement in terms of pairwise distances between nodes. Similar quality measurements were also developed in other studies to evaluate whether drawings have specific layout features (e.g., {{cite:bfa566b181c1759ed582f58b9aa75a780f6eb49d}}).", "While these computational approaches evaluate quality of specific visual properties of drawings directly, empirical methods have been used to evaluate overall quality. Huang et al. {{cite:8110dc47a816e4a0cdc50a651472dba350509dd2}} conducted a study that evaluated quality of sociograms that were drawn based on different drawing conventions and edge crossings. Subjects were asked to perform network-related tasks and indicate their preference for drawings conventions and crossings. Task performance including response times and error rates were also recorded. One of their findings was that a drawing that was the most preferred might not be the one that induced the best performance. Other studies in this category include  {{cite:c8cd3bb1767a8b6226b980e58a90c4f79edc488c}}, {{cite:e1368e0e9c3e031a1dfbd14f391f4a7214e735ae}}, {{cite:754a32e001f096abdefe53874f209c9fb570aee3}}. Evaluating quality based on expert opinion, Hachul and Junger {{cite:05f19a0453da06f1963edc1fa67d62f33c6195d6}} conducted a study that compared algorithms for drawing general large graphs. Rather than using a quantitative measurement, the authors evaluated the drawing quality by commenting how well the layout displayed the graph structure.", "The above-mentioned approaches are useful at certain circumstances of a visualization process and for specific purposes. However, a direct, empirically validated measure is more preferable for a quick assessment of overall quality." ]

[]

[ "Two special cases of the general {{formula:5cc03402-adaf-4eb7-8c29-204352f7bea7}} -version arethe abstract version in which no counters are defined, and theconcrete version in which all counters are defined.In the energy semantics, the abstract version corresponds tothe unknown initial credit problem for multidimensionalenergy parity games, which is coNP-complete {{cite:e6eac7263a28845cf8569be647a145e83884f384}}, {{cite:bd34986aa7b7a33c0bfdd12eb0f03914fb4a2dfe}}.The concrete version corresponds to the fixed initial credit problem.For energy games without the parity condition, the fixed initialcredit problem was solved in {{cite:5b81f0f69e39d3d7f03d001b47ecaa2257c3b82e}}(although it does not explicitly mention energy games butinstead formulates the problem as a zero-reachabilityobjective for Player 1).It follows from {{cite:5b81f0f69e39d3d7f03d001b47ecaa2257c3b82e}} that thefixed initial credit problem for {{formula:6ca8db00-baea-4db1-aa9a-cf484f0cabe3}} -dimensional energy gamescan be solved in {{formula:c647a0b5-6f13-40bf-bac9-093a697ea5e0}} -EXPTIME (resp. {{formula:0ead0fe6-3b86-49a6-afe6-a82e83363a11}} -EXPTIME foroffsets encoded in unary) and even the upward-closed winningsets can be computed.An EXPSPACE lower bound is derived by a reduction fromPetri net coverability.The subcase of one-dimensional energy parity games wasconsidered in {{cite:a5e066775f7d07b35a7f0cbed481aa53719afa42}},where both the unknown and fixed initial credit problems are decidable,and the winning sets (i.e., the minimal required initial energy) can be computed.The assumption of having just one dimension isan important restriction that significantly simplifies the problem.This case is solved using an algorithm which isa generalization of the classical algorithms ofMcNaughton {{cite:def2245c44be93d6e209bd617bddf6bd337d9f64}}andZielonka {{cite:325196e3e151292f40b4380c0b853e9b47151fef}}.", "However, for general multidimensionalenergy parity games, computing the winning sets was an open problem,mentioned, e.g., in {{cite:e6eac7263a28845cf8569be647a145e83884f384}}.", "In contrast, under the vass semantics,all these integer game problems are shown to be undecidablefor dimensions {{formula:c1d50fe5-6815-4093-bb5c-92f631346e7e}} in {{cite:c609db96aaba3729acf887bf3e563fb8aff0f194}}, even for simplesafety/coverability objectives. (The one-dimensional case is aspecial case of parity games on one-counter machines, which isPSPACE-complete).A special subcase are single-sided vass games,where just Player 0 can modify counters while Player 1can only change control-states. This restriction makes the winning setfor Player 0 upward-closed, unlike in general vass games.The paper {{cite:7e8721387abcea61ad0027f0dbd952581558ed74}} shows decidabilityof coverability objectives for single-sided vass games,using a standard backward fixpoint computation." ]

[]

[ "There have been several other pieces of software introduced for both general purpose network analysis as well as for law enforcement. The main contribution of the software presented in this paper if the ORCA software. ORCA is that it is tailor made for specific police requirements dealing with violent street gangs. Further, it leverage new articiial intellignece and data mining techniques such as MANCALog and TIP_DECOMP. Although ORCA is the first attempt to combine these techniques in a piece of intelligence analysis software, there have been other efforts to build software designed for social network-based intelligence analysis that we describe below.", "Previous software designed to support law-enforcement through social network analysis include CrimeFighter {{cite:3fbc536e63f31638689b64b9ff16f01dcdd15f57}} and CrimeLink {{cite:d19db1eda55cf476187eeb899fc5b25c118e109a}}. However, these tools provide complementary capabilities to ORCA. While ORCA is primarily designed for intelligence analysis with respect to street gangs, CrimeLink is designed to support investigations while CrimeFighter is directed more toward targeting of individuals in criminal organizations - rather than understanding membership in and linkages between sub-groups. Further, neither of these software packages support degree of membership, seed set identification, or the creation of ecosystems.", "General-purpose social network analysis software also exist, including the Organization Risk Analyzer (ORA) {{cite:45069cd920f2654fe8da954d206b9f9a2168570c}}, and NodeXL (an add-on to Microsoft Excel) {{cite:93574317278c910ed8392b609c09b7d88a527da4}}. While these tools are very powerful and contain many features, they do not provide the degree of membership calculation or the identification of seed sets as ORCA does. ORCA is also purposely designed for police intelligence analysts to better understand street gangs, and features such as community finding in ORCA are designed for with this application in mind (hence the creation of ecosystems and identification of inter-group connectors). This nuanced use of community detection for street gang intelligence analysis is not ready for use “out of the box” in a general-purpose social network analysis package such as ORA or NodeXL." ]

[]

[ "One of the oldest references on mining television content is a DARPA-sponsored workshop in 1999 with a topic detection and tracking challenge {{cite:9f14ad3a6d49854c93d60f2f4ff89caa846b868b}}.Higher-level applications have been emerging in recent years.For example, {{cite:902807a21f891d6c33fc924255739f34117a1e0f}} describe a system for finding web pages related to television content, and test different methods to synthesize a web search query from a television transcript.{{cite:f5d24e867b437ff1e6d6978998d5ea933f9068ec}} classify videos based on a transcription obtained from speech recognition.{{cite:08fd54e2dc0e1f0fa9113d99dbbf892996299205}} describe a system to rate the credibility of information items on television by looking at how often the same image is described in a similar way by more than one news source.", "Information from closed captions has also been combined with other data sources.{{cite:78464cc06bf7734a2076d1cd883011d8ccf1b740}} present a system for video classification based on closed captions as well as content-based attributes of the video.{{cite:e3a289cebb83469fe9e5812f2467c23639d54820}} also combine closed captions and multimedia content to improve video segmentation.{{cite:2f90f05159911a8cae2ebd61832e2254bfceb2f6}} align the closed captions during a live event (a presidential debate) with social media reactions.{{cite:4fea33b483f0c6caf7d03f15258290051414f350}} combine closed captions with real-time television audience measures to detect ads – which typically are accompanied by sudden drops in viewership (“zapping”).", "Quantitative analysis of media.", "Most content analysis of news reports is based on a time-consuming process of manual coding;automatic methods are less frequently used.{{cite:58cc8a6510fd570a4c91715d394d849f55e02bae}} use an indirect measure of television bias by manually counting references to think tanks in each network, and then by scoring each think tank on a left-right political spectrum by automatically counting their appearances in congressional records of speeches by politicians of different political leaning.{{cite:ae998c010efd9356504ce8e352e47050e27b6a8c}} study news articles available on the web, and analyze the prevalence of different topics and the distribution of readability and subjectivity scores.", "Topics and perspectives.The PhD thesis of {{cite:8a02dde7b515ef497b7801390479b58942b6741d}} and related papers (e.g. {{cite:dbd131f6b25ad16a85ed918057d28d5225786890}}), introduce a joint probabilistic model of the topics and the perspectives from which documents and sentences are written. The parameters of the model are learned from a training corpus for which the ideological perspectives are known. In contrast, the methods we study on this paper are based on NLP algorithms.", "Lin et al. process news videos using content-based methods. First, news boundaries are detected automatically using visual clues, such as scene and background changes. Second, each news segment is annotated with concepts such as sky, person, outdoor, etc. which are inferred automatically from shots in the video. This approach is effective for settings in which closed captions are not available." ]

[]

[ "There have been several papers that focus on the generalizationcapability analysis of the {{formula:b8970189-6966-4d5a-a45a-bea934ce3513}} regularization scheme(REF ).Wu and Zhou{{cite:22ea3042ea2f976a114d216e919691376f9088eb}} were the first, to the best of our knowledge, to showa mathematical foundation of learning algorithms in SDHS. Theyclaimed that the data dependent nature of the algorithm leads to anextra error term called hypothesis error, which is essentiallydifferent form regularization schemes with sample independenthypothesis spaces (SIHSs). Based on this, the authors proposed acoefficient-based regularization strategy and conducted atheoretical analysis of the strategy by dividing the generalizationerror into approximation error, sample error and hypothesis error.Following their work, Xiao and Zhou {{cite:76802514e557abe79b80029cd94d98d1a608419b}} derived alearning rate of {{formula:f21c570b-0944-4a69-8756-9aaa9f6907c0}} regularizer via bounding the regularizationerror, sample error and hypothesis error, respectively. Their resultwas improved in {{cite:e0a240818b829cc5814be67a199ea35f23ba597b}} by adopting a concentrationinequality technique with {{formula:3a23e2dd-b172-4b3a-92f3-b82e4264bb21}} empirical covering numbers to tacklethe sample error. On the other hand, for {{formula:233af1a3-33c8-4927-9f0d-49bb5c16be9e}} {{formula:93d1c874-4c37-4d07-b171-9c032fb13dc3}} regularizers, Tong et al. {{cite:ef727df7cec612f980b1d430e6200fe49216c233}} deduced an upper bound forgeneralization error by using a different method to cope with thehypothesis error.Later, the learning rate of {{cite:ef727df7cec612f980b1d430e6200fe49216c233}} wasimproved further in {{cite:725ede6b5c30cd62a830adb71c2b0c2b4f0fee9b}} by giving a sharper estimationof the sample error.", "In all those researches, some sharp restrictions on the probabilitydistributions (priors) have been imposed, say, both spectrumassumption of the regression function and concentration property ofthe marginal distribution should be satisfied.Noting this, for {{formula:335e1218-9ccf-4b2b-931f-ae86d1b057a3}} regularizer,Sun and Wu {{cite:11da512db9b936b814063541a3eeee375d9390c8}} conducted a generalization capabilityanalysis for {{formula:c86c8052-6357-483d-b821-6df5f0191ebe}} regularizer by using the spectrum assumption tothe regression function only. For {{formula:22f2228b-bbc2-496f-9512-529f23735aa9}} regularizer, by using asophisticated functional analysis method, Zhang et al.{{cite:7470e9a198c3bde1c2cafb4bc0379a7fba0e2076}} and Song et al. {{cite:6aa94efbce1dbaee94b2a4b7739b1d582a32d221}} built theregularized least square algorithm on the reproducing kernel Banachspace (RKBS), and they proved that the regularized least squarealgorithm in RKBS is equivalent to {{formula:cb08ab63-d8b7-45ab-9d54-d7f0597994f9}} regularizer if the kernelsatisfies some restricted conditions. Following this method, Songand Zhang {{cite:267d8c2a9ceb6f67b61458b065d641ad0f2512cd}} deduced a similar learning rate for the{{formula:75a43ceb-0919-4d03-8da9-a619e34dd451}} regularizer and eliminated the concentration propertyassumption on the marginal distribution .", "Limiting {{formula:c8279225-2b42-4e4e-8128-9415d4d45e77}} within {{formula:076d0362-f577-484a-9410-0b98c597df8c}} is certainly incomplete to judgewhether the generalization capability of {{formula:d23246a3-8d06-4a6f-9109-b082c4061c52}} regularizationdepends on the choice of {{formula:188bd2a4-1af9-4163-bcf6-e450b6191ce5}} . Moreover, in the context of learningtheory, to intrinsically characterize the generalization capabilityof a learning strategy, the essential generalization bound{{cite:9a05fbb4bd0d643f5daff5229138bf8c160b22da}} rather than the upper bound is required, that is,we must deduce a lower and an upper bound simultaneously for thelearning strategy and prove that the upper and lower bounds can beasymptotically identical. We notice, however, that most of thepreviously known estimations on generalization capability oflearning schemes (REF )are onlyconcerned with the upper bound estimation. Thus, their results can not serve the answerto Problem 1.Different from the pervious work, the essential bound estimation ofgeneralization error for {{formula:ed9d3e7a-ae80-414b-aea5-d26413de771f}} regularization schemes(REF ) with {{formula:1cf7fc80-603a-4979-a503-7b79eb92eec4}} will be presented in thepresent paper. As a consequence, we provide an affirmative answer toProblem 1." ]

[]