Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
158 tokens/sec
GPT-4o
7 tokens/sec
Gemini 2.5 Pro Pro
45 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Dichotomy of Early and Late Phase Implicit Biases Can Provably Induce Grokking (2311.18817v2)

Published 30 Nov 2023 in cs.LG and cs.AI

Abstract: Recent work by Power et al. (2022) highlighted a surprising "grokking" phenomenon in learning arithmetic tasks: a neural net first "memorizes" the training set, resulting in perfect training accuracy but near-random test accuracy, and after training for sufficiently longer, it suddenly transitions to perfect test accuracy. This paper studies the grokking phenomenon in theoretical setups and shows that it can be induced by a dichotomy of early and late phase implicit biases. Specifically, when training homogeneous neural nets with large initialization and small weight decay on both classification and regression tasks, we prove that the training process gets trapped at a solution corresponding to a kernel predictor for a long time, and then a very sharp transition to min-norm/max-margin predictors occurs, leading to a dramatic change in test accuracy.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (68)
  1. A convergence theory for deep learning via over-parameterization. In Kamalika Chaudhuri and Ruslan Salakhutdinov, editors, Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pages 242–252. PMLR, 09–15 Jun 2019. URL https://proceedings.mlr.press/v97/allen-zhu19a.html.
  2. Implicit regularization in deep matrix factorization. Advances in Neural Information Processing Systems, 32, 2019a.
  3. Fine-grained analysis of optimization and generalization for overparameterized two-layer neural networks. In Kamalika Chaudhuri and Ruslan Salakhutdinov, editors, Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pages 322–332. PMLR, 09–15 Jun 2019b. URL https://proceedings.mlr.press/v97/arora19a.html.
  4. On exact computation with an infinitely wide neural net. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019c. URL https://proceedings.neurips.cc/paper_files/paper/2019/file/dbc4d84bfcfe2284ba11beffb853a8c4-Paper.pdf.
  5. On the rademacher complexity of linear hypothesis sets. arXiv preprint arXiv:2007.11045, 2020.
  6. Hidden progress in deep learning: Sgd learns parities near the computational limit. In S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, editors, Advances in Neural Information Processing Systems, volume 35, pages 21750–21764. Curran Associates, Inc., 2022. URL https://proceedings.neurips.cc/paper_files/paper/2022/file/884baf65392170763b27c914087bde01-Paper-Conference.pdf.
  7. Simplicity bias in transformers and their ability to learn sparse boolean functions. arXiv preprint arXiv:2211.12316, 2022.
  8. Implicit regularization for deep neural networks driven by an ornstein-uhlenbeck like process. In Conference on learning theory, pages 483–513. PMLR, 2020.
  9. Matrix completion with noise. Proceedings of the IEEE, 98(6):925–936, 2010.
  10. Generalization bounds of stochastic gradient descent for wide and deep neural networks. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019. URL https://proceedings.neurips.cc/paper_files/paper/2019/file/cf9dc5e4e194fc21f397b4cac9cc3ae9-Paper.pdf.
  11. François Charton. Can transformers learn the greatest common divisor? arXiv preprint arXiv:2308.15594, 2023.
  12. Implicit bias of gradient descent for wide two-layer neural networks trained with the logistic loss. In Conference on Learning Theory, pages 1305–1338. PMLR, 2020.
  13. On lazy training in differentiable programming. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019. URL https://proceedings.neurips.cc/paper_files/paper/2019/file/ae614c557843b1df326cb29c57225459-Paper.pdf.
  14. A toy model of universality: Reverse engineering how networks learn group operations. In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett, editors, Proceedings of the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machine Learning Research, pages 6243–6267. PMLR, 23–29 Jul 2023. URL https://proceedings.mlr.press/v202/chughtai23a.html.
  15. Francis Clarke. Nonsmooth analysis in control theory: a survey. European Journal of Control, 7(2-3):145–159, 2001.
  16. Frank H Clarke. Generalized gradients and applications. Transactions of the American Mathematical Society, 205:247–262, 1975.
  17. Label noise sgd provably prefers flat global minimizers. Advances in Neural Information Processing Systems, 34:27449–27461, 2021.
  18. Unifying grokking and double descent. In NeurIPS ML Safety Workshop, 2022. URL https://openreview.net/forum?id=JqtHMZtqWm.
  19. Flat minima generalize for low-rank matrix recovery. arXiv preprint arXiv:2203.03756, 2022.
  20. Gradient descent finds global minima of deep neural networks. In Kamalika Chaudhuri and Ruslan Salakhutdinov, editors, Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pages 1675–1685. PMLR, 09–15 Jun 2019a. URL https://proceedings.mlr.press/v97/du19c.html.
  21. Gradient descent provably optimizes over-parameterized neural networks. In International Conference on Learning Representations, 2019b. URL https://openreview.net/forum?id=S1eK3i09YQ.
  22. The goldilocks zone: Towards better understanding of neural network loss landscapes. In Proceedings of the Thirty-Third AAAI Conference on Artificial Intelligence and Thirty-First Innovative Applications of Artificial Intelligence Conference and Ninth AAAI Symposium on Educational Advances in Artificial Intelligence, AAAI’19/IAAI’19/EAAI’19. AAAI Press, 2019. ISBN 978-1-57735-809-1. doi: 10.1609/aaai.v33i01.33013574. URL https://doi.org/10.1609/aaai.v33i01.33013574.
  23. Andrey Gromov. Grokking modular arithmetic. arXiv preprint arXiv:2301.02679, 2023.
  24. Why (and when) does local SGD generalize better than SGD? In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=svCcui6Drl.
  25. Implicit regularization in matrix factorization. Advances in neural information processing systems, 30, 2017.
  26. Characterizing implicit bias in terms of optimization geometry. In International Conference on Machine Learning, pages 1832–1841. PMLR, 2018a.
  27. Implicit bias of gradient descent on linear convolutional networks. In S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 31. Curran Associates, Inc., 2018b. URL https://proceedings.neurips.cc/paper_files/paper/2018/file/0e98aeeb54acf612b9eb4e48a269814c-Paper.pdf.
  28. Shape matters: Understanding the implicit bias of the noise covariance. In Conference on Learning Theory, pages 2315–2357. PMLR, 2021.
  29. Delving deep into rectifiers: Surpassing human-level performance on imagenet classification. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), December 2015.
  30. Latent state models of training dynamics. arXiv preprint arXiv:2308.09543, 2023.
  31. Simple and effective regularization methods for training on noisily labeled data with generalization guarantee. In International Conference on Learning Representations, 2020. URL https://openreview.net/forum?id=Hke3gyHYwH.
  32. Neural tangent kernel: Convergence and generalization in neural networks. Advances in neural information processing systems, 31, 2018.
  33. Directional convergence and alignment in deep learning. Advances in Neural Information Processing Systems, 33:17176–17186, 2020a.
  34. Polylogarithmic width suffices for gradient descent to achieve arbitrarily small test error with shallow relu networks. In International Conference on Learning Representations, 2020b. URL https://openreview.net/forum?id=HygegyrYwH.
  35. V. Koltchinskii and D. Panchenko. Empirical Margin Distributions and Bounding the Generalization Error of Combined Classifiers. The Annals of Statistics, 30(1):1 – 50, 2002. doi: 10.1214/aos/1015362183.
  36. Grokking as the transition from lazy to rich training dynamics. arXiv preprint arXiv:2310.06110, 2023.
  37. The asymmetric maximum margin bias of quasi-homogeneous neural networks. arXiv preprint arXiv:2210.03820, 2022.
  38. Grokking in linear estimators–a solvable model that groks without understanding. arXiv preprint arXiv:2310.16441, 2023.
  39. On the training dynamics of deep networks with l⁢_⁢2𝑙_2l\_2italic_l _ 2 regularization. Advances in Neural Information Processing Systems, 33:4790–4799, 2020.
  40. Towards resolving the implicit bias of gradient descent for matrix factorization: Greedy low-rank learning. In International Conference on Learning Representations, 2021a.
  41. What happens after sgd reaches zero loss?–a mathematical framework. In International Conference on Learning Representations, 2021b.
  42. Towards understanding grokking: An effective theory of representation learning. In S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, editors, Advances in Neural Information Processing Systems, volume 35, pages 34651–34663. Curran Associates, Inc., 2022. URL https://proceedings.neurips.cc/paper_files/paper/2022/file/dfc310e81992d2e4cedc09ac47eff13e-Paper-Conference.pdf.
  43. Omnigrok: Grokking beyond algorithmic data. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=zDiHoIWa0q1.
  44. Gradient descent maximizes the margin of homogeneous neural networks. In International Conference on Learning Representations, 2020.
  45. Understanding the generalization benefit of normalization layers: Sharpness reduction. Advances in Neural Information Processing Systems, 35:34689–34708, 2022.
  46. A tale of two circuits: Grokking as competition of sparse and dense subnetworks. arXiv preprint arXiv:2303.11873, 2023.
  47. Beren Millidge. Grokking. 2022. URL https://www.beren.io/2022-01-11-Grokking-Grokking/.
  48. Foundations of Machine Learning. 2018.
  49. Implicit bias in deep linear classification: Initialization scale vs training accuracy. In H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin, editors, Advances in Neural Information Processing Systems, volume 33, pages 22182–22193. Curran Associates, Inc., 2020. URL https://proceedings.neurips.cc/paper_files/paper/2020/file/fc2022c89b61c76bbef978f1370660bf-Paper.pdf.
  50. Feature emergence via margin maximization: case studies in algebraic tasks. arXiv preprint arXiv:2311.07568, 2023.
  51. Lexicographic and depth-sensitive margins in homogeneous and non-homogeneous deep models. In International Conference on Machine Learning, pages 4683–4692. PMLR, 2019a.
  52. Convergence of gradient descent on separable data. In Kamalika Chaudhuri and Masashi Sugiyama, editors, Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, volume 89 of Proceedings of Machine Learning Research, pages 3420–3428. PMLR, 16–18 Apr 2019b. URL https://proceedings.mlr.press/v89/nacson19b.html.
  53. Progress measures for grokking via mechanistic interpretability. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=9XFSbDPmdW.
  54. Predicting grokking long before it happens: A look into the loss landscape of models which grok. arXiv preprint arXiv:2306.13253, 2023.
  55. Grokking: Generalization beyond overfitting on small algorithmic datasets. arXiv preprint arXiv:2201.02177, 2022.
  56. Mechanism of feature learning in deep fully connected networks and kernel machines that recursively learn features. arXiv preprint arXiv:2212.13881, 2022.
  57. Implicit regularization in deep learning may not be explainable by norms. Advances in neural information processing systems, 33:21174–21187, 2020.
  58. The implicit bias of gradient descent on separable data. The Journal of Machine Learning Research, 19(1):2822–2878, 2018.
  59. Matus Telgarsky. Feature selection and low test error in shallow low-rotation reLU networks. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=swEskiem99.
  60. The slingshot mechanism: An empirical study of adaptive optimizers and the grokking phenomenon. arXiv preprint arXiv:2206.04817, 2022.
  61. Explaining grokking through circuit efficiency. arXiv preprint arXiv:2309.02390, 2023.
  62. The implicit bias for adaptive optimization algorithms on homogeneous neural networks. In International Conference on Machine Learning, pages 10849–10858. PMLR, 2021.
  63. G Alistair Watson. Characterization of the subdifferential of some matrix norms. Linear algebra and its applications, 170(0):33–45, 1992.
  64. How does sharpness-aware minimization minimize sharpness? arXiv preprint arXiv:2211.05729, 2022.
  65. Kernel and rich regimes in overparametrized models. In Conference on Learning Theory, pages 3635–3673. PMLR, 2020.
  66. Benign overfitting and grokking in relu networks for xor cluster data. arXiv preprint arXiv:2310.02541, 2023.
  67. Gradient descent optimizes over-parameterized deep ReLU networks. Machine learning, 109:467–492, 2020.
  68. Grokking phase transitions in learning local rules with gradient descent. arXiv preprint arXiv:2210.15435, 2022.
Citations (25)
View on Semantic Scholar

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now
Github Logo Streamline Icon: https://streamlinehq.com

GitHub

  1. GitHub - vfleaking/grokking-dichotomy (10 stars)