Datasets:

davidberenstein1957

/

daily-papers-docling

like 2

c965977f-2ec3-4a59-929f-c01b1c46ba01

2302.06555v2.pdf

text

Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learning , pages 8748-8763. PMLR.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/171", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 71.14116668701172, "t": 776.1381225585938, "r": 292.1780090332031, "b": 684.1265869140625, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 304]}], "orig": "Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learning , pages 8748-8763. PMLR.", "text": "Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learning , pages 8748-8763. PMLR."}

38e50e86-7373-4db7-b4b9-91daeb82f267

2302.06555v2.pdf

text

Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. 2019. Language models are unsupervised multitask learners.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/172", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 71.21831512451172, "t": 670.2796630859375, "r": 292.0796203613281, "b": 618.6445922851562, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 140]}], "orig": "Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. 2019. Language models are unsupervised multitask learners.", "text": "Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. 2019. Language models are unsupervised multitask learners."}

c8a0f20b-de90-4b54-aebb-74be5a23c0f2

2302.06555v2.pdf

text

William J. Rapaport. 2002. Holism, conceptualrole semantics, and syntactic semantics. Minds and Machines , 12(1):3-59.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/173", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 71.34474182128906, "t": 604.24462890625, "r": 292.07537841796875, "b": 566.3685913085938, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 118]}], "orig": "William J. Rapaport. 2002. Holism, conceptualrole semantics, and syntactic semantics. Minds and Machines , 12(1):3-59.", "text": "William J. Rapaport. 2002. Holism, conceptualrole semantics, and syntactic semantics. Minds and Machines , 12(1):3-59."}

a97c6f67-c753-4991-bdd7-8b9d156a3174

2302.06555v2.pdf

text

Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Li Fei-Fei. 2015. ImageNet Large Scale Visual Recognition Challenge. International Journal of Computer Vision (IJCV) , 115(3):211-252.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/174", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 71.34825897216797, "t": 551.4522705078125, "r": 292.07965087890625, "b": 459.50482177734375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 303]}], "orig": "Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Li Fei-Fei. 2015. ImageNet Large Scale Visual Recognition Challenge. International Journal of Computer Vision (IJCV) , 115(3):211-252.", "text": "Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Li Fei-Fei. 2015. ImageNet Large Scale Visual Recognition Challenge. International Journal of Computer Vision (IJCV) , 115(3):211-252."}

ee98c152-482e-4ea3-9317-de4d3c25fe55

2302.06555v2.pdf

text

Magnus Sahlgren and Fredrik Carlsson. 2021. The singleton fallacy: Why current critiques of language models miss the point. Frontiers in Artificial Intelligence , 4(682578).

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/175", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 71.13983917236328, "t": 445.57135009765625, "r": 292.07958984375, "b": 393.9989929199219, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 173]}], "orig": "Magnus Sahlgren and Fredrik Carlsson. 2021. The singleton fallacy: Why current critiques of language models miss the point. Frontiers in Artificial Intelligence , 4(682578).", "text": "Magnus Sahlgren and Fredrik Carlsson. 2021. The singleton fallacy: Why current critiques of language models miss the point. Frontiers in Artificial Intelligence , 4(682578)."}

4e49d633-f5a6-4196-9ae5-474684c0c889

2302.06555v2.pdf

text

Jona Sassenhagen and Christian J. Fiebach. 2020. Traces of Meaning Itself: Encoding Distributional Word Vectors in Brain Activity. Neurobiology of Language , 1(1):54-76.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/176", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 70.33506774902344, "t": 379.6624450683594, "r": 292.17694091796875, "b": 327.645751953125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 169]}], "orig": "Jona Sassenhagen and Christian J. Fiebach. 2020. Traces of Meaning Itself: Encoding Distributional Word Vectors in Brain Activity. Neurobiology of Language , 1(1):54-76.", "text": "Jona Sassenhagen and Christian J. Fiebach. 2020. Traces of Meaning Itself: Encoding Distributional Word Vectors in Brain Activity. Neurobiology of Language , 1(1):54-76."}

d8a5a2ba-84c7-4348-bb1c-492300b31ef5

2302.06555v2.pdf

text

Peter H Schönemann. 1966. A generalized solution of the orthogonal procrustes problem. Psychometrika , 31(1):1-10.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/177", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 71.24508666992188, "t": 313.40948486328125, "r": 292.0789489746094, "b": 275.96563720703125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 114]}], "orig": "Peter H Sch\u00f6nemann. 1966. A generalized solution of the orthogonal procrustes problem. Psychometrika , 31(1):1-10.", "text": "Peter H Sch\u00f6nemann. 1966. A generalized solution of the orthogonal procrustes problem. Psychometrika , 31(1):1-10."}

6407cc85-86c3-4d8e-8797-3306c9f82ed9

2302.06555v2.pdf

text

Martin Schrimpf, Idan Blank, Greta Tuckute, Carina Kauf, Eghbal A. Hosseini, Nancy Kanwisher, Joshua Tenenbaum, and Evelina Fedorenko. 2021. The neural architecture of language: Integrative modeling converges on predictive processing. bioRxiv .

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/178", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 71.0137939453125, "t": 261.80999755859375, "r": 292.07965087890625, "b": 182.22943115234375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 244]}], "orig": "Martin Schrimpf, Idan Blank, Greta Tuckute, Carina Kauf, Eghbal A. Hosseini, Nancy Kanwisher, Joshua Tenenbaum, and Evelina Fedorenko. 2021. The neural architecture of language: Integrative modeling converges on predictive processing. bioRxiv .", "text": "Martin Schrimpf, Idan Blank, Greta Tuckute, Carina Kauf, Eghbal A. Hosseini, Nancy Kanwisher, Joshua Tenenbaum, and Evelina Fedorenko. 2021. The neural architecture of language: Integrative modeling converges on predictive processing. bioRxiv ."}

1064c112-f7f7-417a-8726-c292c4f057ae

2302.06555v2.pdf

text

Martin Schrimpf, Jonas Kubilius, Ha Hong, Najib Majaj, Rishi Rajalingham, Elias B. Issa, Kohitij Kar, Pouya Bashivan, Jonathan PrescottRoy, Kailyn Schmidt, Daniel L. K. Yamins, and James J. DiCarlo. 2018. Brain-score: Which artificial neural network for object recognition is most brain-like? bioRxiv .

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/179", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 71.23005676269531, "t": 168.40966796875, "r": 292.07965087890625, "b": 76.56763458251953, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 302]}], "orig": "Martin Schrimpf, Jonas Kubilius, Ha Hong, Najib Majaj, Rishi Rajalingham, Elias B. Issa, Kohitij Kar, Pouya Bashivan, Jonathan PrescottRoy, Kailyn Schmidt, Daniel L. K. Yamins, and James J. DiCarlo. 2018. Brain-score: Which artificial neural network for object recognition is most brain-like? bioRxiv .", "text": "Martin Schrimpf, Jonas Kubilius, Ha Hong, Najib Majaj, Rishi Rajalingham, Elias B. Issa, Kohitij Kar, Pouya Bashivan, Jonathan PrescottRoy, Kailyn Schmidt, Daniel L. K. Yamins, and James J. DiCarlo. 2018. Brain-score: Which artificial neural network for object recognition is most brain-like? bioRxiv ."}

3c3cdbd8-4c94-44dd-882a-edce5b01d57c

2302.06555v2.pdf

text

John R. Searle. 1980. Minds, brains, and programs. Behavioral and Brain Sciences , 3:417-424.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/180", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 306.22113037109375, "t": 776.2285766601562, "r": 527.4554443359375, "b": 752.2166137695312, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 93]}], "orig": "John R. Searle. 1980. Minds, brains, and programs. Behavioral and Brain Sciences , 3:417-424.", "text": "John R. Searle. 1980. Minds, brains, and programs. Behavioral and Brain Sciences , 3:417-424."}

a06eb85a-0f05-4769-a0b6-08ab7a3d9394

2302.06555v2.pdf

text

Nicholas Shea. 2018. Representation in Cognitive Science . Oxford University Press.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/181", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 306.5379943847656, "t": 740.894287109375, "r": 525.5486450195312, "b": 716.62060546875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 83]}], "orig": "Nicholas Shea. 2018. Representation in Cognitive Science . Oxford University Press.", "text": "Nicholas Shea. 2018. Representation in Cognitive Science . Oxford University Press."}

ee67461b-22b6-4d9e-b3bb-a515b35d1835

2302.06555v2.pdf

text

Anders Søgaard, Sebastian Ruder, and Ivan Vuli'c. 2018. On the Limitations of Unsupervised Bilingual Dictionary Induction. In Proceedings of ACL 2018 .

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/182", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 306.6299133300781, "t": 705.385498046875, "r": 527.45263671875, "b": 654.4376220703125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 151]}], "orig": "Anders S\u00f8gaard, Sebastian Ruder, and Ivan Vuli'c. 2018. On the Limitations of Unsupervised Bilingual Dictionary Induction. In Proceedings of ACL 2018 .", "text": "Anders S\u00f8gaard, Sebastian Ruder, and Ivan Vuli'c. 2018. On the Limitations of Unsupervised Bilingual Dictionary Induction. In Proceedings of ACL 2018 ."}

28e3a887-36a1-44fa-b63f-aaa67f90f1af

2302.06555v2.pdf

text

Ryan Teehan, Miruna Clinciu, Oleg Serikov, Eliza Szczechla, Natasha Seelam, Shachar Mirkin, and Aaron Gokaslan. 2022. Emergent structures and training dynamics in large language models. In Proceedings of BigScience Episode #5 - Workshop on Challenges & Perspectives in Creating Large Language Models , pages 146-159, virtual+Dublin. Association for Computational Linguistics.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/183", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 307.2760009765625, "t": 643.467529296875, "r": 527.352783203125, "b": 523.699462890625, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 375]}], "orig": "Ryan Teehan, Miruna Clinciu, Oleg Serikov, Eliza Szczechla, Natasha Seelam, Shachar Mirkin, and Aaron Gokaslan. 2022. Emergent structures and training dynamics in large language models. In Proceedings of BigScience Episode #5 - Workshop on Challenges & Perspectives in Creating Large Language Models , pages 146-159, virtual+Dublin. Association for Computational Linguistics.", "text": "Ryan Teehan, Miruna Clinciu, Oleg Serikov, Eliza Szczechla, Natasha Seelam, Shachar Mirkin, and Aaron Gokaslan. 2022. Emergent structures and training dynamics in large language models. In Proceedings of BigScience Episode #5 - Workshop on Challenges & Perspectives in Creating Large Language Models , pages 146-159, virtual+Dublin. Association for Computational Linguistics."}

6140df5b-cbe1-41c5-828f-82164f0fcb0d

2302.06555v2.pdf

text

Mariya Toneva and Leila Wehbe. 2019. Interpreting and improving natural-language processing (in machines) with natural language-processing (in the brain). Advances in neural information processing systems , 32.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/184", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 306.4243469238281, "t": 512.8441162109375, "r": 527.3577270507812, "b": 447.7977294921875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 210]}], "orig": "Mariya Toneva and Leila Wehbe. 2019. Interpreting and improving natural-language processing (in machines) with natural language-processing (in the brain). Advances in neural information processing systems , 32.", "text": "Mariya Toneva and Leila Wehbe. 2019. Interpreting and improving natural-language processing (in machines) with natural language-processing (in the brain). Advances in neural information processing systems , 32."}

4c136523-3928-4aae-a443-58c8dfafd370

2302.06555v2.pdf

text

Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288 .

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/185", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 307.0433654785156, "t": 436.8598937988281, "r": 527.4573364257812, "b": 358.6497802734375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 255]}], "orig": "Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288 .", "text": "Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288 ."}

4c15fd40-fe13-416c-87ee-592fb35ae8d8

2302.06555v2.pdf

text

Iulia Turc, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Well-read students learn better: On the importance of pre-training compact models. arXiv preprint arXiv:1908.08962v2 .

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/186", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 307.0731506347656, "t": 347.679931640625, "r": 527.3555908203125, "b": 282.7406311035156, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 188]}], "orig": "Iulia Turc, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Well-read students learn better: On the importance of pre-training compact models. arXiv preprint arXiv:1908.08962v2 .", "text": "Iulia Turc, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Well-read students learn better: On the importance of pre-training compact models. arXiv preprint arXiv:1908.08962v2 ."}

1e975cb8-0535-42fd-9d33-f494ec518ec4

2302.06555v2.pdf

text

Ivan Vuli'c, Douwe Kiela, Stephen Clark, and Marie-Francine Moens. 2016. Multi-modal representations for improved bilingual lexicon learning. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , pages 188-194, Berlin, Germany. Association for Computational Linguistics.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/187", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 307.1670837402344, "t": 271.49267578125, "r": 527.3556518554688, "b": 165.977294921875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 334]}], "orig": "Ivan Vuli'c, Douwe Kiela, Stephen Clark, and Marie-Francine Moens. 2016. Multi-modal representations for improved bilingual lexicon learning. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , pages 188-194, Berlin, Germany. Association for Computational Linguistics.", "text": "Ivan Vuli'c, Douwe Kiela, Stephen Clark, and Marie-Francine Moens. 2016. Multi-modal representations for improved bilingual lexicon learning. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , pages 188-194, Berlin, Germany. Association for Computational Linguistics."}

bca58df1-69ce-43fc-af73-2f4b7aa30e60

2302.06555v2.pdf

text

Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed H. Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, and William Fedus. 2022. Emergent abilities of large language

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/188", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 15, "bbox": {"l": 306.69537353515625, "t": 154.9891357421875, "r": 527.3553466796875, "b": 75.83184814453125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 273]}], "orig": "Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed H. Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, and William Fedus. 2022. Emergent abilities of large language", "text": "Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed H. Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, and William Fedus. 2022. Emergent abilities of large language"}

e7219246-56c2-429a-8cc2-fa779175000a

2302.06555v2.pdf

text

models. Transactions on Machine Learning Research . Survey Certification.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/189", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 16, "bbox": {"l": 82.18153381347656, "t": 776.12890625, "r": 292.0752868652344, "b": 751.8050537109375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 73]}], "orig": "models. Transactions on Machine Learning Research . Survey Certification.", "text": "models. Transactions on Machine Learning Research . Survey Certification."}

588d2b28-6bec-4593-80a4-aed9acd70a37

2302.06555v2.pdf

text

Daniel Williams. 2018. Predictive processing and the representation wars. Minds and Machines , 28(1):141-172.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/190", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 16, "bbox": {"l": 71.25851440429688, "t": 738.0191650390625, "r": 291.6308288574219, "b": 700.214599609375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 109]}], "orig": "Daniel Williams. 2018. Predictive processing and the representation wars. Minds and Machines , 28(1):141-172.", "text": "Daniel Williams. 2018. Predictive processing and the representation wars. Minds and Machines , 28(1):141-172."}

0da9998f-8b79-4109-ba88-eb48c89786ca

2302.06555v2.pdf

text

Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lhoest, and Alexander Rush. 2020. Transformers: State-of-the-art natural language processing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations , pages 38-45, Online. Association for Computational Linguistics.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/191", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 16, "bbox": {"l": 70.77729797363281, "t": 685.7984008789062, "r": 292.0815124511719, "b": 512.7030029296875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 568]}], "orig": "Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lhoest, and Alexander Rush. 2020. Transformers: State-of-the-art natural language processing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations , pages 38-45, Online. Association for Computational Linguistics.", "text": "Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lhoest, and Alexander Rush. 2020. Transformers: State-of-the-art natural language processing. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations , pages 38-45, Online. Association for Computational Linguistics."}

56e51cda-5afc-4d69-a752-3f367bd9a764

2302.06555v2.pdf

text

Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, and Ping Luo. 2021. Segformer: Simple and efficient design for semantic segmentation with transformers. In Advances in Neural Information Processing Systems , volume 34, pages 12077-12090. Curran Associates, Inc.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/192", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 16, "bbox": {"l": 71.51163482666016, "t": 497.9288635253906, "r": 292.17620849609375, "b": 406.5206604003906, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 282]}], "orig": "Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, and Ping Luo. 2021. Segformer: Simple and efficient design for semantic segmentation with transformers. In Advances in Neural Information Processing Systems , volume 34, pages 12077-12090. Curran Associates, Inc.", "text": "Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, and Ping Luo. 2021. Segformer: Simple and efficient design for semantic segmentation with transformers. In Advances in Neural Information Processing Systems , volume 34, pages 12077-12090. Curran Associates, Inc."}

706ae8aa-f25b-45e9-8676-567c7ad9a6d1

2302.06555v2.pdf

text

Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen, Christopher Dewan, Mona Diab, Xian Li, Xi Victoria Lin, Todor Mihaylov, Myle Ott, Sam Shleifer, Kurt Shuster, Daniel Simig, Punit Singh Koura, Anjali Sridhar, Tianlu Wang, and Luke Zettlemoyer. 2022. Opt: Open pre-trained transformer language models.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/193", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 16, "bbox": {"l": 71.23656463623047, "t": 391.8356018066406, "r": 292.1730041503906, "b": 286.18853759765625, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 332]}], "orig": "Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen, Christopher Dewan, Mona Diab, Xian Li, Xi Victoria Lin, Todor Mihaylov, Myle Ott, Sam Shleifer, Kurt Shuster, Daniel Simig, Punit Singh Koura, Anjali Sridhar, Tianlu Wang, and Luke Zettlemoyer. 2022. Opt: Open pre-trained transformer language models.", "text": "Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen, Christopher Dewan, Mona Diab, Xian Li, Xi Victoria Lin, Todor Mihaylov, Myle Ott, Sam Shleifer, Kurt Shuster, Daniel Simig, Punit Singh Koura, Anjali Sridhar, Tianlu Wang, and Luke Zettlemoyer. 2022. Opt: Open pre-trained transformer language models."}

a725105e-47f8-4002-90e8-ec3c1c284d51

2302.06555v2.pdf

text

Jiawei Zhao and Andrew Gilman. 2020. Nonlinearity in mapping based cross-lingual word embeddings. In Proceedings of the 12th Language Resources and Evaluation Conference , pages 3583-3589, Marseille, France. European Language Resources Association.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/194", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 16, "bbox": {"l": 70.4427261352539, "t": 272.12591552734375, "r": 292.0821228027344, "b": 193.81298828125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 248]}], "orig": "Jiawei Zhao and Andrew Gilman. 2020. Nonlinearity in mapping based cross-lingual word embeddings. In Proceedings of the 12th Language Resources and Evaluation Conference , pages 3583-3589, Marseille, France. European Language Resources Association.", "text": "Jiawei Zhao and Andrew Gilman. 2020. Nonlinearity in mapping based cross-lingual word embeddings. In Proceedings of the 12th Language Resources and Evaluation Conference , pages 3583-3589, Marseille, France. European Language Resources Association."}

f6343bdb-e8e3-4526-8aa4-49850563b008

2302.06555v2.pdf

text

Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Barriuso, and Antonio Torralba. 2017. Scene parsing through ade20k dataset. computer vision and pattern recognition .

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/195", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 16, "bbox": {"l": 71.41193389892578, "t": 179.43560791015625, "r": 292.17620849609375, "b": 127.77508544921875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 173]}], "orig": "Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Barriuso, and Antonio Torralba. 2017. Scene parsing through ade20k dataset. computer vision and pattern recognition .", "text": "Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Barriuso, and Antonio Torralba. 2017. Scene parsing through ade20k dataset. computer vision and pattern recognition ."}

50d045d7-48fa-4202-8bea-233c2e6ebc2d

2302.06555v2.pdf

text

Yukun Zhu, Ryan Kiros, Richard S. Zemel, Ruslan Salakhutdinov, Raquel Urtasun, Antonio Torralba, and Sanja Fidler. 2015. Aligning books

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/196", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 16, "bbox": {"l": 71.28720092773438, "t": 114.25738525390625, "r": 292.0768127441406, "b": 75.94146728515625, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 135]}], "orig": "Yukun Zhu, Ryan Kiros, Richard S. Zemel, Ruslan Salakhutdinov, Raquel Urtasun, Antonio Torralba, and Sanja Fidler. 2015. Aligning books", "text": "Yukun Zhu, Ryan Kiros, Richard S. Zemel, Ruslan Salakhutdinov, Raquel Urtasun, Antonio Torralba, and Sanja Fidler. 2015. Aligning books"}

fcb05201-b6eb-4b30-aa6c-afebdd3bc784

2302.06555v2.pdf

text

and movies: Towards story-like visual explanations by watching movies and reading books. 2015 IEEE International Conference on Computer Vision (ICCV) , pages 19-27.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/197", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 16, "bbox": {"l": 316.86114501953125, "t": 776.248046875, "r": 527.455810546875, "b": 724.501220703125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 164]}], "orig": "and movies: Towards story-like visual explanations by watching movies and reading books. 2015 IEEE International Conference on Computer Vision (ICCV) , pages 19-27.", "text": "and movies: Towards story-like visual explanations by watching movies and reading books. 2015 IEEE International Conference on Computer Vision (ICCV) , pages 19-27."}

9eeb8e6b-7770-4b9f-b466-e6e512803c64

2302.06555v2.pdf

text

Andy Zou, Long Phan, Sarah Chen, James Campbell, Phillip Guo, Richard Ren, Alexander Pan, Xuwang Yin, Mantas Mazeika, Ann-Kathrin Dombrowski, et al. 2023. Representation engineering: A top-down approach to ai transparency. arXiv preprint arXiv:2310.01405 .

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/198", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 16, "bbox": {"l": 306.3648376464844, "t": 710.09228515625, "r": 527.448974609375, "b": 631.9486083984375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 256]}], "orig": "Andy Zou, Long Phan, Sarah Chen, James Campbell, Phillip Guo, Richard Ren, Alexander Pan, Xuwang Yin, Mantas Mazeika, Ann-Kathrin Dombrowski, et al. 2023. Representation engineering: A top-down approach to ai transparency. arXiv preprint arXiv:2310.01405 .", "text": "Andy Zou, Long Phan, Sarah Chen, James Campbell, Phillip Guo, Richard Ren, Alexander Pan, Xuwang Yin, Mantas Mazeika, Ann-Kathrin Dombrowski, et al. 2023. Representation engineering: A top-down approach to ai transparency. arXiv preprint arXiv:2310.01405 ."}

e4daa604-5987-4336-9aab-71f4e67e7762

2302.06555v2.pdf

section_header

A Detailed Experimental Settings

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/199", "parent": {"cref": "#/body"}, "children": [], "label": "section_header", "prov": [{"page_no": 16, "bbox": {"l": 306.2620544433594, "t": 616.3843383789062, "r": 486.592041015625, "b": 604.2142333984375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 32]}], "orig": "A Detailed Experimental Settings", "text": "A Detailed Experimental Settings", "level": 1}

5a95bc5d-fb48-45b6-8ee9-d9030663e9e6

2302.06555v2.pdf

section_header

A.1 Computational Environment

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/200", "parent": {"cref": "#/body"}, "children": [], "label": "section_header", "prov": [{"page_no": 16, "bbox": {"l": 305.94427490234375, "t": 592.9501342773438, "r": 468.4688415527344, "b": 581.759765625, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 29]}], "orig": "A.1 Computational Environment", "text": "A.1 Computational Environment", "level": 1}

f46bd9dc-6e48-4794-a661-111c2ef3f30a

2302.06555v2.pdf

text

Our primary software toolkits included HuggingFace Transformers 4.36.2 (Wolf et al., 2020), PyTorch 2.1.2 (Paszke et al., 2019). We ran our experiments on 2 NVIDIA A100s 40G.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/201", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 16, "bbox": {"l": 305.7516784667969, "t": 573.577392578125, "r": 527.3517456054688, "b": 522.483642578125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 174]}], "orig": "Our primary software toolkits included HuggingFace Transformers 4.36.2 (Wolf et al., 2020), PyTorch 2.1.2 (Paszke et al., 2019). We ran our experiments on 2 NVIDIA A100s 40G.", "text": "Our primary software toolkits included HuggingFace Transformers 4.36.2 (Wolf et al., 2020), PyTorch 2.1.2 (Paszke et al., 2019). We ran our experiments on 2 NVIDIA A100s 40G."}

c7b52e75-0fdb-454e-8414-d33eb838aa54

2302.06555v2.pdf

section_header

A.2 Model Details

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/202", "parent": {"cref": "#/body"}, "children": [], "label": "section_header", "prov": [{"page_no": 16, "bbox": {"l": 306.1889953613281, "t": 508.5110778808594, "r": 398.79168701171875, "b": 497.6560363769531, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 17]}], "orig": "A.2 Model Details", "text": "A.2 Model Details", "level": 1}

0c4f575f-3be4-41f5-b3a5-0f84980a105b

2302.06555v2.pdf

text

Except for the ResNet and CLIP models, all other models used in this study are from the Huggingface Transformers (Table 7).

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/203", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 16, "bbox": {"l": 306.1834716796875, "t": 489.63348388671875, "r": 525.7245483398438, "b": 451.5606384277344, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 123]}], "orig": "Except for the ResNet and CLIP models, all other models used in this study are from the Huggingface Transformers (Table 7).", "text": "Except for the ResNet and CLIP models, all other models used in this study are from the Huggingface Transformers (Table 7)."}

2c8a251a-8e5c-4162-a86e-3f2c762f056d

2302.06555v2.pdf

caption

Table 7: Sources of models in our experiments.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/204", "parent": {"cref": "#/body"}, "children": [], "label": "caption", "prov": [{"page_no": 16, "bbox": {"l": 310.9766845703125, "t": 211.72222900390625, "r": 520.7345581054688, "b": 200.37744140625, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 46]}], "orig": "Table 7: Sources of models in our experiments.", "text": "Table 7: Sources of models in our experiments."}

a1c64ce6-c7a3-4830-bac0-9a7d24cf9145

2302.06555v2.pdf

table

Models	Links
BERT$_{TINY}$ BERT$_{MINI}$ BERT$_{SMALL}$	https://huggingface.co/google/bert_uncased_L-2_H-128_A-2 https://huggingface.co/google/bert_uncased_L-4_H-256_A-4 https://huggingface.co/google/bert_uncased_L-4_H-512_A-8 https://huggingface.co/google/bert_uncased_L-8_H-512_A-8 https://huggingface.co/bert-base-uncased https://huggingface.co/bert-large-uncased
BERT$_{MEDIUM}$
BERT$_{BASE}$
BERT$_{LARGE}$
GPT-2$_{BASE}$	https://huggingface.co/openai-community/gpt2
GPT-2$_{LARGE}$	https://huggingface.co/openai-community/gpt2-large
GPT-2$_{XL}$	https://huggingface.co/openai-community/gpt2-xl
OPT$_{125M}$	https://huggingface.co/facebook/opt-125m
OPT$_{6.7B}$	https://huggingface.co/facebook/opt-6.7b
OPT$_{30B}$	https://huggingface.co/facebook/opt-30b
LLaMA-2$_{7B}$	https://huggingface.co/meta-llama/Llama-2-7b
LLaMA-2$_{13B}$	https://huggingface.co/meta-llama/Llama-2-13b
SegFormer-B0	https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512
SegFormer-B1	https://huggingface.co/nvidia/segformer-b1-finetuned-ade-512-512
SegFormer-B2	https://huggingface.co/nvidia/segformer-b2-finetuned-ade-512-512
SegFormer-B3	https://huggingface.co/nvidia/segformer-b3-finetuned-ade-512-512
SegFormer-B4 SegFormer-B5	https://huggingface.co/nvidia/segformer-b4-finetuned-ade-512-512 https://huggingface.co/nvidia/segformer-b5-finetuned-ade-640-640
MAE$_{BASE}$ MAE$_{LARGE}$	https://huggingface.co/facebook/vit-mae-base https://huggingface.co/facebook/vit-mae-large
MAE$_{HUGE}$	https://huggingface.co/facebook/vit-mae-huge
ResNet18
ResNet34
ResNet50	https://pypi.org/project/img2vec-pytorch/
ResNet101
ResNet152
CLIP-RN50
CLIP-RN101 CLIP-RN50*64	https://github.com/openai/CLIP
CLIP-VIT-B-32

Table 7: Sources of models in our experiments.

image/png

application/pdf

1.0.0

{"self_ref": "#/tables/6", "parent": {"cref": "#/body"}, "children": [], "label": "table", "prov": [{"page_no": 16, "bbox": {"l": 307.3106384277344, "t": 438.501220703125, "r": 525.870361328125, "b": 222.412109375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 0]}], "captions": [{"cref": "#/texts/204"}], "references": [], "footnotes": [], "image": {"mimetype": "image/png", "dpi": 144, "size": {"width": 437.0, "height": 432.0}, "uri": null}, "data": {"table_cells": [{"bbox": {"l": 310.941650390625, "t": 436.0203857421875, "r": 325.2330322265625, "b": 431.8905334472656, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "Models", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 428.6837463378906, "t": 436.0203857421875, "r": 439.92022705078125, "b": 431.8905334472656, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "Links", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 427.8594970703125, "r": 333.781982421875, "b": 412.3416442871094, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "BERT$_{TINY}$ BERT$_{MINI}$ BERT$_{SMALL}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 356.8112487792969, "t": 427.61602783203125, "r": 511.79302978515625, "b": 395.541748046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/google/bert_uncased_L-2_H-128_A-2 https://huggingface.co/google/bert_uncased_L-4_H-256_A-4 https://huggingface.co/google/bert_uncased_L-4_H-512_A-8 https://huggingface.co/google/bert_uncased_L-8_H-512_A-8 https://huggingface.co/bert-base-uncased https://huggingface.co/bert-large-uncased", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 410.74261474609375, "r": 336.3897705078125, "b": 406.6357421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "BERT$_{MEDIUM}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 405.0371398925781, "r": 330.8688049316406, "b": 400.9302673339844, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "BERT$_{BASE}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.9416809082031, "t": 399.3316650390625, "r": 333.59405517578125, "b": 395.22479248046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "BERT$_{LARGE}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 391.22589111328125, "r": 331.48443603515625, "b": 387.1190490722656, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "GPT-2$_{BASE}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 373.55841064453125, "t": 390.982421875, "r": 495.0466003417969, "b": 387.43603515625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/openai-community/gpt2", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 385.52044677734375, "r": 334.209716796875, "b": 381.41357421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "GPT-2$_{LARGE}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 365.184814453125, "t": 385.2769470214844, "r": 503.41754150390625, "b": 381.73052978515625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/openai-community/gpt2-large", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.9416809082031, "t": 379.81494140625, "r": 327.497314453125, "b": 375.70806884765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 7, "end_row_offset_idx": 8, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "GPT-2$_{XL}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 369.3191833496094, "t": 379.57147216796875, "r": 499.28302001953125, "b": 376.0250549316406, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 7, "end_row_offset_idx": 8, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/openai-community/gpt2-xl", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 371.7087707519531, "r": 327.63726806640625, "b": 367.6018981933594, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 8, "end_row_offset_idx": 9, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "OPT$_{125M}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 379.0710144042969, "t": 371.4653015136719, "r": 489.5340576171875, "b": 367.91888427734375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 8, "end_row_offset_idx": 9, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/facebook/opt-125m", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 366.0032958984375, "r": 326.0531311035156, "b": 361.89642333984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 9, "end_row_offset_idx": 10, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "OPT$_{6.7B}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 379.0710144042969, "t": 365.75982666015625, "r": 489.5340576171875, "b": 362.21337890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 9, "end_row_offset_idx": 10, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/facebook/opt-6.7b", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 360.29779052734375, "r": 325.2140808105469, "b": 356.19091796875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 10, "end_row_offset_idx": 11, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "OPT$_{30B}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 380.44927978515625, "t": 360.0543212890625, "r": 488.1560363769531, "b": 356.5079040527344, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 10, "end_row_offset_idx": 11, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/facebook/opt-30b", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 352.1920471191406, "r": 333.73883056640625, "b": 348.0851745605469, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 11, "end_row_offset_idx": 12, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "LLaMA-2$_{7B}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 373.34869384765625, "t": 351.9485778808594, "r": 495.2549743652344, "b": 348.40216064453125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 11, "end_row_offset_idx": 12, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/meta-llama/Llama-2-7b", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.9416809082031, "t": 346.486572265625, "r": 335.41693115234375, "b": 342.37969970703125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 12, "end_row_offset_idx": 13, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "LLaMA-2$_{13B}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 371.9704284667969, "t": 346.24310302734375, "r": 496.6329650878906, "b": 342.6966552734375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 12, "end_row_offset_idx": 13, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/meta-llama/Llama-2-13b", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 338.38079833984375, "r": 338.0176696777344, "b": 334.27392578125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 13, "end_row_offset_idx": 14, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "SegFormer-B0", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 345.5762939453125, "t": 338.1373291015625, "r": 523.0264282226562, "b": 334.5909118652344, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 13, "end_row_offset_idx": 14, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.9416809082031, "t": 332.6749267578125, "r": 338.0176696777344, "b": 328.56805419921875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 14, "end_row_offset_idx": 15, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "SegFormer-B1", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 345.5762939453125, "t": 332.4314270019531, "r": 523.0264282226562, "b": 328.885009765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 14, "end_row_offset_idx": 15, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/nvidia/segformer-b1-finetuned-ade-512-512", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.9416809082031, "t": 326.96942138671875, "r": 338.0176696777344, "b": 322.862548828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 15, "end_row_offset_idx": 16, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "SegFormer-B2", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 345.5762939453125, "t": 326.7259521484375, "r": 523.0264282226562, "b": 323.17950439453125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 15, "end_row_offset_idx": 16, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/nvidia/segformer-b2-finetuned-ade-512-512", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.9416809082031, "t": 321.263916015625, "r": 338.0176696777344, "b": 317.15704345703125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 16, "end_row_offset_idx": 17, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "SegFormer-B3", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 345.5762939453125, "t": 321.02044677734375, "r": 523.0264282226562, "b": 317.47406005859375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 16, "end_row_offset_idx": 17, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/nvidia/segformer-b3-finetuned-ade-512-512", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.9416809082031, "t": 315.5584716796875, "r": 338.0176696777344, "b": 305.74609375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 17, "end_row_offset_idx": 18, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "SegFormer-B4 SegFormer-B5", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 345.5762939453125, "t": 315.3149719238281, "r": 523.0264282226562, "b": 306.06304931640625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 17, "end_row_offset_idx": 18, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/nvidia/segformer-b4-finetuned-ade-512-512 https://huggingface.co/nvidia/segformer-b5-finetuned-ade-640-640", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 301.7472229003906, "r": 332.33538818359375, "b": 291.9344482421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 18, "end_row_offset_idx": 19, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "MAE$_{BASE}$ MAE$_{LARGE}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 372.07568359375, "t": 301.50372314453125, "r": 496.52691650390625, "b": 292.25140380859375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 18, "end_row_offset_idx": 19, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/facebook/vit-mae-base https://huggingface.co/facebook/vit-mae-large", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 290.3358154296875, "r": 330.46929931640625, "b": 286.22894287109375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 19, "end_row_offset_idx": 20, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "MAE$_{HUGE}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 373.45355224609375, "t": 290.09234619140625, "r": 495.1484680175781, "b": 286.5459289550781, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 19, "end_row_offset_idx": 20, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/facebook/vit-mae-huge", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 282.2300720214844, "r": 329.0596923828125, "b": 278.1231994628906, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 20, "end_row_offset_idx": 21, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "ResNet18", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 20, "end_row_offset_idx": 21, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 276.52459716796875, "r": 329.0596923828125, "b": 272.417724609375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 21, "end_row_offset_idx": 22, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "ResNet34", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 21, "end_row_offset_idx": 22, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 270.819091796875, "r": 329.0596923828125, "b": 266.71221923828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 22, "end_row_offset_idx": 23, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "ResNet50", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 377.69317626953125, "t": 270.57562255859375, "r": 490.9125061035156, "b": 267.0292053222656, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 22, "end_row_offset_idx": 23, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://pypi.org/project/img2vec-pytorch/", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 265.1136169433594, "r": 331.35662841796875, "b": 261.0067443847656, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 23, "end_row_offset_idx": 24, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "ResNet101", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 23, "end_row_offset_idx": 24, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 259.40771484375, "r": 331.35662841796875, "b": 255.30084228515625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 24, "end_row_offset_idx": 25, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "ResNet152", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 24, "end_row_offset_idx": 25, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 251.30197143554688, "r": 333.40087890625, "b": 247.19509887695312, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 25, "end_row_offset_idx": 26, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-RN50", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 25, "end_row_offset_idx": 26, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.941650390625, "t": 245.5964813232422, "r": 340.2915954589844, "b": 235.78411865234375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 26, "end_row_offset_idx": 27, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-RN101 CLIP-RN50*64", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 392.9576416015625, "t": 239.64752197265625, "r": 475.6464538574219, "b": 236.10110473632812, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 26, "end_row_offset_idx": 27, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://github.com/openai/CLIP", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 310.9416809082031, "t": 234.18551635742188, "r": 340.3742980957031, "b": 230.07864379882812, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 27, "end_row_offset_idx": 28, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-VIT-B-32", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 27, "end_row_offset_idx": 28, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}], "num_rows": 28, "num_cols": 2, "grid": [[{"bbox": {"l": 310.941650390625, "t": 436.0203857421875, "r": 325.2330322265625, "b": 431.8905334472656, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "Models", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 428.6837463378906, "t": 436.0203857421875, "r": 439.92022705078125, "b": 431.8905334472656, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "Links", "column_header": true, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 427.8594970703125, "r": 333.781982421875, "b": 412.3416442871094, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "BERT$_{TINY}$ BERT$_{MINI}$ BERT$_{SMALL}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 356.8112487792969, "t": 427.61602783203125, "r": 511.79302978515625, "b": 395.541748046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/google/bert_uncased_L-2_H-128_A-2 https://huggingface.co/google/bert_uncased_L-4_H-256_A-4 https://huggingface.co/google/bert_uncased_L-4_H-512_A-8 https://huggingface.co/google/bert_uncased_L-8_H-512_A-8 https://huggingface.co/bert-base-uncased https://huggingface.co/bert-large-uncased", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 410.74261474609375, "r": 336.3897705078125, "b": 406.6357421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "BERT$_{MEDIUM}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 405.0371398925781, "r": 330.8688049316406, "b": 400.9302673339844, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "BERT$_{BASE}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.9416809082031, "t": 399.3316650390625, "r": 333.59405517578125, "b": 395.22479248046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "BERT$_{LARGE}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 391.22589111328125, "r": 331.48443603515625, "b": 387.1190490722656, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "GPT-2$_{BASE}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 373.55841064453125, "t": 390.982421875, "r": 495.0466003417969, "b": 387.43603515625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/openai-community/gpt2", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 385.52044677734375, "r": 334.209716796875, "b": 381.41357421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "GPT-2$_{LARGE}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 365.184814453125, "t": 385.2769470214844, "r": 503.41754150390625, "b": 381.73052978515625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/openai-community/gpt2-large", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.9416809082031, "t": 379.81494140625, "r": 327.497314453125, "b": 375.70806884765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 7, "end_row_offset_idx": 8, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "GPT-2$_{XL}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 369.3191833496094, "t": 379.57147216796875, "r": 499.28302001953125, "b": 376.0250549316406, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 7, "end_row_offset_idx": 8, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/openai-community/gpt2-xl", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 371.7087707519531, "r": 327.63726806640625, "b": 367.6018981933594, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 8, "end_row_offset_idx": 9, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "OPT$_{125M}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 379.0710144042969, "t": 371.4653015136719, "r": 489.5340576171875, "b": 367.91888427734375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 8, "end_row_offset_idx": 9, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/facebook/opt-125m", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 366.0032958984375, "r": 326.0531311035156, "b": 361.89642333984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 9, "end_row_offset_idx": 10, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "OPT$_{6.7B}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 379.0710144042969, "t": 365.75982666015625, "r": 489.5340576171875, "b": 362.21337890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 9, "end_row_offset_idx": 10, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/facebook/opt-6.7b", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 360.29779052734375, "r": 325.2140808105469, "b": 356.19091796875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 10, "end_row_offset_idx": 11, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "OPT$_{30B}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 380.44927978515625, "t": 360.0543212890625, "r": 488.1560363769531, "b": 356.5079040527344, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 10, "end_row_offset_idx": 11, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/facebook/opt-30b", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 352.1920471191406, "r": 333.73883056640625, "b": 348.0851745605469, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 11, "end_row_offset_idx": 12, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "LLaMA-2$_{7B}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 373.34869384765625, "t": 351.9485778808594, "r": 495.2549743652344, "b": 348.40216064453125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 11, "end_row_offset_idx": 12, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/meta-llama/Llama-2-7b", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.9416809082031, "t": 346.486572265625, "r": 335.41693115234375, "b": 342.37969970703125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 12, "end_row_offset_idx": 13, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "LLaMA-2$_{13B}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 371.9704284667969, "t": 346.24310302734375, "r": 496.6329650878906, "b": 342.6966552734375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 12, "end_row_offset_idx": 13, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/meta-llama/Llama-2-13b", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 338.38079833984375, "r": 338.0176696777344, "b": 334.27392578125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 13, "end_row_offset_idx": 14, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "SegFormer-B0", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 345.5762939453125, "t": 338.1373291015625, "r": 523.0264282226562, "b": 334.5909118652344, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 13, "end_row_offset_idx": 14, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.9416809082031, "t": 332.6749267578125, "r": 338.0176696777344, "b": 328.56805419921875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 14, "end_row_offset_idx": 15, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "SegFormer-B1", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 345.5762939453125, "t": 332.4314270019531, "r": 523.0264282226562, "b": 328.885009765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 14, "end_row_offset_idx": 15, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/nvidia/segformer-b1-finetuned-ade-512-512", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.9416809082031, "t": 326.96942138671875, "r": 338.0176696777344, "b": 322.862548828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 15, "end_row_offset_idx": 16, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "SegFormer-B2", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 345.5762939453125, "t": 326.7259521484375, "r": 523.0264282226562, "b": 323.17950439453125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 15, "end_row_offset_idx": 16, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/nvidia/segformer-b2-finetuned-ade-512-512", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.9416809082031, "t": 321.263916015625, "r": 338.0176696777344, "b": 317.15704345703125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 16, "end_row_offset_idx": 17, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "SegFormer-B3", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 345.5762939453125, "t": 321.02044677734375, "r": 523.0264282226562, "b": 317.47406005859375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 16, "end_row_offset_idx": 17, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/nvidia/segformer-b3-finetuned-ade-512-512", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.9416809082031, "t": 315.5584716796875, "r": 338.0176696777344, "b": 305.74609375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 17, "end_row_offset_idx": 18, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "SegFormer-B4 SegFormer-B5", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 345.5762939453125, "t": 315.3149719238281, "r": 523.0264282226562, "b": 306.06304931640625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 17, "end_row_offset_idx": 18, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/nvidia/segformer-b4-finetuned-ade-512-512 https://huggingface.co/nvidia/segformer-b5-finetuned-ade-640-640", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 301.7472229003906, "r": 332.33538818359375, "b": 291.9344482421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 18, "end_row_offset_idx": 19, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "MAE$_{BASE}$ MAE$_{LARGE}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 372.07568359375, "t": 301.50372314453125, "r": 496.52691650390625, "b": 292.25140380859375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 18, "end_row_offset_idx": 19, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/facebook/vit-mae-base https://huggingface.co/facebook/vit-mae-large", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 290.3358154296875, "r": 330.46929931640625, "b": 286.22894287109375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 19, "end_row_offset_idx": 20, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "MAE$_{HUGE}$", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 373.45355224609375, "t": 290.09234619140625, "r": 495.1484680175781, "b": 286.5459289550781, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 19, "end_row_offset_idx": 20, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://huggingface.co/facebook/vit-mae-huge", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 282.2300720214844, "r": 329.0596923828125, "b": 278.1231994628906, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 20, "end_row_offset_idx": 21, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "ResNet18", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 20, "end_row_offset_idx": 21, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 276.52459716796875, "r": 329.0596923828125, "b": 272.417724609375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 21, "end_row_offset_idx": 22, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "ResNet34", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 21, "end_row_offset_idx": 22, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 270.819091796875, "r": 329.0596923828125, "b": 266.71221923828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 22, "end_row_offset_idx": 23, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "ResNet50", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 377.69317626953125, "t": 270.57562255859375, "r": 490.9125061035156, "b": 267.0292053222656, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 22, "end_row_offset_idx": 23, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://pypi.org/project/img2vec-pytorch/", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 265.1136169433594, "r": 331.35662841796875, "b": 261.0067443847656, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 23, "end_row_offset_idx": 24, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "ResNet101", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 23, "end_row_offset_idx": 24, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 259.40771484375, "r": 331.35662841796875, "b": 255.30084228515625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 24, "end_row_offset_idx": 25, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "ResNet152", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 24, "end_row_offset_idx": 25, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 251.30197143554688, "r": 333.40087890625, "b": 247.19509887695312, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 25, "end_row_offset_idx": 26, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-RN50", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 25, "end_row_offset_idx": 26, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.941650390625, "t": 245.5964813232422, "r": 340.2915954589844, "b": 235.78411865234375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 26, "end_row_offset_idx": 27, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-RN101 CLIP-RN50*64", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 392.9576416015625, "t": 239.64752197265625, "r": 475.6464538574219, "b": 236.10110473632812, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 26, "end_row_offset_idx": 27, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "https://github.com/openai/CLIP", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 310.9416809082031, "t": 234.18551635742188, "r": 340.3742980957031, "b": 230.07864379882812, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 27, "end_row_offset_idx": 28, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-VIT-B-32", "column_header": false, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 27, "end_row_offset_idx": 28, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "", "column_header": false, "row_header": false, "row_section": false}]]}}

e895876f-ed1f-4dee-b6a9-f8c381e3a053

2302.06555v2.pdf

section_header

B Dispersion Details

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/205", "parent": {"cref": "#/body"}, "children": [], "label": "section_header", "prov": [{"page_no": 16, "bbox": {"l": 306.82098388671875, "t": 170.36370849609375, "r": 419.8581237792969, "b": 158.57666015625, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 20]}], "orig": "B Dispersion Details", "text": "B Dispersion Details", "level": 1}

8c4e241e-2494-45f3-ab19-60fc77f1a40c

2302.06555v2.pdf

section_header

B.1 Image Dispersion

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/206", "parent": {"cref": "#/body"}, "children": [], "label": "section_header", "prov": [{"page_no": 16, "bbox": {"l": 306.6719055175781, "t": 146.79437255859375, "r": 415.15606689453125, "b": 136.0662841796875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 20]}], "orig": "B.1 Image Dispersion", "text": "B.1 Image Dispersion", "level": 1}

0335479c-87ac-4855-90cd-daf4e8ca77a3

2302.06555v2.pdf

text

The image dispersion d of a concept alias a is defined as the average pairwise cosine distance between all the image representations i$_{1}$, i$_{2}$...i$_{n}$ in the set of n images for a given alias (Kiela et al.,

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/207", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 16, "bbox": {"l": 306.2406921386719, "t": 127.46954345703125, "r": 527.3583984375, "b": 75.8514404296875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 215]}], "orig": "The image dispersion d of a concept alias a is defined as the average pairwise cosine distance between all the image representations i$_{1}$, i$_{2}$...i$_{n}$ in the set of n images for a given alias (Kiela et al.,", "text": "The image dispersion d of a concept alias a is defined as the average pairwise cosine distance between all the image representations i$_{1}$, i$_{2}$...i$_{n}$ in the set of n images for a given alias (Kiela et al.,"}

650a2730-51c7-412d-a473-4a58f4a6ad31

2302.06555v2.pdf

caption

Table 8: The cumulative of explained variance ratios for different models and sizes.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/208", "parent": {"cref": "#/body"}, "children": [], "label": "caption", "prov": [{"page_no": 17, "bbox": {"l": 71.0458984375, "t": 701.2876586914062, "r": 290.2685241699219, "b": 677.588623046875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 84]}], "orig": "Table 8: The cumulative of explained variance ratios for different models and sizes.", "text": "Table 8: The cumulative of explained variance ratios for different models and sizes."}

68880cdd-02ce-461a-a416-0704b0d907e1

2302.06555v2.pdf

table

Model	Explained Variance Ratio (Sum)
	256	512	768	1024	1280	2048	Max
MAE$_{Huge}$	0.9735	0.9922	0.9975	0.9994	1.0000	-	1.0000
ResNet152	0.9795	0.9942	0.9974	0.9987	0.9993	1.0000	1.0000
SegFormer-B5	0.9685	1.0000	-	-	-	-	1.0000
LLaMA-2$_{13B}$	0.5708	0.6662	0.7277	0.7725	0.8077	0.8814	1.0000
OPT$_{30B}$	0.4926	0.6002	0.6664	0.7164	0.7554	0.8360	1.0000

Table 8: The cumulative of explained variance ratios for different models and sizes.

image/png

application/pdf

1.0.0

{"self_ref": "#/tables/7", "parent": {"cref": "#/body"}, "children": [], "label": "table", "prov": [{"page_no": 17, "bbox": {"l": 72.62456512451172, "t": 779.7711181640625, "r": 290.5558166503906, "b": 712.1237182617188, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 0]}], "captions": [{"cref": "#/texts/208"}], "references": [], "footnotes": [], "image": {"mimetype": "image/png", "dpi": 144, "size": {"width": 436.0, "height": 136.0}, "uri": null}, "data": {"table_cells": [{"bbox": {"l": 77.09851837158203, "t": 771.510986328125, "r": 93.7813949584961, "b": 765.798828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "Model", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 162.82977294921875, "t": 775.478515625, "r": 245.70103454589844, "b": 769.766357421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 7, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 1, "end_col_offset_idx": 8, "text": "Explained Variance Ratio (Sum)", "column_header": true, "row_header": false, "row_section": false}, {"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 125.7531967163086, "t": 764.3327026367188, "r": 135.33738708496094, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "256", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 150.32705688476562, "t": 764.3327026367188, "r": 159.9112548828125, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "512", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 174.9009246826172, "t": 764.3327026367188, "r": 184.485107421875, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "768", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 197.87741088867188, "t": 764.3327026367188, "r": 210.65634155273438, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "1024", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 222.45127868652344, "t": 764.3327026367188, "r": 235.23019409179688, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "1280", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 247.0187530517578, "t": 764.3327026367188, "r": 259.79766845703125, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 6, "end_col_offset_idx": 7, "text": "2048", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 272.12933349609375, "t": 764.3327026367188, "r": 283.8411865234375, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 7, "end_col_offset_idx": 8, "text": "Max", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 77.09851837158203, "t": 753.0579833984375, "r": 101.40696716308594, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "MAE$_{Huge}$", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 121.75990295410156, "t": 753.0579833984375, "r": 139.33091735839844, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "0.9735", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 146.33377075195312, "t": 753.0579833984375, "r": 163.90478515625, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "0.9922", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 170.90762329101562, "t": 753.0579833984375, "r": 188.4786376953125, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "0.9975", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 195.48147583007812, "t": 753.0579833984375, "r": 213.052490234375, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "0.9994", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 220.05532836914062, "t": 753.0579833984375, "r": 237.62635803222656, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 252.34765625, "t": 753.0579833984375, "r": 254.475341796875, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 6, "end_col_offset_idx": 7, "text": "-", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 269.1966552734375, "t": 753.0579833984375, "r": 286.76763916015625, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 7, "end_col_offset_idx": 8, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 77.09852600097656, "t": 745.122314453125, "r": 105.49327087402344, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "ResNet152", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 121.76083374023438, "t": 745.122314453125, "r": 139.33184814453125, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "0.9795", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 146.33470153808594, "t": 745.122314453125, "r": 163.9057159423828, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "0.9942", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 170.90855407714844, "t": 745.122314453125, "r": 188.4795684814453, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "0.9974", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 195.48240661621094, "t": 745.122314453125, "r": 213.0534210205078, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "0.9987", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 220.0498809814453, "t": 745.122314453125, "r": 237.6208953857422, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "0.9993", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 244.6237335205078, "t": 745.122314453125, "r": 262.19476318359375, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 6, "end_col_offset_idx": 7, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 269.1976013183594, "t": 745.122314453125, "r": 286.7685852050781, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 7, "end_col_offset_idx": 8, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 77.09852600097656, "t": 737.1866455078125, "r": 114.75800323486328, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "SegFormer-B5", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 121.76083374023438, "t": 737.1866455078125, "r": 139.33184814453125, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "0.9685", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 146.33470153808594, "t": 737.1866455078125, "r": 163.9057159423828, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 178.6270294189453, "t": 737.1866455078125, "r": 180.7547149658203, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "-", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 203.20089721679688, "t": 737.1866455078125, "r": 205.32858276367188, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "-", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 227.77474975585938, "t": 737.1866455078125, "r": 229.90243530273438, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "-", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 252.34860229492188, "t": 737.1866455078125, "r": 254.47628784179688, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 6, "end_col_offset_idx": 7, "text": "-", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 269.1976013183594, "t": 737.1866455078125, "r": 286.7685852050781, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 7, "end_col_offset_idx": 8, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 77.09852600097656, "t": 729.2509765625, "r": 111.14067077636719, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "LLaMA-2$_{13B}$", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 121.75990295410156, "t": 729.2509765625, "r": 139.33091735839844, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "0.5708", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 146.33377075195312, "t": 729.2509765625, "r": 163.90478515625, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "0.6662", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 170.90762329101562, "t": 729.2509765625, "r": 188.4786376953125, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "0.7277", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 195.48147583007812, "t": 729.2509765625, "r": 213.052490234375, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "0.7725", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 220.05532836914062, "t": 729.2509765625, "r": 237.62635803222656, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "0.8077", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 244.622802734375, "t": 729.2509765625, "r": 262.1938171386719, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 6, "end_col_offset_idx": 7, "text": "0.8814", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 269.1966552734375, "t": 729.2509765625, "r": 286.7676696777344, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 7, "end_col_offset_idx": 8, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 77.09852600097656, "t": 721.3153076171875, "r": 96.94973754882812, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "OPT$_{30B}$", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 121.75990295410156, "t": 721.3153076171875, "r": 139.33091735839844, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "0.4926", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 146.33377075195312, "t": 721.3153076171875, "r": 163.90478515625, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "0.6002", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 170.90762329101562, "t": 721.3153076171875, "r": 188.4786376953125, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "0.6664", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 195.48147583007812, "t": 721.3153076171875, "r": 213.052490234375, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "0.7164", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 220.05532836914062, "t": 721.3153076171875, "r": 237.62635803222656, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "0.7554", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 244.622802734375, "t": 721.3153076171875, "r": 262.1938171386719, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 6, "end_col_offset_idx": 7, "text": "0.8360", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 269.1966552734375, "t": 721.3153076171875, "r": 286.7676696777344, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 7, "end_col_offset_idx": 8, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}], "num_rows": 7, "num_cols": 8, "grid": [[{"bbox": {"l": 77.09851837158203, "t": 771.510986328125, "r": 93.7813949584961, "b": 765.798828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "Model", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 162.82977294921875, "t": 775.478515625, "r": 245.70103454589844, "b": 769.766357421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 7, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 1, "end_col_offset_idx": 8, "text": "Explained Variance Ratio (Sum)", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 162.82977294921875, "t": 775.478515625, "r": 245.70103454589844, "b": 769.766357421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 7, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 1, "end_col_offset_idx": 8, "text": "Explained Variance Ratio (Sum)", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 162.82977294921875, "t": 775.478515625, "r": 245.70103454589844, "b": 769.766357421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 7, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 1, "end_col_offset_idx": 8, "text": "Explained Variance Ratio (Sum)", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 162.82977294921875, "t": 775.478515625, "r": 245.70103454589844, "b": 769.766357421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 7, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 1, "end_col_offset_idx": 8, "text": "Explained Variance Ratio (Sum)", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 162.82977294921875, "t": 775.478515625, "r": 245.70103454589844, "b": 769.766357421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 7, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 1, "end_col_offset_idx": 8, "text": "Explained Variance Ratio (Sum)", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 162.82977294921875, "t": 775.478515625, "r": 245.70103454589844, "b": 769.766357421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 7, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 1, "end_col_offset_idx": 8, "text": "Explained Variance Ratio (Sum)", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 162.82977294921875, "t": 775.478515625, "r": 245.70103454589844, "b": 769.766357421875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 7, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 1, "end_col_offset_idx": 8, "text": "Explained Variance Ratio (Sum)", "column_header": true, "row_header": false, "row_section": false}], [{"bbox": null, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 125.7531967163086, "t": 764.3327026367188, "r": 135.33738708496094, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "256", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 150.32705688476562, "t": 764.3327026367188, "r": 159.9112548828125, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "512", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 174.9009246826172, "t": 764.3327026367188, "r": 184.485107421875, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "768", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 197.87741088867188, "t": 764.3327026367188, "r": 210.65634155273438, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "1024", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 222.45127868652344, "t": 764.3327026367188, "r": 235.23019409179688, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "1280", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 247.0187530517578, "t": 764.3327026367188, "r": 259.79766845703125, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 6, "end_col_offset_idx": 7, "text": "2048", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 272.12933349609375, "t": 764.3327026367188, "r": 283.8411865234375, "b": 758.6204833984375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 7, "end_col_offset_idx": 8, "text": "Max", "column_header": true, "row_header": false, "row_section": false}], [{"bbox": {"l": 77.09851837158203, "t": 753.0579833984375, "r": 101.40696716308594, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "MAE$_{Huge}$", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 121.75990295410156, "t": 753.0579833984375, "r": 139.33091735839844, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "0.9735", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 146.33377075195312, "t": 753.0579833984375, "r": 163.90478515625, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "0.9922", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 170.90762329101562, "t": 753.0579833984375, "r": 188.4786376953125, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "0.9975", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 195.48147583007812, "t": 753.0579833984375, "r": 213.052490234375, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "0.9994", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 220.05532836914062, "t": 753.0579833984375, "r": 237.62635803222656, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 252.34765625, "t": 753.0579833984375, "r": 254.475341796875, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 6, "end_col_offset_idx": 7, "text": "-", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 269.1966552734375, "t": 753.0579833984375, "r": 286.76763916015625, "b": 747.3458251953125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 7, "end_col_offset_idx": 8, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 77.09852600097656, "t": 745.122314453125, "r": 105.49327087402344, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "ResNet152", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 121.76083374023438, "t": 745.122314453125, "r": 139.33184814453125, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "0.9795", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 146.33470153808594, "t": 745.122314453125, "r": 163.9057159423828, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "0.9942", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 170.90855407714844, "t": 745.122314453125, "r": 188.4795684814453, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "0.9974", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 195.48240661621094, "t": 745.122314453125, "r": 213.0534210205078, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "0.9987", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 220.0498809814453, "t": 745.122314453125, "r": 237.6208953857422, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "0.9993", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 244.6237335205078, "t": 745.122314453125, "r": 262.19476318359375, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 6, "end_col_offset_idx": 7, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 269.1976013183594, "t": 745.122314453125, "r": 286.7685852050781, "b": 739.41015625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 7, "end_col_offset_idx": 8, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 77.09852600097656, "t": 737.1866455078125, "r": 114.75800323486328, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "SegFormer-B5", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 121.76083374023438, "t": 737.1866455078125, "r": 139.33184814453125, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "0.9685", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 146.33470153808594, "t": 737.1866455078125, "r": 163.9057159423828, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 178.6270294189453, "t": 737.1866455078125, "r": 180.7547149658203, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "-", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 203.20089721679688, "t": 737.1866455078125, "r": 205.32858276367188, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "-", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 227.77474975585938, "t": 737.1866455078125, "r": 229.90243530273438, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "-", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 252.34860229492188, "t": 737.1866455078125, "r": 254.47628784179688, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 6, "end_col_offset_idx": 7, "text": "-", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 269.1976013183594, "t": 737.1866455078125, "r": 286.7685852050781, "b": 731.4744873046875, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 7, "end_col_offset_idx": 8, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 77.09852600097656, "t": 729.2509765625, "r": 111.14067077636719, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "LLaMA-2$_{13B}$", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 121.75990295410156, "t": 729.2509765625, "r": 139.33091735839844, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "0.5708", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 146.33377075195312, "t": 729.2509765625, "r": 163.90478515625, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "0.6662", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 170.90762329101562, "t": 729.2509765625, "r": 188.4786376953125, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "0.7277", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 195.48147583007812, "t": 729.2509765625, "r": 213.052490234375, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "0.7725", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 220.05532836914062, "t": 729.2509765625, "r": 237.62635803222656, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "0.8077", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 244.622802734375, "t": 729.2509765625, "r": 262.1938171386719, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 6, "end_col_offset_idx": 7, "text": "0.8814", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 269.1966552734375, "t": 729.2509765625, "r": 286.7676696777344, "b": 723.538818359375, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 7, "end_col_offset_idx": 8, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 77.09852600097656, "t": 721.3153076171875, "r": 96.94973754882812, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "OPT$_{30B}$", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 121.75990295410156, "t": 721.3153076171875, "r": 139.33091735839844, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "0.4926", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 146.33377075195312, "t": 721.3153076171875, "r": 163.90478515625, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "0.6002", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 170.90762329101562, "t": 721.3153076171875, "r": 188.4786376953125, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "0.6664", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 195.48147583007812, "t": 721.3153076171875, "r": 213.052490234375, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "0.7164", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 220.05532836914062, "t": 721.3153076171875, "r": 237.62635803222656, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "0.7554", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 244.622802734375, "t": 721.3153076171875, "r": 262.1938171386719, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 6, "end_col_offset_idx": 7, "text": "0.8360", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 269.1966552734375, "t": 721.3153076171875, "r": 286.7676696777344, "b": 715.6031494140625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 7, "end_col_offset_idx": 8, "text": "1.0000", "column_header": false, "row_header": false, "row_section": false}]]}}

2911d964-5fb7-416b-a46b-8c4364f9de45

2302.06555v2.pdf

caption

Table 9: Evaluation of POS impact on OPT$_{30B}$ and different CLIP models using EN-CLDI set. "Mix" denotes a combination of all POS categories.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/209", "parent": {"cref": "#/body"}, "children": [], "label": "caption", "prov": [{"page_no": 17, "bbox": {"l": 70.96319580078125, "t": 562.115234375, "r": 290.2682189941406, "b": 524.6176147460938, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 144]}], "orig": "Table 9: Evaluation of POS impact on OPT$_{30B}$ and different CLIP models using EN-CLDI set. \"Mix\" denotes a combination of all POS categories.", "text": "Table 9: Evaluation of POS impact on OPT$_{30B}$ and different CLIP models using EN-CLDI set. \"Mix\" denotes a combination of all POS categories."}

363e8eaa-884e-4e93-901e-40ff23cf1b4a

2302.06555v2.pdf

table

Models	Train	Test	P@1	P@10	P@100
CLIP-ViT-L	Noun	Adj.	12.7	52.2	85.4
CLIP-RN50x64	1337	157	7.0	45.2	84.7
CLIP-ViT-L	Noun	Verb.	12.2	55.1	93.9
CLIP-RN50x64	1337	196	9.2	46.4	89.3
CLIP-ViT-L	Mix	Mix.	39.1	81.0	94.1
CLIP-RN50x64	1337	353	33.7	79.9	93.8

Table 9: Evaluation of POS impact on OPT$_{30B}$ and different CLIP models using EN-CLDI set. "Mix" denotes a combination of all POS categories.

image/png

application/pdf

1.0.0

{"self_ref": "#/tables/8", "parent": {"cref": "#/body"}, "children": [], "label": "table", "prov": [{"page_no": 17, "bbox": {"l": 78.54300689697266, "t": 664.9376831054688, "r": 285.4424743652344, "b": 572.4727783203125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 0]}], "captions": [{"cref": "#/texts/209"}], "references": [], "footnotes": [], "image": {"mimetype": "image/png", "dpi": 144, "size": {"width": 414.0, "height": 185.0}, "uri": null}, "data": {"table_cells": [{"bbox": {"l": 84.09164428710938, "t": 659.4486083984375, "r": 109.04653930664062, "b": 652.011962890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "Models", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 146.65635681152344, "t": 659.4486083984375, "r": 164.38265991210938, "b": 652.011962890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "Train", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 173.49951171875, "t": 659.4486083984375, "r": 187.24134826660156, "b": 652.011962890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "Test", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 200.21047973632812, "t": 659.4486083984375, "r": 216.65576171875, "b": 652.011962890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "P@1", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 225.77259826660156, "t": 659.4486083984375, "r": 246.3770294189453, "b": 652.011962890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "P@10", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 255.49388122558594, "t": 659.4486083984375, "r": 280.2574462890625, "b": 652.011962890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "P@100", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 84.09164428710938, "t": 644.77099609375, "r": 124.87626647949219, "b": 637.3344116210938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-ViT-L", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 146.65635681152344, "t": 644.77099609375, "r": 164.7699432373047, "b": 637.3344116210938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "Noun", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 173.49899291992188, "t": 644.77099609375, "r": 188.05601501464844, "b": 637.3344116210938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "Adj.", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 202.09844970703125, "t": 644.870849609375, "r": 216.6554718017578, "b": 637.3926391601562, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "12.7", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 231.8197479248047, "t": 644.870849609375, "r": 246.3767547607422, "b": 637.3926391601562, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "52.2", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 265.7001647949219, "t": 644.870849609375, "r": 280.2571716308594, "b": 637.3926391601562, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "85.4", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 84.09164428710938, "t": 634.439697265625, "r": 137.2372589111328, "b": 627.003173828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-RN50x64", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 146.03262329101562, "t": 634.439697265625, "r": 162.66921997070312, "b": 627.003173828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "1337", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 172.87449645996094, "t": 634.439697265625, "r": 185.35194396972656, "b": 627.003173828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "157", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 206.25718688964844, "t": 634.439697265625, "r": 216.65505981445312, "b": 627.003173828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "7.0", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 231.81932067871094, "t": 634.439697265625, "r": 246.37632751464844, "b": 627.003173828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "45.2", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 265.6997375488281, "t": 634.439697265625, "r": 280.2567443847656, "b": 627.003173828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "84.7", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 84.09164428710938, "t": 619.76220703125, "r": 124.87626647949219, "b": 612.3256225585938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-ViT-L", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 146.65635681152344, "t": 619.76220703125, "r": 164.7699432373047, "b": 612.3256225585938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "Noun", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 173.49899291992188, "t": 619.76220703125, "r": 190.95079040527344, "b": 612.3256225585938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "Verb.", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 202.09844970703125, "t": 619.862060546875, "r": 216.6554718017578, "b": 612.3838500976562, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "12.2", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 231.8197479248047, "t": 619.862060546875, "r": 246.3767547607422, "b": 612.3838500976562, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "55.1", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 265.7001647949219, "t": 619.862060546875, "r": 280.2571716308594, "b": 612.3838500976562, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "93.9", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 84.09164428710938, "t": 609.430908203125, "r": 137.2372589111328, "b": 601.994384765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-RN50x64", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 146.03262329101562, "t": 609.430908203125, "r": 162.66921997070312, "b": 601.994384765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "1337", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 172.87449645996094, "t": 609.430908203125, "r": 185.35194396972656, "b": 601.994384765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "196", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 206.25718688964844, "t": 609.430908203125, "r": 216.65505981445312, "b": 601.994384765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "9.2", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 231.81932067871094, "t": 609.430908203125, "r": 246.37632751464844, "b": 601.994384765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "46.4", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 265.6997375488281, "t": 609.430908203125, "r": 280.2567443847656, "b": 601.994384765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "89.3", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 84.09164428710938, "t": 594.7526245117188, "r": 124.87626647949219, "b": 587.3160400390625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-ViT-L", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 146.65635681152344, "t": 594.7526245117188, "r": 160.52296447753906, "b": 587.3160400390625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "Mix", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 173.49900817871094, "t": 594.7526245117188, "r": 189.44517517089844, "b": 587.3160400390625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "Mix.", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 202.09844970703125, "t": 594.8524169921875, "r": 216.6554718017578, "b": 587.374267578125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "39.1", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 231.8197479248047, "t": 594.8524169921875, "r": 246.3767547607422, "b": 587.374267578125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "81.0", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 265.7001647949219, "t": 594.8524169921875, "r": 280.2571716308594, "b": 587.374267578125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "94.1", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 84.09164428710938, "t": 584.42138671875, "r": 137.2372589111328, "b": 576.9848022460938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-RN50x64", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 146.0326385498047, "t": 584.42138671875, "r": 162.66921997070312, "b": 576.9848022460938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "1337", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 173.49900817871094, "t": 584.42138671875, "r": 185.9764404296875, "b": 576.9848022460938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "353", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 202.09844970703125, "t": 584.42138671875, "r": 216.6554718017578, "b": 576.9848022460938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "33.7", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 231.8197479248047, "t": 584.42138671875, "r": 246.3767547607422, "b": 576.9848022460938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "79.9", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 265.7001647949219, "t": 584.42138671875, "r": 280.2571716308594, "b": 576.9848022460938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "93.8", "column_header": false, "row_header": false, "row_section": false}], "num_rows": 7, "num_cols": 6, "grid": [[{"bbox": {"l": 84.09164428710938, "t": 659.4486083984375, "r": 109.04653930664062, "b": 652.011962890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "Models", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 146.65635681152344, "t": 659.4486083984375, "r": 164.38265991210938, "b": 652.011962890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "Train", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 173.49951171875, "t": 659.4486083984375, "r": 187.24134826660156, "b": 652.011962890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "Test", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 200.21047973632812, "t": 659.4486083984375, "r": 216.65576171875, "b": 652.011962890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "P@1", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 225.77259826660156, "t": 659.4486083984375, "r": 246.3770294189453, "b": 652.011962890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "P@10", "column_header": true, "row_header": false, "row_section": false}, {"bbox": {"l": 255.49388122558594, "t": 659.4486083984375, "r": 280.2574462890625, "b": 652.011962890625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 0, "end_row_offset_idx": 1, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "P@100", "column_header": true, "row_header": false, "row_section": false}], [{"bbox": {"l": 84.09164428710938, "t": 644.77099609375, "r": 124.87626647949219, "b": 637.3344116210938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-ViT-L", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 146.65635681152344, "t": 644.77099609375, "r": 164.7699432373047, "b": 637.3344116210938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "Noun", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 173.49899291992188, "t": 644.77099609375, "r": 188.05601501464844, "b": 637.3344116210938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "Adj.", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 202.09844970703125, "t": 644.870849609375, "r": 216.6554718017578, "b": 637.3926391601562, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "12.7", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 231.8197479248047, "t": 644.870849609375, "r": 246.3767547607422, "b": 637.3926391601562, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "52.2", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 265.7001647949219, "t": 644.870849609375, "r": 280.2571716308594, "b": 637.3926391601562, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 1, "end_row_offset_idx": 2, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "85.4", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 84.09164428710938, "t": 634.439697265625, "r": 137.2372589111328, "b": 627.003173828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-RN50x64", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 146.03262329101562, "t": 634.439697265625, "r": 162.66921997070312, "b": 627.003173828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "1337", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 172.87449645996094, "t": 634.439697265625, "r": 185.35194396972656, "b": 627.003173828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "157", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 206.25718688964844, "t": 634.439697265625, "r": 216.65505981445312, "b": 627.003173828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "7.0", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 231.81932067871094, "t": 634.439697265625, "r": 246.37632751464844, "b": 627.003173828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "45.2", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 265.6997375488281, "t": 634.439697265625, "r": 280.2567443847656, "b": 627.003173828125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 2, "end_row_offset_idx": 3, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "84.7", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 84.09164428710938, "t": 619.76220703125, "r": 124.87626647949219, "b": 612.3256225585938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-ViT-L", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 146.65635681152344, "t": 619.76220703125, "r": 164.7699432373047, "b": 612.3256225585938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "Noun", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 173.49899291992188, "t": 619.76220703125, "r": 190.95079040527344, "b": 612.3256225585938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "Verb.", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 202.09844970703125, "t": 619.862060546875, "r": 216.6554718017578, "b": 612.3838500976562, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "12.2", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 231.8197479248047, "t": 619.862060546875, "r": 246.3767547607422, "b": 612.3838500976562, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "55.1", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 265.7001647949219, "t": 619.862060546875, "r": 280.2571716308594, "b": 612.3838500976562, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 3, "end_row_offset_idx": 4, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "93.9", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 84.09164428710938, "t": 609.430908203125, "r": 137.2372589111328, "b": 601.994384765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-RN50x64", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 146.03262329101562, "t": 609.430908203125, "r": 162.66921997070312, "b": 601.994384765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "1337", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 172.87449645996094, "t": 609.430908203125, "r": 185.35194396972656, "b": 601.994384765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "196", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 206.25718688964844, "t": 609.430908203125, "r": 216.65505981445312, "b": 601.994384765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "9.2", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 231.81932067871094, "t": 609.430908203125, "r": 246.37632751464844, "b": 601.994384765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "46.4", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 265.6997375488281, "t": 609.430908203125, "r": 280.2567443847656, "b": 601.994384765625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 4, "end_row_offset_idx": 5, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "89.3", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 84.09164428710938, "t": 594.7526245117188, "r": 124.87626647949219, "b": 587.3160400390625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-ViT-L", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 146.65635681152344, "t": 594.7526245117188, "r": 160.52296447753906, "b": 587.3160400390625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "Mix", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 173.49900817871094, "t": 594.7526245117188, "r": 189.44517517089844, "b": 587.3160400390625, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "Mix.", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 202.09844970703125, "t": 594.8524169921875, "r": 216.6554718017578, "b": 587.374267578125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "39.1", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 231.8197479248047, "t": 594.8524169921875, "r": 246.3767547607422, "b": 587.374267578125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "81.0", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 265.7001647949219, "t": 594.8524169921875, "r": 280.2571716308594, "b": 587.374267578125, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 5, "end_row_offset_idx": 6, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "94.1", "column_header": false, "row_header": false, "row_section": false}], [{"bbox": {"l": 84.09164428710938, "t": 584.42138671875, "r": 137.2372589111328, "b": 576.9848022460938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 0, "end_col_offset_idx": 1, "text": "CLIP-RN50x64", "column_header": false, "row_header": true, "row_section": false}, {"bbox": {"l": 146.0326385498047, "t": 584.42138671875, "r": 162.66921997070312, "b": 576.9848022460938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 1, "end_col_offset_idx": 2, "text": "1337", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 173.49900817871094, "t": 584.42138671875, "r": 185.9764404296875, "b": 576.9848022460938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 2, "end_col_offset_idx": 3, "text": "353", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 202.09844970703125, "t": 584.42138671875, "r": 216.6554718017578, "b": 576.9848022460938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 3, "end_col_offset_idx": 4, "text": "33.7", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 231.8197479248047, "t": 584.42138671875, "r": 246.3767547607422, "b": 576.9848022460938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 4, "end_col_offset_idx": 5, "text": "79.9", "column_header": false, "row_header": false, "row_section": false}, {"bbox": {"l": 265.7001647949219, "t": 584.42138671875, "r": 280.2571716308594, "b": 576.9848022460938, "coord_origin": "BOTTOMLEFT"}, "row_span": 1, "col_span": 1, "start_row_offset_idx": 6, "end_row_offset_idx": 7, "start_col_offset_idx": 5, "end_col_offset_idx": 6, "text": "93.8", "column_header": false, "row_header": false, "row_section": false}]]}}

e30d4f0b-48c9-47d0-b5e9-d7c3621543b6

2302.06555v2.pdf

text

2015):

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/210", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 17, "bbox": {"l": 71.56356811523438, "t": 500.7362060546875, "r": 100.48365783691406, "b": 490.0966491699219, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 6]}], "orig": "2015):", "text": "2015):"}

f914a2ae-988b-4f2b-ba11-3d4d5e7734db

2302.06555v2.pdf

formula

d ( a ) = 2 n ( n - 1) ∑ k<j ≤ n 1 - i$_{j}$ · i$_{k}$ | i$_{j}$ || i$_{k}$ |

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/211", "parent": {"cref": "#/body"}, "children": [], "label": "formula", "prov": [{"page_no": 17, "bbox": {"l": 101.1298599243164, "t": 480.173828125, "r": 258.9724426269531, "b": 449.6528015136719, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 77]}], "orig": "d ( a ) = 2 n ( n - 1) \u2211 k<j \u2264 n 1 - i$_{j}$ \u00b7 i$_{k}$ | i$_{j}$ || i$_{k}$ |", "text": "d ( a ) = 2 n ( n - 1) \u2211 k<j \u2264 n 1 - i$_{j}$ \u00b7 i$_{k}$ | i$_{j}$ || i$_{k}$ |"}

5de5fea0-60d0-4c9e-bade-68e854c61177

2302.06555v2.pdf

section_header

B.2 Language Dispersion

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/212", "parent": {"cref": "#/body"}, "children": [], "label": "section_header", "prov": [{"page_no": 17, "bbox": {"l": 71.37012481689453, "t": 438.3519592285156, "r": 196.8655548095703, "b": 427.3279724121094, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 23]}], "orig": "B.2 Language Dispersion", "text": "B.2 Language Dispersion", "level": 1}

72c4b925-088f-49e8-bd4b-074758cd8349

2302.06555v2.pdf

text

The language dispersion d of a concept alias a is defined as the average pairwise cosine distance between all the corresponding word representations w$_{1}$, w$_{2}$...w$_{n}$ in the set of n sentences for a given alias:

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/213", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 17, "bbox": {"l": 71.2419662475586, "t": 420.04296875, "r": 290.2678527832031, "b": 355.6986389160156, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 220]}], "orig": "The language dispersion d of a concept alias a is defined as the average pairwise cosine distance between all the corresponding word representations w$_{1}$, w$_{2}$...w$_{n}$ in the set of n sentences for a given alias:", "text": "The language dispersion d of a concept alias a is defined as the average pairwise cosine distance between all the corresponding word representations w$_{1}$, w$_{2}$...w$_{n}$ in the set of n sentences for a given alias:"}

fbf92b92-9314-483e-80a9-29e9f8d163e0

2302.06555v2.pdf

formula

d ( a ) = 2 n ( n - 1) ∑ k<j ≤ n 1 - w$_{j}$ · w$_{k}$ | w$_{j}$ || w$_{k}$ |

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/214", "parent": {"cref": "#/body"}, "children": [], "label": "formula", "prov": [{"page_no": 17, "bbox": {"l": 97.32268524169922, "t": 347.8448181152344, "r": 263.3014831542969, "b": 316.15997314453125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 77]}], "orig": "d ( a ) = 2 n ( n - 1) \u2211 k<j \u2264 n 1 - w$_{j}$ \u00b7 w$_{k}$ | w$_{j}$ || w$_{k}$ |", "text": "d ( a ) = 2 n ( n - 1) \u2211 k<j \u2264 n 1 - w$_{j}$ \u00b7 w$_{k}$ | w$_{j}$ || w$_{k}$ |"}

573f0c41-b9c7-49a3-9471-51991b8ec111

2302.06555v2.pdf

section_header

C More Results

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/215", "parent": {"cref": "#/body"}, "children": [], "label": "section_header", "prov": [{"page_no": 17, "bbox": {"l": 71.76544189453125, "t": 305.74468994140625, "r": 160.4810333251953, "b": 294.17938232421875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 14]}], "orig": "C More Results", "text": "C More Results", "level": 1}

8d5a5666-a319-4d6e-96ce-f284bd4456eb

2302.06555v2.pdf

text

Cumulative percentage of variance explained. In Table 8, we present the cumulative percentage of variance explained by each selected component after PCA.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/216", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 17, "bbox": {"l": 71.33179473876953, "t": 284.09747314453125, "r": 292.175537109375, "b": 232.8676300048828, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 153]}], "orig": "Cumulative percentage of variance explained. In Table 8, we present the cumulative percentage of variance explained by each selected component after PCA.", "text": "Cumulative percentage of variance explained. In Table 8, we present the cumulative percentage of variance explained by each selected component after PCA."}

ff749416-42e2-4a87-9843-56853ee3018a

2302.06555v2.pdf

text

CLIP Results. We also investigate the effects of incorporating text signals during vision pretraining by comparing pure vision models against selected CLIP (Radford et al., 2021) vision encoders (ResNet50, ResNet101, ResNet50x60, ViTBase-Patch32, and ViT-Large-Patch14). The results align with our expectations, indicating that the CLIP vision encoders exhibit better alignment with LMs. The findings also support our previous observation that larger LMs tend to demonstrate better alignment. However, it would be unfair to

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/217", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 17, "bbox": {"l": 71.11685943603516, "t": 222.97711181640625, "r": 292.077880859375, "b": 76.12005615234375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 523]}], "orig": "CLIP Results. We also investigate the effects of incorporating text signals during vision pretraining by comparing pure vision models against selected CLIP (Radford et al., 2021) vision encoders (ResNet50, ResNet101, ResNet50x60, ViTBase-Patch32, and ViT-Large-Patch14). The results align with our expectations, indicating that the CLIP vision encoders exhibit better alignment with LMs. The findings also support our previous observation that larger LMs tend to demonstrate better alignment. However, it would be unfair to", "text": "CLIP Results. We also investigate the effects of incorporating text signals during vision pretraining by comparing pure vision models against selected CLIP (Radford et al., 2021) vision encoders (ResNet50, ResNet101, ResNet50x60, ViTBase-Patch32, and ViT-Large-Patch14). The results align with our expectations, indicating that the CLIP vision encoders exhibit better alignment with LMs. The findings also support our previous observation that larger LMs tend to demonstrate better alignment. However, it would be unfair to"}

807b210d-f163-47ed-bbd5-9f2d5e0cb2be

2302.06555v2.pdf

caption

Figure 9: Illustrating the impact of scaling CLIP models up on Exclude-1K set. The incremental growth in P@100 for scaled-up CLIP models is marginal, contrasting with the more substantial increase observed when scaling up LMs in the same family.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/218", "parent": {"cref": "#/body"}, "children": [], "label": "caption", "prov": [{"page_no": 17, "bbox": {"l": 306.6195983886719, "t": 421.1622009277344, "r": 527.356201171875, "b": 342.441650390625, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 245]}], "orig": "Figure 9: Illustrating the impact of scaling CLIP models up on Exclude-1K set. The incremental growth in P@100 for scaled-up CLIP models is marginal, contrasting with the more substantial increase observed when scaling up LMs in the same family.", "text": "Figure 9: Illustrating the impact of scaling CLIP models up on Exclude-1K set. The incremental growth in P@100 for scaled-up CLIP models is marginal, contrasting with the more substantial increase observed when scaling up LMs in the same family."}

5962a406-fd03-46b6-8e01-61b889bd56df

2302.06555v2.pdf

picture

Figure 9: Illustrating the impact of scaling CLIP models up on Exclude-1K set. The incremental growth in P@100 for scaled-up CLIP models is marginal, contrasting with the more substantial increase observed when scaling up LMs in the same family.

image/png

application/pdf

1.0.0

{"self_ref": "#/pictures/23", "parent": {"cref": "#/body"}, "children": [], "label": "picture", "prov": [{"page_no": 17, "bbox": {"l": 317.4569091796875, "t": 768.2399291992188, "r": 512.4075927734375, "b": 609.8652954101562, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 245]}], "captions": [{"cref": "#/texts/218"}], "references": [], "footnotes": [], "image": {"mimetype": "image/png", "dpi": 144, "size": {"width": 390.0, "height": 317.0}, "uri": null}, "annotations": []}

c7df0093-309d-4f72-9a96-5db7424615be

2302.06555v2.pdf

picture

image/png

application/pdf

1.0.0

{"self_ref": "#/pictures/24", "parent": {"cref": "#/body"}, "children": [], "label": "picture", "prov": [{"page_no": 17, "bbox": {"l": 317.1781921386719, "t": 595.3639526367188, "r": 512.4075927734375, "b": 449.40948486328125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 0]}], "captions": [], "references": [], "footnotes": [], "image": {"mimetype": "image/png", "dpi": 144, "size": {"width": 391.0, "height": 292.0}, "uri": null}, "annotations": []}

beeadb91-d612-4de4-aee0-43528f793782

2302.06555v2.pdf

text

CLIP-VIT-L-14

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/219", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 17, "bbox": {"l": 333.468994140625, "t": 444.474609375, "r": 383.1658935546875, "b": 436.07421875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 13]}], "orig": "CLIP-VIT-L-14", "text": "CLIP-VIT-L-14"}

9a4cc1b1-57b7-4df3-9871-7b554b97b92b

2302.06555v2.pdf

text

CLIP-VIT-B-32

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/220", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 17, "bbox": {"l": 390.9438781738281, "t": 444.474609375, "r": 441.5317687988281, "b": 436.07421875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 13]}], "orig": "CLIP-VIT-B-32", "text": "CLIP-VIT-B-32"}

75fda726-3abb-4770-9039-0c4f9ebb8fef

2302.06555v2.pdf

text

directly compare the results from CLIP with pure vision models, as the pretraining datasets they utilize differ significantly in scale and scope. Detailed results are presented in Figure 9 and Figure 10.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/221", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 17, "bbox": {"l": 306.244873046875, "t": 318.658203125, "r": 527.3578491210938, "b": 267.01153564453125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 203]}], "orig": "directly compare the results from CLIP with pure vision models, as the pretraining datasets they utilize differ significantly in scale and scope. Detailed results are presented in Figure 9 and Figure 10.", "text": "directly compare the results from CLIP with pure vision models, as the pretraining datasets they utilize differ significantly in scale and scope. Detailed results are presented in Figure 9 and Figure 10."}

cd22f130-095d-4521-b9ee-1c2cf2480ae2

2302.06555v2.pdf

text

POS impact on CLIP and OPT. In Table 9, we report the POS impact on OPT$_{30B}$ and two best CLIP vision encoders in our experiments.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/222", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 17, "bbox": {"l": 306.3138427734375, "t": 257.1636962890625, "r": 525.539794921875, "b": 219.1971435546875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 133]}], "orig": "POS impact on CLIP and OPT. In Table 9, we report the POS impact on OPT$_{30B}$ and two best CLIP vision encoders in our experiments.", "text": "POS impact on CLIP and OPT. In Table 9, we report the POS impact on OPT$_{30B}$ and two best CLIP vision encoders in our experiments."}

c935ad4a-c949-4ced-9a58-c8bc3e06be98

2304.09355v5.pdf

page_header

arXiv:2304.09355v5 [cs.LG] 21 Nov 2023

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/0", "parent": {"cref": "#/body"}, "children": [], "label": "page_header", "prov": [{"page_no": 1, "bbox": {"l": 17.048152923583984, "t": 584.1799926757812, "r": 36.339778900146484, "b": 231.9525146484375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 38]}], "orig": "arXiv:2304.09355v5 [cs.LG] 21 Nov 2023", "text": "arXiv:2304.09355v5 [cs.LG] 21 Nov 2023"}

bc299c52-9dcc-44a9-a0cd-b5b9cf7b9aa2

2304.09355v5.pdf

page_header

To Compress or Not to Compress

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/1", "parent": {"cref": "#/body"}, "children": [], "label": "page_header", "prov": [{"page_no": 1, "bbox": {"l": 216.67510986328125, "t": 752.1227416992188, "r": 391.006103515625, "b": 741.8831176757812, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 30]}], "orig": "To Compress or Not to Compress", "text": "To Compress or Not to Compress"}

df371f1d-6e5c-4bc6-872b-4e850a0ddbc7

2304.09355v5.pdf

section_header

To Compress or Not to Compress - Self-Supervised Learning and Information Theory: A Review

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/2", "parent": {"cref": "#/body"}, "children": [], "label": "section_header", "prov": [{"page_no": 1, "bbox": {"l": 90.24012756347656, "t": 695.8875122070312, "r": 521.2555541992188, "b": 663.5675659179688, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 90]}], "orig": "To Compress or Not to Compress - Self-Supervised Learning and Information Theory: A Review", "text": "To Compress or Not to Compress - Self-Supervised Learning and Information Theory: A Review", "level": 1}

eba286c6-7443-4769-8fde-8ba8d8f35061

2304.09355v5.pdf

text

Ravid Shwartz-Ziv New York University

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/3", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 1, "bbox": {"l": 89.51315307617188, "t": 647.0247802734375, "r": 288.3852233886719, "b": 635.8550415039062, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 37]}], "orig": "Ravid Shwartz-Ziv New York University", "text": "Ravid Shwartz-Ziv New York University"}

2e954c16-4aa1-458c-9973-538469ab5a27

2304.09355v5.pdf

text

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/4", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 1, "bbox": {"l": 386.21044921875, "t": 645.8736572265625, "r": 521.9979858398438, "b": 636.7821655273438, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 25]}], "orig": "[email protected]", "text": "[email protected]"}

62d1143f-715a-46b1-a64f-8c03907955b4

2304.09355v5.pdf

text

Yann LeCun New York University & Meta AI - FAIR

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/5", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 1, "bbox": {"l": 88.8544921875, "t": 633.96142578125, "r": 340.334228515625, "b": 623.0888671875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 47]}], "orig": "Yann LeCun New York University & Meta AI - FAIR", "text": "Yann LeCun New York University & Meta AI - FAIR"}

7a4c7bae-d7a2-42ac-b46a-7a68dcc5056e

2304.09355v5.pdf

section_header

Abstract

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/6", "parent": {"cref": "#/body"}, "children": [], "label": "section_header", "prov": [{"page_no": 1, "bbox": {"l": 279.8447265625, "t": 585.865234375, "r": 331.7017517089844, "b": 574.3516845703125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 8]}], "orig": "Abstract", "text": "Abstract", "level": 1}

d81777c5-2a34-4d37-af6c-535e121b0595

2304.09355v5.pdf

text

Deep neural networks excel in supervised learning tasks but are constrained by the need for extensive labeled data. Self-supervised learning emerges as a promising alternative, allowing models to learn without explicit labels. Information theory, and notably the information bottleneck principle, has been pivotal in shaping deep neural networks. This principle focuses on optimizing the trade-off between compression and preserving relevant information, providing a foundation for efficient network design in supervised contexts. However, its precise role and adaptation in self-supervised learning remain unclear. In this work, we scrutinize various self-supervised learning approaches from an informationtheoretic perspective, introducing a unified framework that encapsulates the self-supervised information-theoretic learning problem . We weave together existing research into a cohesive narrative, delve into contemporary self-supervised methodologies, and spotlight potential research avenues and inherent challenges. Additionally, we discuss the empirical evaluation of information-theoretic quantities and their estimation methods. Overall, this paper furnishes an exhaustive review of the intersection of information theory, self-supervised learning, and deep neural networks.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/7", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 1, "bbox": {"l": 108.97402954101562, "t": 559.693359375, "r": 504.0065612792969, "b": 382.31695556640625, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 1286]}], "orig": "Deep neural networks excel in supervised learning tasks but are constrained by the need for extensive labeled data. Self-supervised learning emerges as a promising alternative, allowing models to learn without explicit labels. Information theory, and notably the information bottleneck principle, has been pivotal in shaping deep neural networks. This principle focuses on optimizing the trade-off between compression and preserving relevant information, providing a foundation for efficient network design in supervised contexts. However, its precise role and adaptation in self-supervised learning remain unclear. In this work, we scrutinize various self-supervised learning approaches from an informationtheoretic perspective, introducing a unified framework that encapsulates the self-supervised information-theoretic learning problem . We weave together existing research into a cohesive narrative, delve into contemporary self-supervised methodologies, and spotlight potential research avenues and inherent challenges. Additionally, we discuss the empirical evaluation of information-theoretic quantities and their estimation methods. Overall, this paper furnishes an exhaustive review of the intersection of information theory, self-supervised learning, and deep neural networks.", "text": "Deep neural networks excel in supervised learning tasks but are constrained by the need for extensive labeled data. Self-supervised learning emerges as a promising alternative, allowing models to learn without explicit labels. Information theory, and notably the information bottleneck principle, has been pivotal in shaping deep neural networks. This principle focuses on optimizing the trade-off between compression and preserving relevant information, providing a foundation for efficient network design in supervised contexts. However, its precise role and adaptation in self-supervised learning remain unclear. In this work, we scrutinize various self-supervised learning approaches from an informationtheoretic perspective, introducing a unified framework that encapsulates the self-supervised information-theoretic learning problem . We weave together existing research into a cohesive narrative, delve into contemporary self-supervised methodologies, and spotlight potential research avenues and inherent challenges. Additionally, we discuss the empirical evaluation of information-theoretic quantities and their estimation methods. Overall, this paper furnishes an exhaustive review of the intersection of information theory, self-supervised learning, and deep neural networks."}

afc0f5f7-7730-4ed8-8aa9-157448dee7ce

2304.09355v5.pdf

text

Keywords: Self-Supervised Learning, Information Theory, Representation Learning

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/8", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 1, "bbox": {"l": 109.1109848022461, "t": 370.41619873046875, "r": 483.8846740722656, "b": 359.9473876953125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 79]}], "orig": "Keywords: Self-Supervised Learning, Information Theory, Representation Learning", "text": "Keywords: Self-Supervised Learning, Information Theory, Representation Learning"}

63b72cfa-0b0f-45f8-83d7-c57f5f877641

2304.09355v5.pdf

section_header

1. Introduction

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/9", "parent": {"cref": "#/body"}, "children": [], "label": "section_header", "prov": [{"page_no": 1, "bbox": {"l": 89.9918441772461, "t": 331.13165283203125, "r": 180.34707641601562, "b": 319.9496765136719, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 15]}], "orig": "1. Introduction", "text": "1. Introduction", "level": 1}

92ebe21c-6395-4e5a-90a1-95b4bbcc724e

2304.09355v5.pdf

text

Deep neural networks (DNNs) have revolutionized fields such as computer vision, natural language processing, and speech recognition due to their remarkable performance in supervised learning tasks (Alam et al., 2020; He et al., 2015; LeCun et al., 2015). However, the success of DNNs is often limited by the need for vast amounts of labeled data, which can be both time-consuming and expensive to acquire. Self-supervised learning (SSL) emerges as a promising alternative, enabling models to learn from data without explicit labels by leveraging the underlying structure and relationships within the data itself.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/10", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 1, "bbox": {"l": 88.8239974975586, "t": 309.8063659667969, "r": 523.817626953125, "b": 217.18157958984375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 612]}], "orig": "Deep neural networks (DNNs) have revolutionized fields such as computer vision, natural language processing, and speech recognition due to their remarkable performance in supervised learning tasks (Alam et al., 2020; He et al., 2015; LeCun et al., 2015). However, the success of DNNs is often limited by the need for vast amounts of labeled data, which can be both time-consuming and expensive to acquire. Self-supervised learning (SSL) emerges as a promising alternative, enabling models to learn from data without explicit labels by leveraging the underlying structure and relationships within the data itself.", "text": "Deep neural networks (DNNs) have revolutionized fields such as computer vision, natural language processing, and speech recognition due to their remarkable performance in supervised learning tasks (Alam et al., 2020; He et al., 2015; LeCun et al., 2015). However, the success of DNNs is often limited by the need for vast amounts of labeled data, which can be both time-consuming and expensive to acquire. Self-supervised learning (SSL) emerges as a promising alternative, enabling models to learn from data without explicit labels by leveraging the underlying structure and relationships within the data itself."}

2c78e44b-77e8-4f29-a322-80c6271b75f8

2304.09355v5.pdf

text

Recent advances in SSL have been driven by joint embedding architectures, such as Siamese Nets (Bromley et al., 1993), DrLIM (Chopra et al., 2005; Hadsell et al., 2006), and SimCLR (Chen et al., 2020a). These approaches define a loss function that encourages representations of different versions of the same image to be similar while pushing representations of distinct images apart. After optimizing the surrogate objective, the pre-trained model can be employed as a feature extractor, with the learned features serving as inputs for downstream supervised tasks like image classification, object detection, instance segmentation, or pose estimation (Caron et al., 2021; Chen et al., 2020a; Misra and van der Maaten, 2020; ShwartzZiv et al., 2022b). Although SSL methods have shown promising results in practice, the

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/11", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 1, "bbox": {"l": 88.7239990234375, "t": 205.8809814453125, "r": 523.8172607421875, "b": 86.47491455078125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 818]}], "orig": "Recent advances in SSL have been driven by joint embedding architectures, such as Siamese Nets (Bromley et al., 1993), DrLIM (Chopra et al., 2005; Hadsell et al., 2006), and SimCLR (Chen et al., 2020a). These approaches define a loss function that encourages representations of different versions of the same image to be similar while pushing representations of distinct images apart. After optimizing the surrogate objective, the pre-trained model can be employed as a feature extractor, with the learned features serving as inputs for downstream supervised tasks like image classification, object detection, instance segmentation, or pose estimation (Caron et al., 2021; Chen et al., 2020a; Misra and van der Maaten, 2020; ShwartzZiv et al., 2022b). Although SSL methods have shown promising results in practice, the", "text": "Recent advances in SSL have been driven by joint embedding architectures, such as Siamese Nets (Bromley et al., 1993), DrLIM (Chopra et al., 2005; Hadsell et al., 2006), and SimCLR (Chen et al., 2020a). These approaches define a loss function that encourages representations of different versions of the same image to be similar while pushing representations of distinct images apart. After optimizing the surrogate objective, the pre-trained model can be employed as a feature extractor, with the learned features serving as inputs for downstream supervised tasks like image classification, object detection, instance segmentation, or pose estimation (Caron et al., 2021; Chen et al., 2020a; Misra and van der Maaten, 2020; ShwartzZiv et al., 2022b). Although SSL methods have shown promising results in practice, the"}

6a615eda-22e3-46ec-b111-538418f40309

2304.09355v5.pdf

page_footer

1

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/12", "parent": {"cref": "#/body"}, "children": [], "label": "page_footer", "prov": [{"page_no": 1, "bbox": {"l": 303.3566589355469, "t": 66.76409912109375, "r": 308.49029541015625, "b": 57.317256927490234, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 1]}], "orig": "1", "text": "1"}

5f6d419e-d3e6-4ddc-a214-f9c50540b26d

2304.09355v5.pdf

text

theoretical underpinnings behind their effectiveness remain an open question (Arora et al., 2019; Lee et al., 2021a).

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/13", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 2, "bbox": {"l": 89.18399810791016, "t": 698.4678344726562, "r": 523.51904296875, "b": 674.0634155273438, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 117]}], "orig": "theoretical underpinnings behind their effectiveness remain an open question (Arora et al., 2019; Lee et al., 2021a).", "text": "theoretical underpinnings behind their effectiveness remain an open question (Arora et al., 2019; Lee et al., 2021a)."}

fdf81ebd-d0e6-406b-bb19-7e29f306c781

2304.09355v5.pdf

text

Information theory has played a crucial role in understanding and optimizing deep neural networks, from practical applications like the variational information bottleneck (Alemi et al., 2016) to theoretical investigations of generalization bounds induced by mutual information (Steinke and Zakynthinou, 2020; Xu and Raginsky, 2017). Building upon these foundations, several researchers have attempted to enhance self-supervised and semisupervised learning algorithms using information-theoretic principles, such as the Mutual Information Neural Estimator (MINE) (Belghazi et al., 2018b) combined with the information maximization (InfoMax) principle (Linsker, 1988). However, the plethora of objective functions, contradicting assumptions, and various estimation techniques in the literature can make it challenging to grasp the underlying principles and their implications.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/14", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 2, "bbox": {"l": 89.1538314819336, "t": 662.443603515625, "r": 523.8245849609375, "b": 529.1944580078125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 874]}], "orig": "Information theory has played a crucial role in understanding and optimizing deep neural networks, from practical applications like the variational information bottleneck (Alemi et al., 2016) to theoretical investigations of generalization bounds induced by mutual information (Steinke and Zakynthinou, 2020; Xu and Raginsky, 2017). Building upon these foundations, several researchers have attempted to enhance self-supervised and semisupervised learning algorithms using information-theoretic principles, such as the Mutual Information Neural Estimator (MINE) (Belghazi et al., 2018b) combined with the information maximization (InfoMax) principle (Linsker, 1988). However, the plethora of objective functions, contradicting assumptions, and various estimation techniques in the literature can make it challenging to grasp the underlying principles and their implications.", "text": "Information theory has played a crucial role in understanding and optimizing deep neural networks, from practical applications like the variational information bottleneck (Alemi et al., 2016) to theoretical investigations of generalization bounds induced by mutual information (Steinke and Zakynthinou, 2020; Xu and Raginsky, 2017). Building upon these foundations, several researchers have attempted to enhance self-supervised and semisupervised learning algorithms using information-theoretic principles, such as the Mutual Information Neural Estimator (MINE) (Belghazi et al., 2018b) combined with the information maximization (InfoMax) principle (Linsker, 1988). However, the plethora of objective functions, contradicting assumptions, and various estimation techniques in the literature can make it challenging to grasp the underlying principles and their implications."}

d4d508ee-f428-48f5-b765-531dbf54758e

2304.09355v5.pdf

text

In this paper, we aim to achieve two objectives. First, we propose a unified framework that synthesizes existing research on self-supervised and semi-supervised learning from an information-theoretic standpoint. This framework allows us to present and compare current methods, analyze their assumptions and difficulties, and discuss the optimal representation for neural networks in general and self-supervised networks in particular. Second, we explore different methods and estimators for optimizing information-theoretic quantities in deep neural networks and investigate how recent models optimize various theoretical-information terms.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/15", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 2, "bbox": {"l": 89.29786682128906, "t": 517.6422119140625, "r": 522.2979125976562, "b": 412.41864013671875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 640]}], "orig": "In this paper, we aim to achieve two objectives. First, we propose a unified framework that synthesizes existing research on self-supervised and semi-supervised learning from an information-theoretic standpoint. This framework allows us to present and compare current methods, analyze their assumptions and difficulties, and discuss the optimal representation for neural networks in general and self-supervised networks in particular. Second, we explore different methods and estimators for optimizing information-theoretic quantities in deep neural networks and investigate how recent models optimize various theoretical-information terms.", "text": "In this paper, we aim to achieve two objectives. First, we propose a unified framework that synthesizes existing research on self-supervised and semi-supervised learning from an information-theoretic standpoint. This framework allows us to present and compare current methods, analyze their assumptions and difficulties, and discuss the optimal representation for neural networks in general and self-supervised networks in particular. Second, we explore different methods and estimators for optimizing information-theoretic quantities in deep neural networks and investigate how recent models optimize various theoretical-information terms."}

1366b73a-52ce-4247-b1d5-c168269b2acf

2304.09355v5.pdf

text

By reviewing the literature on various aspects of information-theoretic learning, we provide a comprehensive understanding of the interplay between information theory, self-supervised learning, and deep neural networks. We discuss the application of the information bottleneck principle (Tishby et al., 1999a), connections between information theory and generalization, and recent information-theoretic learning algorithms. Furthermore, we examine how the information-theoretic perspective can offer insights into the design of better self-supervised learning algorithms and the potential benefits of using information theory in SSL across a wide range of applications.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/16", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 2, "bbox": {"l": 89.21715545654297, "t": 400.3964538574219, "r": 523.5151977539062, "b": 294.5848083496094, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 669]}], "orig": "By reviewing the literature on various aspects of information-theoretic learning, we provide a comprehensive understanding of the interplay between information theory, self-supervised learning, and deep neural networks. We discuss the application of the information bottleneck principle (Tishby et al., 1999a), connections between information theory and generalization, and recent information-theoretic learning algorithms. Furthermore, we examine how the information-theoretic perspective can offer insights into the design of better self-supervised learning algorithms and the potential benefits of using information theory in SSL across a wide range of applications.", "text": "By reviewing the literature on various aspects of information-theoretic learning, we provide a comprehensive understanding of the interplay between information theory, self-supervised learning, and deep neural networks. We discuss the application of the information bottleneck principle (Tishby et al., 1999a), connections between information theory and generalization, and recent information-theoretic learning algorithms. Furthermore, we examine how the information-theoretic perspective can offer insights into the design of better self-supervised learning algorithms and the potential benefits of using information theory in SSL across a wide range of applications."}

c401e2d1-41de-43ac-8745-13dee3d04d6d

2304.09355v5.pdf

text

In addition to the main structure of the paper, we dedicate a section to the challenges and opportunities in extending the information-theoretic perspective to other learning paradigms, such as energy-based models. We highlight the potential advantages of incorporating these extensions into self-supervised learning algorithms and discuss the technical and conceptual challenges that must be addressed.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/17", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 2, "bbox": {"l": 89.00598907470703, "t": 282.90179443359375, "r": 523.5209350585938, "b": 217.75103759765625, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 403]}], "orig": "In addition to the main structure of the paper, we dedicate a section to the challenges and opportunities in extending the information-theoretic perspective to other learning paradigms, such as energy-based models. We highlight the potential advantages of incorporating these extensions into self-supervised learning algorithms and discuss the technical and conceptual challenges that must be addressed.", "text": "In addition to the main structure of the paper, we dedicate a section to the challenges and opportunities in extending the information-theoretic perspective to other learning paradigms, such as energy-based models. We highlight the potential advantages of incorporating these extensions into self-supervised learning algorithms and discuss the technical and conceptual challenges that must be addressed."}

4566090b-b276-42db-805e-e8e0bec215d5

2304.09355v5.pdf

text

The structure of the paper is as follows. Section 2 introduces the key concepts in supervised, semi-supervised, self-supervised learning, information theory, and representation learning. Section 3 presents a unified framework for multiview learning based on information theory. We first discuss what an optimal representation is and why compression is beneficial for learning. Next, we explore optimal representation in single-view supervised learning models and how they can be extended to unsupervised, semi-supervised, and multiview contexts. The focus then shifts to self-supervised learning, where the optimal representation remains an open question. Using the unified framework, we compare recent self-supervised algorithms and discuss their differences. We analyze the assumptions behind these models, their effects

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/18", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 2, "bbox": {"l": 88.85343170166016, "t": 205.67364501953125, "r": 524.1221313476562, "b": 86.9251708984375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 822]}], "orig": "The structure of the paper is as follows. Section 2 introduces the key concepts in supervised, semi-supervised, self-supervised learning, information theory, and representation learning. Section 3 presents a unified framework for multiview learning based on information theory. We first discuss what an optimal representation is and why compression is beneficial for learning. Next, we explore optimal representation in single-view supervised learning models and how they can be extended to unsupervised, semi-supervised, and multiview contexts. The focus then shifts to self-supervised learning, where the optimal representation remains an open question. Using the unified framework, we compare recent self-supervised algorithms and discuss their differences. We analyze the assumptions behind these models, their effects", "text": "The structure of the paper is as follows. Section 2 introduces the key concepts in supervised, semi-supervised, self-supervised learning, information theory, and representation learning. Section 3 presents a unified framework for multiview learning based on information theory. We first discuss what an optimal representation is and why compression is beneficial for learning. Next, we explore optimal representation in single-view supervised learning models and how they can be extended to unsupervised, semi-supervised, and multiview contexts. The focus then shifts to self-supervised learning, where the optimal representation remains an open question. Using the unified framework, we compare recent self-supervised algorithms and discuss their differences. We analyze the assumptions behind these models, their effects"}

221c6a02-90ac-4372-ab95-62dd62c819e5

2304.09355v5.pdf

page_footer

2

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/19", "parent": {"cref": "#/body"}, "children": [], "label": "page_footer", "prov": [{"page_no": 2, "bbox": {"l": 302.83428955078125, "t": 66.747802734375, "r": 308.49029541015625, "b": 57.317256927490234, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 1]}], "orig": "2", "text": "2"}

50337d6f-7d1c-47c7-ad69-071869989e21

2304.09355v5.pdf

page_header

To Compress or Not to Compress

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/20", "parent": {"cref": "#/body"}, "children": [], "label": "page_header", "prov": [{"page_no": 3, "bbox": {"l": 216.72885131835938, "t": 752.1143798828125, "r": 390.9455871582031, "b": 741.8831176757812, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 30]}], "orig": "To Compress or Not to Compress", "text": "To Compress or Not to Compress"}

30f65453-77f8-4459-a58c-1f55a0d34b46

2304.09355v5.pdf

text

on the learned representation, and their varying perspectives on important information within the network.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/21", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 3, "bbox": {"l": 89.13358306884766, "t": 698.26904296875, "r": 522.005615234375, "b": 674.5136108398438, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 106]}], "orig": "on the learned representation, and their varying perspectives on important information within the network.", "text": "on the learned representation, and their varying perspectives on important information within the network."}

2de25490-e7e8-4c00-82c0-295d0c861fe8

2304.09355v5.pdf

text

Section 5 addresses several technical challenges, discussing both theoretical and practical issues in estimating theoretical information terms. We present recent methods for estimating these quantities, including variational bounds and estimators. Section 6 concludes the paper by offering insights into potential future research directions at the intersection of information theory, self-supervised learning, and deep neural networks. Our aim is to inspire further research that leverages information theory to advance our understanding of self-supervised learning and to develop more efficient and effective models for a broad range of applications.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/22", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 3, "bbox": {"l": 89.21458435058594, "t": 661.738525390625, "r": 524.1165161132812, "b": 569.0333862304688, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 651]}], "orig": "Section 5 addresses several technical challenges, discussing both theoretical and practical issues in estimating theoretical information terms. We present recent methods for estimating these quantities, including variational bounds and estimators. Section 6 concludes the paper by offering insights into potential future research directions at the intersection of information theory, self-supervised learning, and deep neural networks. Our aim is to inspire further research that leverages information theory to advance our understanding of self-supervised learning and to develop more efficient and effective models for a broad range of applications.", "text": "Section 5 addresses several technical challenges, discussing both theoretical and practical issues in estimating theoretical information terms. We present recent methods for estimating these quantities, including variational bounds and estimators. Section 6 concludes the paper by offering insights into potential future research directions at the intersection of information theory, self-supervised learning, and deep neural networks. Our aim is to inspire further research that leverages information theory to advance our understanding of self-supervised learning and to develop more efficient and effective models for a broad range of applications."}

d7caceb7-9c5d-48c1-b227-2681ec5f83fb

2304.09355v5.pdf

section_header

2. Background and Fundamental Concepts

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/23", "parent": {"cref": "#/body"}, "children": [], "label": "section_header", "prov": [{"page_no": 3, "bbox": {"l": 89.69479370117188, "t": 540.44921875, "r": 341.64739990234375, "b": 528.4921875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 38]}], "orig": "2. Background and Fundamental Concepts", "text": "2. Background and Fundamental Concepts", "level": 1}

0d8605ab-dc14-456a-a575-9b50b0672e1d

2304.09355v5.pdf

section_header

2.1 Multiview Representation Learning

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/24", "parent": {"cref": "#/body"}, "children": [], "label": "section_header", "prov": [{"page_no": 3, "bbox": {"l": 89.4501724243164, "t": 519.122802734375, "r": 307.7369384765625, "b": 508.2320556640625, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 37]}], "orig": "2.1 Multiview Representation Learning", "text": "2.1 Multiview Representation Learning", "level": 1}

7d25a754-68f8-4f09-9fb3-e5296ca4f5c2

2304.09355v5.pdf

text

Multiview learning has gained increasing attention and great practical success by using complementary information from multiple features or modalities. The multiview learning paradigm divides the input variable into multiple views from which the target variable should be predicted (Zhao et al., 2017b). Using this paradigm, one can eliminate hypotheses that contradict predictions from other views and provide a natural semi-supervised and self-supervised learning setting. A multiview dataset consists of data captured from multiple sources, modalities, and forms but with similar high-level semantics (Yan et al., 2021). This mechanism was initially used for natural-world data, combining image, text, audio, and video measurements. For example, photos of objects are taken from various angles, and our supervised task is to identify the objects. Another example is identifying a person by analyzing the video stream as one view and the audio stream as the other.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/25", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 3, "bbox": {"l": 89.10533142089844, "t": 499.7522888183594, "r": 522.4059448242188, "b": 353.42889404296875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 966]}], "orig": "Multiview learning has gained increasing attention and great practical success by using complementary information from multiple features or modalities. The multiview learning paradigm divides the input variable into multiple views from which the target variable should be predicted (Zhao et al., 2017b). Using this paradigm, one can eliminate hypotheses that contradict predictions from other views and provide a natural semi-supervised and self-supervised learning setting. A multiview dataset consists of data captured from multiple sources, modalities, and forms but with similar high-level semantics (Yan et al., 2021). This mechanism was initially used for natural-world data, combining image, text, audio, and video measurements. For example, photos of objects are taken from various angles, and our supervised task is to identify the objects. Another example is identifying a person by analyzing the video stream as one view and the audio stream as the other.", "text": "Multiview learning has gained increasing attention and great practical success by using complementary information from multiple features or modalities. The multiview learning paradigm divides the input variable into multiple views from which the target variable should be predicted (Zhao et al., 2017b). Using this paradigm, one can eliminate hypotheses that contradict predictions from other views and provide a natural semi-supervised and self-supervised learning setting. A multiview dataset consists of data captured from multiple sources, modalities, and forms but with similar high-level semantics (Yan et al., 2021). This mechanism was initially used for natural-world data, combining image, text, audio, and video measurements. For example, photos of objects are taken from various angles, and our supervised task is to identify the objects. Another example is identifying a person by analyzing the video stream as one view and the audio stream as the other."}

d09a4822-0422-492c-81f4-4af9b729dfb7

2304.09355v5.pdf

text

Although these views often provide different and complementary information about the same data, directly integrating them does not produce satisfactory results due to biases between multiple views (Yan et al., 2021). Thus, multiview representation learning involves identifying the underlying data structure and integrating the different views into a common feature space, resulting in high performance. In recent decades, multiview learning has been used for many machine learning tasks and influenced many algorithms, such as co-training mechanisms (Kumar and Daum' e, 2011), subspace learning methods (Xue et al., 2019), and multiple kernel learning (MKL) (Bach and Jordan, 2002). Li et al. (2018) proposed two categories for multiview representation learning: (i) multiview representation fusion, which combines different features from multiple views into a single compact representation, and (ii) alignment of multiview representation, which attempts to capture the relationships among multiple different views through feature alignment. In this case, a learned mapping function embeds the data of each view, and the representations are regularized to form a multiviewaligned space. In this research direction, an early study is the Canonical Correlation Analysis (CCA) (Hotelling, 1936) and its kernel extensions (Bach and Jordan, 2003; Hardoon et al., 2004; Sun, 2013). In addition to CCA, multiview representation learning has penetrated a variety of learning methods, such as dimensionality reduction (Sun et al., 2010), clustering analysis (Yan et al., 2015), multiview sparse coding (Cao et al., 2013; Jia et al., 2010; Liu et al., 2014), and multimodal topic learning (Pu et al., 2020). However, despite their

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/26", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 3, "bbox": {"l": 88.7239990234375, "t": 341.7623291015625, "r": 523.8228759765625, "b": 86.5010986328125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 1721]}], "orig": "Although these views often provide different and complementary information about the same data, directly integrating them does not produce satisfactory results due to biases between multiple views (Yan et al., 2021). Thus, multiview representation learning involves identifying the underlying data structure and integrating the different views into a common feature space, resulting in high performance. In recent decades, multiview learning has been used for many machine learning tasks and influenced many algorithms, such as co-training mechanisms (Kumar and Daum' e, 2011), subspace learning methods (Xue et al., 2019), and multiple kernel learning (MKL) (Bach and Jordan, 2002). Li et al. (2018) proposed two categories for multiview representation learning: (i) multiview representation fusion, which combines different features from multiple views into a single compact representation, and (ii) alignment of multiview representation, which attempts to capture the relationships among multiple different views through feature alignment. In this case, a learned mapping function embeds the data of each view, and the representations are regularized to form a multiviewaligned space. In this research direction, an early study is the Canonical Correlation Analysis (CCA) (Hotelling, 1936) and its kernel extensions (Bach and Jordan, 2003; Hardoon et al., 2004; Sun, 2013). In addition to CCA, multiview representation learning has penetrated a variety of learning methods, such as dimensionality reduction (Sun et al., 2010), clustering analysis (Yan et al., 2015), multiview sparse coding (Cao et al., 2013; Jia et al., 2010; Liu et al., 2014), and multimodal topic learning (Pu et al., 2020). However, despite their", "text": "Although these views often provide different and complementary information about the same data, directly integrating them does not produce satisfactory results due to biases between multiple views (Yan et al., 2021). Thus, multiview representation learning involves identifying the underlying data structure and integrating the different views into a common feature space, resulting in high performance. In recent decades, multiview learning has been used for many machine learning tasks and influenced many algorithms, such as co-training mechanisms (Kumar and Daum' e, 2011), subspace learning methods (Xue et al., 2019), and multiple kernel learning (MKL) (Bach and Jordan, 2002). Li et al. (2018) proposed two categories for multiview representation learning: (i) multiview representation fusion, which combines different features from multiple views into a single compact representation, and (ii) alignment of multiview representation, which attempts to capture the relationships among multiple different views through feature alignment. In this case, a learned mapping function embeds the data of each view, and the representations are regularized to form a multiviewaligned space. In this research direction, an early study is the Canonical Correlation Analysis (CCA) (Hotelling, 1936) and its kernel extensions (Bach and Jordan, 2003; Hardoon et al., 2004; Sun, 2013). In addition to CCA, multiview representation learning has penetrated a variety of learning methods, such as dimensionality reduction (Sun et al., 2010), clustering analysis (Yan et al., 2015), multiview sparse coding (Cao et al., 2013; Jia et al., 2010; Liu et al., 2014), and multimodal topic learning (Pu et al., 2020). However, despite their"}

09c78988-7391-4c7b-87c7-3cbac278771a

2304.09355v5.pdf

page_footer

3

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/27", "parent": {"cref": "#/body"}, "children": [], "label": "page_footer", "prov": [{"page_no": 3, "bbox": {"l": 302.9095764160156, "t": 66.6334228515625, "r": 308.49029541015625, "b": 57.317256927490234, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 1]}], "orig": "3", "text": "3"}

783bb0ae-863f-4690-914e-1f0190025223

2304.09355v5.pdf

text

promising results, these methods use handcrafted features and linear embedding functions, which cannot capture the nonlinear properties of multiview data.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/28", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 4, "bbox": {"l": 89.12200164794922, "t": 698.1217041015625, "r": 523.5210571289062, "b": 674.1952514648438, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 154]}], "orig": "promising results, these methods use handcrafted features and linear embedding functions, which cannot capture the nonlinear properties of multiview data.", "text": "promising results, these methods use handcrafted features and linear embedding functions, which cannot capture the nonlinear properties of multiview data."}

1250b077-0ce4-4669-af11-785108f29afd

2304.09355v5.pdf

text

The emergence of deep learning has provided a powerful way to learn complex, nonlinear, and hierarchical representations of data. By incorporating multiple hierarchical layers, deep learning algorithms can learn complex, subtle, and abstract representations of target data. The success of deep learning in various application domains has led to a growing interest in deep multiview methods, which have shown promising results. Examples of these methods include deep multiview canonical correlation analysis (Andrew et al., 2013) as an extension of CCA, multiview clustering via deep matrix factorization (Zhao et al., 2017a), and the deep multiview spectral network (Huang et al., 2019). Moreover, deep architectures have been employed to generate effective representations in methods such as multiview convolutional neural networks (Liu et al., 2021a), multimodal deep Boltzmann machines (Srivastava and Salakhutdinov, 2014), multimodal deep autoencoders (Ngiam et al., 2011; Wang et al., 2015), and multimodal recurrent neural networks (Donahue et al., 2015; Karpathy and Fei-Fei, 2015; Mao et al., 2014).

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/29", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 4, "bbox": {"l": 88.89997863769531, "t": 661.9761962890625, "r": 524.1187133789062, "b": 487.9737548828125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 1107]}], "orig": "The emergence of deep learning has provided a powerful way to learn complex, nonlinear, and hierarchical representations of data. By incorporating multiple hierarchical layers, deep learning algorithms can learn complex, subtle, and abstract representations of target data. The success of deep learning in various application domains has led to a growing interest in deep multiview methods, which have shown promising results. Examples of these methods include deep multiview canonical correlation analysis (Andrew et al., 2013) as an extension of CCA, multiview clustering via deep matrix factorization (Zhao et al., 2017a), and the deep multiview spectral network (Huang et al., 2019). Moreover, deep architectures have been employed to generate effective representations in methods such as multiview convolutional neural networks (Liu et al., 2021a), multimodal deep Boltzmann machines (Srivastava and Salakhutdinov, 2014), multimodal deep autoencoders (Ngiam et al., 2011; Wang et al., 2015), and multimodal recurrent neural networks (Donahue et al., 2015; Karpathy and Fei-Fei, 2015; Mao et al., 2014).", "text": "The emergence of deep learning has provided a powerful way to learn complex, nonlinear, and hierarchical representations of data. By incorporating multiple hierarchical layers, deep learning algorithms can learn complex, subtle, and abstract representations of target data. The success of deep learning in various application domains has led to a growing interest in deep multiview methods, which have shown promising results. Examples of these methods include deep multiview canonical correlation analysis (Andrew et al., 2013) as an extension of CCA, multiview clustering via deep matrix factorization (Zhao et al., 2017a), and the deep multiview spectral network (Huang et al., 2019). Moreover, deep architectures have been employed to generate effective representations in methods such as multiview convolutional neural networks (Liu et al., 2021a), multimodal deep Boltzmann machines (Srivastava and Salakhutdinov, 2014), multimodal deep autoencoders (Ngiam et al., 2011; Wang et al., 2015), and multimodal recurrent neural networks (Donahue et al., 2015; Karpathy and Fei-Fei, 2015; Mao et al., 2014)."}

c1fddcd9-76f5-4f62-81dc-7e6c1581e07a

2304.09355v5.pdf

section_header

2.2 Self-Supervised Learning

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/30", "parent": {"cref": "#/body"}, "children": [], "label": "section_header", "prov": [{"page_no": 4, "bbox": {"l": 89.47816467285156, "t": 471.16668701171875, "r": 249.0968780517578, "b": 459.9459533691406, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 28]}], "orig": "2.2 Self-Supervised Learning", "text": "2.2 Self-Supervised Learning", "level": 1}

bb66a9da-fe70-4291-a614-6030cded3918

2304.09355v5.pdf

text

Self-supervised learning (SSL) is a powerful technique that leverages unlabeled data to learn useful representations. In contrast to supervised learning, which relies on labeled data, SSL employs self-defined signals to establish a proxy objective between the input and the signal. The model is initially trained using this proxy objective and subsequently fine-tuned on the target task. Self-supervised signals, derived from the inherent co-occurrence relationships in the data, serve as self-supervision. Various such signals have been used to learn representations, including generative and joint embedding architectures (Bachman et al., 2019; Bar et al., 2022; Chen et al., 2020a,b).

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/31", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 4, "bbox": {"l": 89.25768280029297, "t": 451.8852233886719, "r": 522.4007568359375, "b": 345.9452209472656, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 687]}], "orig": "Self-supervised learning (SSL) is a powerful technique that leverages unlabeled data to learn useful representations. In contrast to supervised learning, which relies on labeled data, SSL employs self-defined signals to establish a proxy objective between the input and the signal. The model is initially trained using this proxy objective and subsequently fine-tuned on the target task. Self-supervised signals, derived from the inherent co-occurrence relationships in the data, serve as self-supervision. Various such signals have been used to learn representations, including generative and joint embedding architectures (Bachman et al., 2019; Bar et al., 2022; Chen et al., 2020a,b).", "text": "Self-supervised learning (SSL) is a powerful technique that leverages unlabeled data to learn useful representations. In contrast to supervised learning, which relies on labeled data, SSL employs self-defined signals to establish a proxy objective between the input and the signal. The model is initially trained using this proxy objective and subsequently fine-tuned on the target task. Self-supervised signals, derived from the inherent co-occurrence relationships in the data, serve as self-supervision. Various such signals have been used to learn representations, including generative and joint embedding architectures (Bachman et al., 2019; Bar et al., 2022; Chen et al., 2020a,b)."}

b42278f4-fcb7-4918-939f-6b1e82c76993

2304.09355v5.pdf

text

Two main categories of SSL architectures exist: (1) generative architectures based on reconstruction or prediction and (2) joint embedding architectures (Liu et al., 2021b). Both architecture classes can be trained using either contrastive or non-contrastive methods.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/32", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 4, "bbox": {"l": 89.14336395263672, "t": 333.6808776855469, "r": 521.999755859375, "b": 295.71673583984375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 267]}], "orig": "Two main categories of SSL architectures exist: (1) generative architectures based on reconstruction or prediction and (2) joint embedding architectures (Liu et al., 2021b). Both architecture classes can be trained using either contrastive or non-contrastive methods.", "text": "Two main categories of SSL architectures exist: (1) generative architectures based on reconstruction or prediction and (2) joint embedding architectures (Liu et al., 2021b). Both architecture classes can be trained using either contrastive or non-contrastive methods."}

ccd2a021-a9d8-4efe-95bd-3fe4c1287b1d

2304.09355v5.pdf

text

We begin by discussing these two main types of architectures:

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/33", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 4, "bbox": {"l": 88.83647918701172, "t": 283.6369934082031, "r": 385.017333984375, "b": 272.6397705078125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 61]}], "orig": "We begin by discussing these two main types of architectures:", "text": "We begin by discussing these two main types of architectures:"}

cb65dc35-de76-4f0d-8fd0-36211e76bfeb

2304.09355v5.pdf

list_item

1. Generative Architecture: Generative architectures employ an objective function that measures the divergence between input data and predicted reconstructions, such as squared error. The architecture reconstructs data from a latent variable or a corrupted version, potentially with a latent variable's assistance. Notable examples of generative architectures include auto-encoders, sparse coding, sparse auto-encoders, and variational auto-encoders (Kingma and Welling, 2013; Lee et al., 2006; Ng et al., 2011). As the reconstruction task lacks a single correct answer, most generative architectures utilize a latent variable, which, when varied, generates multiple reconstructions. The latent variable's information content requires regularization to ensure the system reconstructs regions of high data density while avoiding a collapse by reconstructing the entire space. PCA regularizes the latent variable by limiting its dimensions, while sparse coding and sparse auto-encoders restrict the number of non-zero components. Variational auto-encoders regularize the latent variable by rendering it stochastic

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/34", "parent": {"cref": "#/groups/0"}, "children": [], "label": "list_item", "prov": [{"page_no": 4, "bbox": {"l": 103.18352508544922, "t": 260.2718505859375, "r": 524.1220092773438, "b": 86.83660888671875, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 1111]}], "orig": "1. Generative Architecture: Generative architectures employ an objective function that measures the divergence between input data and predicted reconstructions, such as squared error. The architecture reconstructs data from a latent variable or a corrupted version, potentially with a latent variable's assistance. Notable examples of generative architectures include auto-encoders, sparse coding, sparse auto-encoders, and variational auto-encoders (Kingma and Welling, 2013; Lee et al., 2006; Ng et al., 2011). As the reconstruction task lacks a single correct answer, most generative architectures utilize a latent variable, which, when varied, generates multiple reconstructions. The latent variable's information content requires regularization to ensure the system reconstructs regions of high data density while avoiding a collapse by reconstructing the entire space. PCA regularizes the latent variable by limiting its dimensions, while sparse coding and sparse auto-encoders restrict the number of non-zero components. Variational auto-encoders regularize the latent variable by rendering it stochastic", "text": "1. Generative Architecture: Generative architectures employ an objective function that measures the divergence between input data and predicted reconstructions, such as squared error. The architecture reconstructs data from a latent variable or a corrupted version, potentially with a latent variable's assistance. Notable examples of generative architectures include auto-encoders, sparse coding, sparse auto-encoders, and variational auto-encoders (Kingma and Welling, 2013; Lee et al., 2006; Ng et al., 2011). As the reconstruction task lacks a single correct answer, most generative architectures utilize a latent variable, which, when varied, generates multiple reconstructions. The latent variable's information content requires regularization to ensure the system reconstructs regions of high data density while avoiding a collapse by reconstructing the entire space. PCA regularizes the latent variable by limiting its dimensions, while sparse coding and sparse auto-encoders restrict the number of non-zero components. Variational auto-encoders regularize the latent variable by rendering it stochastic", "enumerated": false, "marker": "-"}

a00b5fb3-828f-4e60-9494-2b00c8e158ae

2304.09355v5.pdf

page_footer

4

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/35", "parent": {"cref": "#/body"}, "children": [], "label": "page_footer", "prov": [{"page_no": 4, "bbox": {"l": 302.75335693359375, "t": 66.92059326171875, "r": 308.49029541015625, "b": 57.317256927490234, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 1]}], "orig": "4", "text": "4"}

90d5d228-a832-4990-a7f7-1c32b6abaf9c

2304.09355v5.pdf

page_header

To Compress or Not to Compress

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/36", "parent": {"cref": "#/body"}, "children": [], "label": "page_header", "prov": [{"page_no": 5, "bbox": {"l": 216.6831512451172, "t": 752.0804443359375, "r": 390.9455871582031, "b": 741.8831176757812, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 30]}], "orig": "To Compress or Not to Compress", "text": "To Compress or Not to Compress"}

c938ef65-defa-4054-8562-9cfbfbb2863a

2304.09355v5.pdf

text

and maximizing the entropy of the distribution relative to a prior. Vector quantized variational auto-encoders (VQ-VAE) employ binary stochastic variables to achieve similar results (Van Den Oord et al., 2017).

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/37", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 5, "bbox": {"l": 116.46167755126953, "t": 698.3206787109375, "r": 522.00439453125, "b": 660.656494140625, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 210]}], "orig": "and maximizing the entropy of the distribution relative to a prior. Vector quantized variational auto-encoders (VQ-VAE) employ binary stochastic variables to achieve similar results (Van Den Oord et al., 2017).", "text": "and maximizing the entropy of the distribution relative to a prior. Vector quantized variational auto-encoders (VQ-VAE) employ binary stochastic variables to achieve similar results (Van Den Oord et al., 2017)."}

b9dec7a5-1a21-4a23-b64c-a90bc9949c4f

2304.09355v5.pdf

list_item

2. Joint Embedding Architectures (JEA): These architectures process multiple views of an input signal through encoders, producing representations of the views. The system is trained to ensure that these representations are both informative and mutually predictable. Examples include Siamese networks, where two identical encoders share weights (Chen et al., 2020a; Chen and He, 2021; Grill et al., 2020; He et al., 2020), and methods permitting encoders to differ (Bardes et al., 2021). A primary challenge with JEA is preventing informational collapse, in which the representations contain minimal information about the inputs, thereby facilitating their mutual prediction. JEA's advantage lies in the encoders' ability to eliminate noisy, unpredictable, or irrelevant information from the input within the representation space.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/38", "parent": {"cref": "#/groups/1"}, "children": [], "label": "list_item", "prov": [{"page_no": 5, "bbox": {"l": 102.54643249511719, "t": 645.849853515625, "r": 524.1174926757812, "b": 512.730224609375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 829]}], "orig": "2. Joint Embedding Architectures (JEA): These architectures process multiple views of an input signal through encoders, producing representations of the views. The system is trained to ensure that these representations are both informative and mutually predictable. Examples include Siamese networks, where two identical encoders share weights (Chen et al., 2020a; Chen and He, 2021; Grill et al., 2020; He et al., 2020), and methods permitting encoders to differ (Bardes et al., 2021). A primary challenge with JEA is preventing informational collapse, in which the representations contain minimal information about the inputs, thereby facilitating their mutual prediction. JEA's advantage lies in the encoders' ability to eliminate noisy, unpredictable, or irrelevant information from the input within the representation space.", "text": "2. Joint Embedding Architectures (JEA): These architectures process multiple views of an input signal through encoders, producing representations of the views. The system is trained to ensure that these representations are both informative and mutually predictable. Examples include Siamese networks, where two identical encoders share weights (Chen et al., 2020a; Chen and He, 2021; Grill et al., 2020; He et al., 2020), and methods permitting encoders to differ (Bardes et al., 2021). A primary challenge with JEA is preventing informational collapse, in which the representations contain minimal information about the inputs, thereby facilitating their mutual prediction. JEA's advantage lies in the encoders' ability to eliminate noisy, unpredictable, or irrelevant information from the input within the representation space.", "enumerated": false, "marker": "-"}

e0e5846f-ae2b-4399-939a-9363f704c232

2304.09355v5.pdf

text

To effectively train these architectures, it is essential to ensure that the representations of different signals are distinct. This can be achieved through either contrastive or noncontrastive methods:

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/39", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 5, "bbox": {"l": 88.84542846679688, "t": 498.2249755859375, "r": 523.8182983398438, "b": 460.6176452636719, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 202]}], "orig": "To effectively train these architectures, it is essential to ensure that the representations of different signals are distinct. This can be achieved through either contrastive or noncontrastive methods:", "text": "To effectively train these architectures, it is essential to ensure that the representations of different signals are distinct. This can be achieved through either contrastive or noncontrastive methods:"}

132c7817-11d5-4c7f-9a03-dc251b4bb726

2304.09355v5.pdf

list_item

· Contrastive Methods: Contrastive methods utilize data points from the training set as positive samples and generate points outside the region of high data density as contrastive samples . The energy (e.g., reconstruction error for generative architectures or representation predictive error for JEA) should be low for positive samples and higher for contrastive samples. Various loss functions involving the energies of pairs or sets of samples can be minimized to achieve this objective.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/40", "parent": {"cref": "#/groups/2"}, "children": [], "label": "list_item", "prov": [{"page_no": 5, "bbox": {"l": 105.7928695678711, "t": 445.7911376953125, "r": 522.2193603515625, "b": 366.61651611328125, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 490]}], "orig": "\u00b7 Contrastive Methods: Contrastive methods utilize data points from the training set as positive samples and generate points outside the region of high data density as contrastive samples . The energy (e.g., reconstruction error for generative architectures or representation predictive error for JEA) should be low for positive samples and higher for contrastive samples. Various loss functions involving the energies of pairs or sets of samples can be minimized to achieve this objective.", "text": "\u00b7 Contrastive Methods: Contrastive methods utilize data points from the training set as positive samples and generate points outside the region of high data density as contrastive samples . The energy (e.g., reconstruction error for generative architectures or representation predictive error for JEA) should be low for positive samples and higher for contrastive samples. Various loss functions involving the energies of pairs or sets of samples can be minimized to achieve this objective.", "enumerated": false, "marker": "-"}

b7aa1139-2d13-45d2-b2f7-84901a4d7bf5

2304.09355v5.pdf

list_item

· Non-Contrastive Methods: Non-contrastive methods prevent the energy landscape's collapse by limiting the volume of space that can take low energy, either through architectural constraints or through a regularizer in the energy or training objective. In latent-variable generative architectures, preventing collapse is achieved by limiting or minimizing the information content of the latent variable. In JEA, collapse is prevented by maximizing the information content of the representations.

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/41", "parent": {"cref": "#/groups/2"}, "children": [], "label": "list_item", "prov": [{"page_no": 5, "bbox": {"l": 106.07862091064453, "t": 352.3957214355469, "r": 523.8255615234375, "b": 273.4095458984375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 494]}], "orig": "\u00b7 Non-Contrastive Methods: Non-contrastive methods prevent the energy landscape's collapse by limiting the volume of space that can take low energy, either through architectural constraints or through a regularizer in the energy or training objective. In latent-variable generative architectures, preventing collapse is achieved by limiting or minimizing the information content of the latent variable. In JEA, collapse is prevented by maximizing the information content of the representations.", "text": "\u00b7 Non-Contrastive Methods: Non-contrastive methods prevent the energy landscape's collapse by limiting the volume of space that can take low energy, either through architectural constraints or through a regularizer in the energy or training objective. In latent-variable generative architectures, preventing collapse is achieved by limiting or minimizing the information content of the latent variable. In JEA, collapse is prevented by maximizing the information content of the representations.", "enumerated": false, "marker": "-"}

2aed7c8d-402c-44e6-a193-bb9f8131058d

2304.09355v5.pdf

text

We now present a few concrete examples of popular models that employ various combinations of generative architectures, joint embedding architectures, contrastive training, and noncontrastive training:

image/png

application/pdf

1.0.0

{"self_ref": "#/texts/42", "parent": {"cref": "#/body"}, "children": [], "label": "text", "prov": [{"page_no": 5, "bbox": {"l": 89.00202941894531, "t": 258.313232421875, "r": 523.8259887695312, "b": 220.55511474609375, "coord_origin": "BOTTOMLEFT"}, "charspan": [0, 200]}], "orig": "We now present a few concrete examples of popular models that employ various combinations of generative architectures, joint embedding architectures, contrastive training, and noncontrastive training:", "text": "We now present a few concrete examples of popular models that employ various combinations of generative architectures, joint embedding architectures, contrastive training, and noncontrastive training:"}