Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Learning with Noisy Foundation Models (2403.06869v1)

Published 11 Mar 2024 in cs.LG, cs.AI, cs.CL, and cs.CV

Abstract: Foundation models are usually pre-trained on large-scale datasets and then adapted to downstream tasks through tuning. However, the large-scale pre-training datasets, often inaccessible or too expensive to handle, can contain label noise that may adversely affect the generalization of the model and pose unexpected risks. This paper stands out as the first work to comprehensively understand and analyze the nature of noise in pre-training datasets and then effectively mitigate its impacts on downstream tasks. Specifically, through extensive experiments of fully-supervised and image-text contrastive pre-training on synthetic noisy ImageNet-1K, YFCC15M, and CC12M datasets, we demonstrate that, while slight noise in pre-training can benefit in-domain (ID) performance, where the training and testing data share a similar distribution, it always deteriorates out-of-domain (OOD) performance, where training and testing distributions are significantly different. These observations are agnostic to scales of pre-training datasets, pre-training noise types, model architectures, pre-training objectives, downstream tuning methods, and downstream applications. We empirically ascertain that the reason behind this is that the pre-training noise shapes the feature space differently. We then propose a tuning method (NMTune) to affine the feature space to mitigate the malignant effect of noise and improve generalization, which is applicable in both parameter-efficient and black-box tuning manners. We additionally conduct extensive experiments on popular vision and LLMs, including APIs, which are supervised and self-supervised pre-trained on realistic noisy data for evaluation. Our analysis and results demonstrate the importance of this novel and fundamental research direction, which we term as Noisy Model Learning.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (158)
  1. R. Bommasani, D. A. Hudson, E. Adeli, R. Altman, S. Arora, S. von Arx, M. S. Bernstein, J. Bohg, A. Bosselut, E. Brunskill et al., “On the opportunities and risks of foundation models,” arXiv preprint arXiv:2108.07258, 2021.
  2. S. Kornblith, J. Shlens, and Q. V. Le, “Do better imagenet models transfer better?” in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).   IEEE, Jun 2019.
  3. E. J. Hu, Y. Shen, P. Wallis, Z. Allen-Zhu, Y. Li, S. Wang, L. Wang, and W. Chen, “Lora: Low-rank adaptation of large language models,” arXiv preprint arXiv:2106.09685, 2021.
  4. J. He, C. Zhou, X. Ma, T. Berg-Kirkpatrick, and G. Neubig, “Towards a unified view of parameter-efficient transfer learning,” arXiv preprint arXiv:2110.04366, 2021.
  5. K. He, H. Fan, Y. Wu, S. Xie, and R. Girshick, “Momentum contrast for unsupervised visual representation learning,” arXiv preprint arXiv:1911.05722, 2019.
  6. A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning transferable visual models from natural language supervision,” in ICML, 2021.
  7. K. He, X. Chen, S. Xie, Y. Li, P. Dollar, and R. Girshick, “Masked autoencoders are scalable vision learners,” 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Jun 2022.
  8. T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell et al., “Language models are few-shot learners,” Advances in neural information processing systems, vol. 33, pp. 1877–1901, 2020.
  9. C. Schuhmann, R. Beaumont, R. Vencu, C. W. Gordon, R. Wightman, M. Cherti, T. Coombes, A. Katta, C. Mullis, M. Wortsman, P. Schramowski, S. R. Kundurthy, K. Crowson, L. Schmidt, R. Kaczmarczyk, and J. Jitsev, “LAION-5b: An open large-scale dataset for training next generation image-text models,” in Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track, 2022.
  10. T. Computer, “Redpajama: an open dataset for training large language models,” 2023. [Online]. Available: https://github.com/togethercomputer/RedPajama-Data
  11. K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in CVPR, 2016.
  12. A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly et al., “An image is worth 16x16 words: Transformers for image recognition at scale,” arXiv preprint arXiv:2010.11929, 2020.
  13. Z. Liu, Y. Lin, Y. Cao, H. Hu, Y. Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” 2021 IEEE/CVF International Conference on Computer Vision (ICCV), Oct 2021.
  14. Z. Liu, H. Mao, C.-Y. Wu, C. Feichtenhofer, T. Darrell, and S. Xie, “A convnet for the 2020s,” 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Jun 2022.
  15. A. Kolesnikov, L. Beyer, X. Zhai, J. Puigcerver, J. Yung, S. Gelly, and N. Houlsby, “Big transfer (bit): General visual representation learning,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part V 16.   Springer, 2020, pp. 491–507.
  16. Q. Xie, M.-T. Luong, E. Hovy, and Q. V. Le, “Self-training with noisy student improves imagenet classification,” 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Jun 2020.
  17. T. Ridnik, E. Ben-Baruch, A. Noy, and L. Zelnik-Manor, “Imagenet-21k pretraining for the masses,” in Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 1), 2021.
  18. M. Byeon, B. Park, H. Kim, S. Lee, W. Baek, and S. Kim, “Coyo-700m: Image-text pair dataset,” https://github.com/kakaobrain/coyo-dataset, 2022.
  19. S. Ren, K. He, R. Girshick, and J. Sun, “Faster r-cnn: Towards real-time object detection with region proposal networks,” Advances in neural information processing systems, vol. 28, 2015.
  20. T.-Y. Lin, P. Goyal, R. Girshick, K. He, and P. Dollár, “Focal loss for dense object detection,” in Proceedings of the IEEE international conference on computer vision, 2017, pp. 2980–2988.
  21. K. He, G. Gkioxari, P. Dollár, and R. Girshick, “Mask r-cnn,” in Proceedings of the IEEE international conference on computer vision, 2017, pp. 2961–2969.
  22. N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” in European conference on computer vision.   Springer, 2020, pp. 213–229.
  23. B. Cheng, A. Schwing, and A. Kirillov, “Per-pixel classification is not all you need for semantic segmentation,” Advances in Neural Information Processing Systems, vol. 34, pp. 17 864–17 875, 2021.
  24. OpenAI, “Gpt-4 technical report,” 2023.
  25. Google, “Gemini: a family of highly capable multimodal models,” arXiv preprint arXiv:2312.11805, 2023.
  26. J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,” arXiv preprint arXiv:1810.04805, 2018.
  27. Y. Liu, M. Ott, N. Goyal, J. Du, M. Joshi, D. Chen, O. Levy, M. Lewis, L. Zettlemoyer, and V. Stoyanov, “Roberta: A robustly optimized bert pretraining approach,” arXiv preprint arXiv:1907.11692, 2019.
  28. A. Radford, K. Narasimhan, T. Salimans, I. Sutskever et al., “Improving language understanding by generative pre-training,” 2018.
  29. A. Radford, J. Wu, R. Child, D. Luan, D. Amodei, I. Sutskever et al., “Language models are unsupervised multitask learners,” OpenAI blog, vol. 1, no. 8, p. 9, 2019.
  30. H. Touvron, T. Lavril, G. Izacard, X. Martinet, M.-A. Lachaux, T. Lacroix, B. Rozière, N. Goyal, E. Hambro, F. Azhar et al., “Llama: Open and efficient foundation language models,” arXiv preprint arXiv:2302.13971, 2023.
  31. M. Chen, K. Goel, N. S. Sohoni, F. Poms, K. Fatahalian, and C. Ré, “Mandoline: Model evaluation under distribution shift,” in International conference on machine learning.   PMLR, 2021, pp. 1617–1629.
  32. A. Kumar, A. Raghunathan, R. Jones, T. Ma, and P. Liang, “Fine-tuning can distort pretrained features and underperform out-of-distribution,” arXiv preprint arXiv:2202.10054, 2022.
  33. K. Sohn, D. Berthelot, N. Carlini, Z. Zhang, H. Zhang, C. A. Raffel, E. D. Cubuk, A. Kurakin, and C.-L. Li, “Fixmatch: Simplifying semi-supervised learning with consistency and confidence,” Advances in Neural Information Processing Systems (NeurIPS), vol. 33, 2020.
  34. Y. Wang, H. Chen, Y. Fan, W. Sun, R. Tao, W. Hou, R. Wang, L. Yang, Z. Zhou, L.-Z. Guo, H. Qi, Z. Wu, Y.-F. Li, S. Nakamura, W. Ye, M. Savvides, B. Raj, T. Shinozaki, B. Schiele, J. Wang, X. Xie, and Y. Zhang, “Usb: A unified semi-supervised learning benchmark,” in Advances in Neural Information Processing Systems (NeurIPS), 2022.
  35. Y. Zhang, B. Kang, B. Hooi, S. Yan, and J. Feng, “Deep long-tailed learning: A survey,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023.
  36. Y. Wang, Z. Yu, J. Wang, Q. Heng, H. Chen, W. Ye, R. Xie, X. Xie, and S. Zhang, “Exploring vision-language models for imbalanced learning,” International Journal of Computer Vision (IJCV), 2023.
  37. H. Song, M. Kim, D. Park, Y. Shin, and J.-G. Lee, “Learning from noisy labels with deep neural networks: A survey,” IEEE Transactions on Neural Networks and Learning Systems, p. 1–19, 2022.
  38. S. Li, X. Xia, S. Ge, and T. Liu, “Selective-supervised contrastive learning with noisy labels,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 316–325.
  39. S. Y. Gadre, G. Ilharco, A. Fang, J. Hayase, G. Smyrnis, T. Nguyen, R. Marten, M. Wortsman, D. Ghosh, J. Zhang, E. Orgad, R. Entezari, G. Daras, S. Pratt, V. Ramanujan, Y. Bitton, K. Marathe, S. Mussmann, R. Vencu, M. Cherti, R. Krishna, P. W. Koh, O. Saukh, A. J. Ratner, S. Song, H. Hajishirzi, A. Farhadi, R. Beaumont, S. Oh, A. G. Dimakis, J. Jitsev, Y. Carmon, V. Shankar, and L. Schmidt, “Datacomp: In search of the next generation of multimodal datasets,” ArXiv, 2023.
  40. R. Entezari, M. Wortsman, O. Saukh, M. M. Shariatnia, H. Sedghi, and L. Schmidt, “The role of pre-training data in transfer learning,” arXiv preprint arXiv:2302.13602, 2023.
  41. J. Zhang, B. Wang, Z. Hu, P. W. Koh, and A. Ratner, “On the trade-off of intra-/inter-class diversity for supervised pre-training,” arXiv preprint arXiv:2305.12224, 2023.
  42. T. Nguyen, G. Ilharco, M. Wortsman, S. Oh, and L. Schmidt, “Quality not quantity: On the interaction between dataset design and robustness of clip,” Advances in Neural Information Processing Systems, vol. 35, pp. 21 455–21 469, 2022.
  43. K. Lee, D. Ippolito, A. Nystrom, C. Zhang, D. Eck, C. Callison-Burch, and N. Carlini, “Deduplicating training data makes language models better,” in Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (ACL), 2022.
  44. S. Gunasekar, Y. Zhang, J. Aneja, C. C. T. Mendes, A. Del Giorno, S. Gopi, M. Javaheripi, P. Kauffmann, G. de Rosa, O. Saarikivi et al., “Textbooks are all you need,” arXiv preprint arXiv:2306.11644, 2023.
  45. S. Longpre, G. Yauney, E. Reif, K. Lee, A. Roberts, B. Zoph, D. Zhou, J. Wei, K. Robinson, D. Mimno et al., “A pretrainer’s guide to training data: Measuring the effects of data age, domain coverage, quality, & toxicity,” arXiv preprint arXiv:2305.13169, 2023.
  46. S. L. Blodgett and B. O’Connor, “Racial disparity in natural language processing: A case study of social media african-american english,” arXiv preprint arXiv:1707.00061, 2017.
  47. H. Chang, T. D. Nguyen, S. K. Murakonda, E. Kazemi, and R. Shokri, “On adversarial bias and the robustness of fair machine learning,” arXiv preprint arXiv:2006.08669, 2020.
  48. A. Birhane, V. Prabhu, S. Han, V. N. Boddeti, and A. S. Luccioni, “Into the laions den: Investigating hate in multimodal datasets,” arXiv preprint arXiv:2311.03449, 2023.
  49. Y. Elazar, A. Bhagia, I. Magnusson, A. Ravichander, D. Schwenk, A. Suhr, P. Walsh, D. Groeneveld, L. Soldaini, S. Singh et al., “What’s in my big data?” arXiv preprint arXiv:2310.20707, 2023.
  50. N. Carlini, D. Ippolito, M. Jagielski, K. Lee, F. Tramer, and C. Zhang, “Quantifying memorization across neural language models,” in The Eleventh International Conference on Learning Representations, 2022.
  51. D. Thiel, “Identifying and eliminating csam in generative ml training data and models,” Technical report, Stanford University, Palo Alto, CA, 2023, Tech. Rep., 2023. [Online]. Available: https://doi.org/10.25740/kh752sm9123
  52. V. Vasudevan, B. Caine, R. Gontijo-Lopes, S. Fridovich-Keil, and R. Roelofs, “When does dough become a bagel? analyzing the remaining mistakes on imagenet,” arXiv preprint arXiv:2205.04596, 2022.
  53. M. Cherti, R. Beaumont, R. Wightman, M. Wortsman, G. Ilharco, C. Gordon, C. Schuhmann, L. Schmidt, and J. Jitsev, “Reproducible scaling laws for contrastive language-image learning,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 2818–2829.
  54. Z. Yang, Z. Dai, Y. Yang, J. Carbonell, R. R. Salakhutdinov, and Q. V. Le, “Xlnet: Generalized autoregressive pretraining for language understanding,” Advances in neural information processing systems, vol. 32, 2019.
  55. A. Ghosh, H. Kumar, and P. S. Sastry, “Robust loss functions under label noise for deep neural networks,” in Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), 2017.
  56. J. Li, R. Socher, and S. C. Hoi, “Dividemix: Learning with noisy labels as semi-supervised learning,” in International Conference on Learning Representations (ICLR), 2020.
  57. C. G. Northcutt, L. Jiang, and I. L. Chuang, “Confident learning: Estimating uncertainty in dataset labels,” Journal of Artificial Intelligence Research (JAIR), vol. 70, pp. 1373–1411, 2021.
  58. H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y. Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosale et al., “Llama 2: Open foundation and fine-tuned chat models,” arXiv preprint arXiv:2307.09288, 2023.
  59. H. Chen, J. Wang, A. Shah, R. Tao, H. Wei, X. Xie, M. Sugiyama, and B. Raj, “Understanding and mitigating the label noise in pre-training on downstream tasks,” arXiv preprint arXiv:2309.17002, 2023.
  60. B. Thomee, D. A. Shamma, G. Friedland, B. Elizalde, K. Ni, D. Poland, D. Borth, and L.-J. Li, “Yfcc100m,” Communications of the ACM, vol. 59, no. 2, p. 64–73, Jan 2016.
  61. S. Changpinyo, P. Sharma, N. Ding, and R. Soricut, “Conceptual 12M: Pushing web-scale image-text pre-training to recognize long-tail visual concepts,” in CVPR, 2021.
  62. X. Chen, S. Wang, M. Long, and J. Wang, “Transferability vs. discriminability: Batch spectral penalization for adversarial domain adaptation,” in Proceedings of the 36th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, K. Chaudhuri and R. Salakhutdinov, Eds., vol. 97.   PMLR, 09–15 Jun 2019, pp. 1081–1090.
  63. Z. Zhang and M. Sabuncu, “Generalized cross entropy loss for training deep neural networks with noisy labels,” Advances in Neural Information Processing Systems (NeurIPS), vol. 31, 2018.
  64. Y. Wang, X. Ma, Z. Chen, Y. Luo, J. Yi, and J. Bailey, “Symmetric cross entropy for robust learning with noisy labels,” Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp. 322–330, 2019.
  65. X. Ma, H. Huang, Y. Wang, S. Romano, S. M. Erfani, and J. Bailey, “Normalized loss functions for deep learning with noisy labels,” in Proceedings of the International Conference on Machine Learning (ICML), 2020.
  66. T. Xiao, T. Xia, Y. Yang, C. Huang, and X. Wang, “Learning from massive noisy labeled data for image classification,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 2691–2699.
  67. J. Goldberger and E. Ben-Reuven, “Training deep neural-networks using a noise adaptation layer,” in International Conference on Learning Representations (ICLR), 2016.
  68. S. Liu, J. Niles-Weed, N. Razavian, and C. Fernandez-Granda, “Early-learning regularization prevents memorization of noisy labels,” Advances in Neural Information Processing Systems (NeurIPS), vol. 33, 2020.
  69. X. Li, T. Liu, B. Han, G. Niu, and M. Sugiyama, “Provably end-to-end label-noise learning without anchor points,” in Proceedings of 38th International Conference on Machine Learning (ICML2021), 2021, pp. 6403–6413.
  70. B. Han, Q. Yao, X. Yu, G. Niu, M. Xu, W. Hu, I. W.-H. Tsang, and M. Sugiyama, “Co-teaching: Robust training of deep neural networks with extremely noisy labels,” Advances in Neural Information Processing Systems (NeurIPS), 2018.
  71. Y. Zhang, G. Niu, and M. Sugiyama, “Learning noise transition matrix from only noisy labels via total variation regularization,” in International Conference on Machine Learning.   PMLR, 2021, pp. 12 501–12 512.
  72. S. Liu, Z. Zhu, Q. Qu, and C. You, “Robust training under label noise by over-parameterization,” in Proceedings of the International Conference on Machine Learning (ICML), K. Chaudhuri, S. Jegelka, L. Song, C. Szepesvari, G. Niu, and S. Sabato, Eds., vol. 162.   PMLR, 17–23 Jul 2022, pp. 14 153–14 172.
  73. T. Kim, J. Ko, J. Choi, S.-Y. Yun et al., “Fine samples for learning with noisy labels,” Advances in Neural Information Processing Systems, vol. 34, pp. 24 137–24 149, 2021.
  74. H. Chen, A. Shah, J. Wang, R. Tao, Y. Wang, X. Xie, M. Sugiyama, R. Singh, and B. Raj, “Imprecise label learning: A unified framework for learning with various imprecise label configurations,” arXiv preprint arXiv:2305.12715, 2023.
  75. J. Wei, Z. Zhu, H. Cheng, T. Liu, G. Niu, and Y. Liu, “Learning with noisy labels revisited: A study using real-world human annotations,” in International Conference on Learning Representations (ICLR), 2022.
  76. H. Cheng, Z. Zhu, X. Sun, and Y. Liu, “Mitigating memorization of noisy labels via regularization between representations,” in International Conference on Learning Representations (ICLR), 2023.
  77. K. Wen, J. Teng, and J. Zhang, “Benign overfitting in classification: Provably counter label noise with larger models,” in The Eleventh International Conference on Learning Representations, 2022.
  78. Y. Xue, K. Whitecross, and B. Mirzasoleiman, “Investigating why contrastive learning benefits robustness against label noise,” in International Conference on Machine Learning.   PMLR, 2022, pp. 24 851–24 871.
  79. B. Zhang, Y. Wang, W. Hou, H. Wu, J. Wang, M. Okumura, and T. Shinozaki, “Flexmatch: Boosting semi-supervised learning with curriculum pseudo labeling,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 34, 2021.
  80. Y. Wang, H. Chen, Q. Heng, W. Hou, Y. Fan, Z. Wu, J. Wang, M. Savvides, T. Shinozaki, B. Raj, B. Schiele, and X. Xie, “Freematch: Self-adaptive thresholding for semi-supervised learning,” in International Conference on Learning Representations (ICLR), 2023.
  81. B. Kang, S. Xie, M. Rohrbach, Z. Yan, A. Gordo, J. Feng, and Y. Kalantidis, “Decoupling representation and classifier for long-tailed recognition,” in International Conference on Learning Representations, 2019.
  82. Y. Wang, B. Zhang, W. Hou, Z. Wu, J. Wang, and T. Shinozaki, “Margin calibration for long-tailed visual recognition,” in Asian Conference on Machine Learning.   PMLR, 2023, pp. 1101–1116.
  83. J. Wei, H. Liu, T. Liu, G. Niu, M. Sugiyama, and Y. Liu, “To smooth or not? when label smoothing meets noisy labels,” in International Conference on Machine Learning (ICML), 2022.
  84. J. Kaplan, S. McCandlish, T. Henighan, T. B. Brown, B. Chess, R. Child, S. Gray, A. Radford, J. Wu, and D. Amodei, “Scaling laws for neural language models,” arXiv preprint arXiv:2001.08361, 2020.
  85. I. Magar and R. Schwartz, “Data contamination: From memorization to exploitation,” in Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), 2022, pp. 157–165.
  86. Y.-Y. Tsai, P.-Y. Chen, and T.-Y. Ho, “Transfer learning without knowing: Reprogramming black-box machine learning models with scarce data and limited resources,” in International Conference on Machine Learning.   PMLR, 2020, pp. 9614–9624.
  87. M. Wortsman, G. Ilharco, J. W. Kim, M. Li, S. Kornblith, R. Roelofs, R. G. Lopes, H. Hajishirzi, A. Farhadi, H. Namkoong et al., “Robust fine-tuning of zero-shot models,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 7959–7971.
  88. S. Goyal, A. Kumar, S. Garg, Z. Kolter, and A. Raghunathan, “Finetune like you pretrain: Improved finetuning of zero-shot vision models,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 19 338–19 347.
  89. S. Xu, J. Yao, R. Luo, S. Zhang, Z. Lian, M. Tan, and Y. Wang, “Towards efficient task-driven model reprogramming with foundation models,” arXiv preprint arXiv:2304.02263, 2023.
  90. C. Oh, H. Hwang, H.-y. Lee, Y. Lim, G. Jung, J. Jung, H. Choi, and K. Song, “Blackvip: Black-box visual prompting for robust transfer learning,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 24 224–24 235.
  91. N. Houlsby, A. Giurgiu, S. Jastrzebski, B. Morrone, Q. De Laroussilhe, A. Gesmundo, M. Attariyan, and S. Gelly, “Parameter-efficient transfer learning for nlp,” in International Conference on Machine Learning.   PMLR, 2019, pp. 2790–2799.
  92. X. Liu, K. Ji, Y. Fu, W. Tam, Z. Du, Z. Yang, and J. Tang, “P-tuning: Prompt tuning can be comparable to fine-tuning across scales and tasks,” in Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), 2022, pp. 61–68.
  93. X. Liu, K. Ji, Y. Fu, W. L. Tam, Z. Du, Z. Yang, and J. Tang, “P-tuning v2: Prompt tuning can be comparable to fine-tuning universally across scales and tasks,” arXiv preprint arXiv:2110.07602, 2021.
  94. N. Kandpal, H. Deng, A. Roberts, E. Wallace, and C. Raffel, “Large language models struggle to learn long-tail knowledge,” in International Conference on Machine Learning.   PMLR, 2023, pp. 15 696–15 707.
  95. H. Kotek, R. Dockum, and D. Q. Sun, “Gender bias in llms,” 2023. [Online]. Available: https://arxiv.org/abs/2308.14921
  96. J. A. Omiye, J. C. Lester, S. Spichak, V. Rotemberg, and R. Daneshjou, “Large language models propagate race-based medicine,” NPJ Digital Medicine, vol. 6, no. 1, p. 195, 2023.
  97. A. Wei, N. Haghtalab, and J. Steinhardt, “Jailbroken: How does llm safety training fail?” arXiv preprint arXiv:2307.02483, 2023.
  98. A. Zou, Z. Wang, J. Z. Kolter, and M. Fredrikson, “Universal and transferable adversarial attacks on aligned language models,” arXiv preprint arXiv:2307.15043, 2023.
  99. B. Recht, R. Roelofs, L. Schmidt, and V. Shankar, “Do imagenet classifiers generalize to imagenet?” in International conference on machine learning.   PMLR, 2019, pp. 5389–5400.
  100. P. Nakkiran, G. Kaplun, Y. Bansal, T. Yang, B. Barak, and I. Sutskever, “Deep double descent: where bigger models and more data hurt,” Journal of Statistical Mechanics: Theory and Experiment, vol. 2021, 2019. [Online]. Available: https://api.semanticscholar.org/CorpusID:207808916
  101. O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, A. C. Berg, and L. Fei-Fei, “ImageNet Large Scale Visual Recognition Challenge,” International Journal of Computer Vision (IJCV), vol. 115, no. 3, pp. 211–252, 2015.
  102. C. Jia, Y. Yang, Y. Xia, Y.-T. Chen, Z. Parekh, H. Pham, Q. Le, Y.-H. Sung, Z. Li, and T. Duerig, “Scaling up visual and vision-language representation learning with noisy text supervision,” in International conference on machine learning.   PMLR, 2021, pp. 4904–4916.
  103. K. Desai, G. Kaul, Z. Aysola, and J. Johnson, “RedCaps: Web-curated image-text data created by the people, for the people,” in NeurIPS Datasets and Benchmarks, 2021.
  104. C. Schuhmann, R. Vencu, R. Beaumont, R. Kaczmarczyk, C. Mullis, A. Katta, T. Coombes, J. Jitsev, and A. Komatsuzaki, “Laion-400m: Open dataset of clip-filtered 400 million image-text pairs,” arXiv preprint arXiv:2111.02114, 2021.
  105. L. Beyer, O. J. Hénaff, A. Kolesnikov, X. Zhai, and A. v. d. Oord, “Are we done with imagenet?” arXiv preprint arXiv:2006.07159, 2020.
  106. S. Yun, S. J. Oh, B. Heo, D. Han, J. Choe, and S. Chun, “Re-labeling imagenet: from single to multi-labels, from global to localized labels,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2021.
  107. J. Djolonga, J. Yung, M. Tschannen, R. Romijnders, L. Beyer, A. Kolesnikov, J. Puigcerver, M. Minderer, A. D’Amour, D. Moldovan, S. Gelly, N. Houlsby, X. Zhai, and M. Lucic, “On robustness and transferability of convolutional neural networks,” 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Jun 2021.
  108. H. Liu, J. Z. HaoChen, A. Gaidon, and T. Ma, “Self-supervised learning is more robust to dataset imbalance,” in International Conference on Learning Representations, 2021.
  109. T.-Y. Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Dollár, and C. L. Zitnick, “Microsoft coco: Common objects in context,” in Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13.   Springer, 2014, pp. 740–755.
  110. Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE, vol. 86, no. 11, pp. 2278–2324, 1998.
  111. A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information processing systems, vol. 30, 2017.
  112. R. Wightman, H. Touvron, and H. Jégou, “Resnet strikes back: An improved training procedure in timm,” arXiv preprint arXiv:2110.00476, 2021.
  113. Y. You, J. Li, S. Reddi, J. Hseu, S. Kumar, S. Bhojanapalli, X. Song, J. Demmel, K. Keutzer, and C.-J. Hsieh, “Large batch optimization for deep learning: Training bert in 76 minutes,” arXiv preprint arXiv:1904.00962, 2019.
  114. T. He, Z. Zhang, H. Zhang, Z. Zhang, J. Xie, and M. Li, “Bag of tricks for image classification with convolutional neural networks,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2019.
  115. G. Huang, Y. Sun, Z. Liu, D. Sedra, and K. Q. Weinberger, “Deep networks with stochastic depth,” in Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part IV 14.   Springer, 2016, pp. 646–661.
  116. M. Berman, H. Jégou, A. Vedaldi, I. Kokkinos, and M. Douze, “Multigrain: a unified image embedding for classes and instances,” arXiv preprint arXiv:1902.05509, 2019.
  117. E. D. Cubuk, B. Zoph, J. Shlens, and Q. V. Le, “Randaugment: Practical automated data augmentation with a reduced search space,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshops, 2020, pp. 702–703.
  118. H. Zhang, M. Cisse, Y. N. Dauphin, and D. Lopez-Paz, “mixup: Beyond empirical risk minimization,” arXiv preprint arXiv:1710.09412, 2017.
  119. S. Yun, D. Han, S. J. Oh, S. Chun, J. Choe, and Y. Yoo, “Cutmix: Regularization strategy to train strong classifiers with localizable features,” in Proceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 6023–6032.
  120. H. Touvron, M. Cord, and H. Jégou, “Deit iii: Revenge of the vit,” in European Conference on Computer Vision.   Springer, 2022, pp. 516–533.
  121. H. Touvron, M. Cord, A. Sablayrolles, G. Synnaeve, and H. Jégou, “Going deeper with image transformers,” in Proceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 32–42.
  122. A. Krizhevsky, G. Hinton et al., “Learning multiple layers of features from tiny images,” 2009.
  123. M.-E. Nilsback and A. Zisserman, “Automated flower classification over a large number of classes,” in 2008 Sixth Indian conference on computer vision, graphics & image processing.   IEEE, 2008, pp. 722–729.
  124. L. Bossard, M. Guillaumin, and L. Van Gool, “Food-101 – mining discriminative components with random forests,” in European Conference on Computer Vision, 2014.
  125. O. M. Parkhi, A. Vedaldi, A. Zisserman, and C. V. Jawahar, “Cats and dogs,” in IEEE Conference on Computer Vision and Pattern Recognition, 2012.
  126. J. Krause, M. Stark, J. Deng, and L. Fei-Fei, “3d object representations for fine-grained categorization,” in 2013 IEEE International Conference on Computer Vision Workshops, 2013, pp. 554–561.
  127. S. Maji, E. Rahtu, J. Kannala, M. Blaschko, and A. Vedaldi, “Fine-grained visual classification of aircraft,” arXiv preprint arXiv:1306.5151, 2013.
  128. Y. Netzer, T. Wang, A. Coates, A. Bissacco, B. Wu, and A. Y. Ng, “Reading digits in natural images with unsupervised feature learning,” in NIPS Workshop on Deep Learning and Unsupervised Feature Learning 2011, 2011.
  129. M. Cimpoi, S. Maji, I. Kokkinos, S. Mohamed, , and A. Vedaldi, “Describing textures in the wild,” in Proceedings of the IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2014.
  130. L. Fei-Fei, R. Fergus, and P. Perona, “Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories,” Computer Vision and Pattern Recognition Workshop, 2004.
  131. P. Helber, B. Bischke, A. Dengel, and D. Borth, “Introducing eurosat: A novel dataset and deep learning benchmark for land use and land cover classification,” in IGARSS 2018-2018 IEEE International Geoscience and Remote Sensing Symposium.   IEEE, 2018, pp. 204–207.
  132. ——, “Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2019.
  133. B. S. Veeling, J. Linmans, J. Winkens, T. Cohen, and M. Welling, “Rotation equivariant CNNs for digital pathology,” Jun. 2018.
  134. G. Cheng, J. Han, and X. Lu, “Remote sensing image scene classification: Benchmark and state of the art,” Proceedings of the IEEE, vol. 105, no. 10, pp. 1865–1883, Oct 2017.
  135. R. Socher, A. Perelygin, J. Wu, J. Chuang, C. D. Manning, A. Ng, and C. Potts, “Recursive deep models for semantic compositionality over a sentiment treebank,” in Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing.   Seattle, Washington, USA: Association for Computational Linguistics, Oct. 2013, pp. 1631–1642.
  136. X. Peng, Q. Bai, X. Xia, Z. Huang, K. Saenko, and B. Wang, “Moment matching for multi-source domain adaptation,” in Proceedings of the IEEE International Conference on Computer Vision, 2019, pp. 1406–1415.
  137. D. Hendrycks, K. Zhao, S. Basart, J. Steinhardt, and D. Song, “Natural adversarial examples,” 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Jun 2021.
  138. D. Hendrycks, S. Basart, N. Mu, S. Kadavath, F. Wang, E. Dorundo, R. Desai, T. Zhu, S. Parajuli, M. Guo, D. Song, J. Steinhardt, and J. Gilmer, “The many faces of robustness: A critical analysis of out-of-distribution generalization,” 2021 IEEE/CVF International Conference on Computer Vision (ICCV), Oct 2021.
  139. H. Wang, S. Ge, E. P. Xing, and Z. C. Lipton, “Learning robust global representations by penalizing local predictive power,” in Advances in Neural Information Processing Systems (NeurIPS), 2019.
  140. V. Shankar, A. Dave, R. Roelofs, D. Ramanan, B. Recht, and L. Schmidt, “Do image classifiers generalize across time?” 2021 IEEE/CVF International Conference on Computer Vision (ICCV), Oct 2021.
  141. A. Barbu, D. Mayo, J. Alverio, W. Luo, C. Wang, D. Gutfreund, J. Tenenbaum, and B. Katz, “Objectnet: A large-scale bias-controlled dataset for pushing the limits of object recognition models,” in Advances in Neural Information Processing Systems, H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, and R. Garnett, Eds., vol. 32, 2019.
  142. D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014.
  143. I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” arXiv preprint arXiv:1711.05101, 2017.
  144. S. Mangrulkar, S. Gugger, L. Debut, Y. Belkada, S. Paul, and B. Bossan, “Peft: State-of-the-art parameter-efficient fine-tuning methods,” https://github.com/huggingface/peft, 2022.
  145. X. Wang, R. Zhang, T. Kong, L. Li, and C. Shen, “Solov2: Dynamic and fast instance segmentation,” Advances in Neural information processing systems, vol. 33, pp. 17 721–17 732, 2020.
  146. S. Oymak, Z. Fabian, M. Li, and M. Soltanolkotabi, “Generalization guarantees for neural networks via harnessing the low-rank structure of the jacobian,” arXiv preprint arXiv:1906.05392, 2019.
  147. C. Wu, F. Wu, T. Qi, and Y. Huang, “Noisytune: A little noise can help you finetune pretrained language models better,” Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), 2022.
  148. C. Zhang, S. Bengio, M. Hardt, B. Recht, and O. Vinyals, “Understanding deep learning (still) requires rethinking generalization,” Communications of the ACM, vol. 64, no. 3, pp. 107–115, 2021.
  149. J. Zbontar, L. Jing, I. Misra, Y. LeCun, and S. Deny, “Barlow twins: Self-supervised learning via redundancy reduction,” in International Conference on Machine Learning.   PMLR, 2021, pp. 12 310–12 320.
  150. A. Bardes, J. Ponce, and Y. LeCun, “Vicreg: Variance-invariance-covariance regularization for self-supervised learning,” in International Conference on Learning Representations (ICLR), 2022.
  151. M. Tan and Q. Le, “Efficientnet: Rethinking model scaling for convolutional neural networks,” in International conference on machine learning.   PMLR, 2019, pp. 6105–6114.
  152. G. Hinton, O. Vinyals, and J. Dean, “Distilling the knowledge in a neural network,” arXiv preprint arXiv:1503.02531, 2015.
  153. K. He, X. Zhang, S. Ren, and J. Sun, “Identity mappings in deep residual networks,” in Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part IV 14.   Springer, 2016, pp. 630–645.
  154. F. Chollet, “Xception: Deep learning with depthwise separable convolutions,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 1251–1258.
  155. R. Wightman, “Pytorch image models,” https://github.com/rwightman/pytorch-image-models, 2019.
  156. A. Wang, A. Singh, J. Michael, F. Hill, O. Levy, and S. Bowman, “Glue: A multi-task benchmark and analysis platform for natural language understanding,” Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, 2018.
  157. L. Yang, S. Zhang, L. Qin, Y. Li, Y. Wang, H. Liu, J. Wang, X. Xie, and Y. Zhang, “Glue-x: Evaluating natural language understanding models from an out-of-distribution generalization perspective,” in Findings of ACL, 2023.
  158. Y. Zhu, R. Kiros, R. Zemel, R. Salakhutdinov, R. Urtasun, A. Torralba, and S. Fidler, “Aligning books and movies: Towards story-like visual explanations by watching movies and reading books,” 2015 IEEE International Conference on Computer Vision (ICCV), Dec 2015.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (8)
  1. Hao Chen (1006 papers)
  2. Jindong Wang (150 papers)
  3. Zihan Wang (181 papers)
  4. Ran Tao (82 papers)
  5. Hongxin Wei (45 papers)
  6. Xing Xie (220 papers)
  7. Masashi Sugiyama (286 papers)
  8. Bhiksha Raj (180 papers)
Citations (3)
View on Semantic Scholar

Summary

Enhancements in Understanding Noise Impact on Foundation Models Across Architectures

Introduction to the Study

The researchers have expanded upon their prior work to delve into the influence of pre-training noise on foundation models. This revised paper stands out for broadening the scope of examination to include both convolutional and transformer-based architectures, specifically adding the Vision Transformer (ViT-B-16) to their analysis alongside the previously studied ResNet-50. The incorporation of these models facilitates a nuanced understanding of how pre-training noise affects various aspects of model performance across different architectures.

Key Findings and Contributions

Expanded Architectural Analysis

The extension to include ViT-B-16 markedly enhances the research, allowing for a comprehensive analysis across a broader spectrum of model architectures. This architectural inclusivity is instrumental in corroborating the architecture-agnostic nature of the observed phenomena related to pre-training noise. Through this expanded lens, the paper demonstrates the robustness and adaptability of foundation models to noise, showcasing consistent phenomena across a range of tasks including classification, detection, and segmentation.

In-depth Feature Analysis

By analyzing both ViT-B-16 and ResNet-50, the paper provides insightful revelations on how pre-training noise influences feature representation and processing within these models. This aspect is pivotal, offering a granular view of the impacts of noise and prompting a broader discussion on the resilience of model architectures to varying noise levels in the pre-training phase.

Methodological Advancements with NMTune

The refinement of the proposed NMTune methodology represents a significant breakthrough, enhancing versatility and efficacy across multiple tuning paradigms including black-box and parameter-efficient approaches. NMTune’s adaptability across different architectural frameworks underscores its potential utility in tackling practical challenges related to noise in foundation models.

Asymmetric Noise Analysis

The novel investigation into asymmetric pre-training noise unveils critical insights into the differential impact of various noise types on model learning and generalization. This segment of research challenges existing paradigms and fosters a deeper understanding of optimizing pre-training strategies to mitigate noise effects.

Implications and Future Directions

The findings from this paper have profound implications for the development and optimization of foundation models. By highlighting the resilience of such models to pre-training noise and proposing effective methods to enhance their performance, the research opens new vistas for leveraging noisy data in model training. The architectural and methodological inclusivity of this paper paves the way for future explorations into noise resilience across emerging model frameworks.

Furthermore, the insights on asymmetric pre-training noise introduce intriguing prospects for refining pre-training techniques and noise mitigation strategies. This could lead to more robust models capable of better generalization and performance in real-world applications subjected to noisy inputs.

Concluding Remarks

This enhanced paper contributes significantly to the foundational understanding of noise impacts on various models, including convolutional and transformer-based architectures. The methodological advancements, coupled with a deeper analysis of noise effects, offer valuable pathways for future research on optimizing foundation models. The versatility and broad applicability of these findings underscore the potential for practical implementations, making this paper a pivotal reference for researchers and practitioners aiming to leverage noisy data effectively in model training processes.

File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown