The DOI matches its Crossref record; 30 of the 30 references it cites were found in a registry or an index.
Towards Understanding Convergence and Generalization of AdamW
Abstract
AdamW modifies Adam by adding a decoupled weight decay to decay network weights per training iteration. For adaptive algorithms, this decoupled weight decay does not affect specific optimization steps, and differs from the widely used$\ell _{2}$-regularizer which changes optimization steps via changing the first- and second-order gradient moments. Despite its great practical success, for AdamW, its convergence behavior and generalization improvement over Adam and$\ell _{2}$-regularized Adam ($\ell _{2}$-Adam) remain absent yet. To solve this issue, we prove the convergence of AdamW and justify its generalization advantages over Adam and$\ell _{2}$-Adam. Specifically, AdamW provably converges but minimizes a dynamically regularized loss that combines vanilla loss and a dynamical regularization induced by decoupled weight decay, thus yielding different behaviors with Adam and$\ell _{2}$-Adam. Moreover, on both general nonconvex problems and PŁ-conditioned problems, we establish stochastic gradient complexity of AdamW to find a stationary point. Such complexity is also applicable to Adam and$\ell _{2}$-Adam, and improves their previously known complexity, especially for over-parametrized networks. Besides, we prove that AdamW enjoys smaller generalization errors than Adam and$\ell _{2}$-Adam from the Bayesian posterior aspect. This result, for the first time, explicitly reveals the benefits of decoupled weight decay in AdamW. Experimental results validate our theory.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148
ImageNet: A large-scale hierarchical image database
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009) · 2009 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2009.5206848 · Cited by 65,333
An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., et al. (2020) · arXiv · doi:10.48550/arxiv.2010.11929 · Cited by 47,472
Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification
He, K., Zhang, X., Ren, S., & Sun, J. (2015) · 2015 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2015.123 · Cited by 19,741
A Stochastic Approximation Method
Robbins, H., & Monro, S. (1951) · The Annals of Mathematical Statistics · doi:10.1214/aoms/1177729586 · Cited by 9,643
Liu, Z., Mao, H., Wu, C.-Y., Feichtenhofer, C., Darrell, T., & Xie, S. (2022) · 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr52688.2022.01167 · Cited by 8,494
Emerging Properties in Self-Supervised Vision Transformers
Caron, M., Touvron, H., Misra, I., Jegou, H., Mairal, J., Bojanowski, P., & Joulin, A. (2021) · 2021 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00951 · Cited by 5,638 · cited 1×
“Here we allow λ k to evolve along iteration number k , as in practice, an evolving λ k often shows better performance than a fixed one [4], [31], [32], [33].”
Measurement Reliability and Validity
(2000) · Research Methods in Applied Settings · doi:10.4324/9781410605337-29 · Cited by 2,541
Convolutional Neural Networks for Speech Recognition
Abdel-Hamid, O., Mohamed, A.-R., Jiang, H., Deng, L., Penn, G., & Yu, D. (2014) · IEEE/ACM Transactions on Audio, Speech, and Language Processing · doi:10.1109/taslp.2014.2339736 · Cited by 2,262
Estimation of Dependences Based on Empirical Data
Vapnik, V. (2006) · Information Science and Statistics · doi:10.1007/0-387-34239-7 · Cited by 2,238
On the Convergence of Adam and Beyond
Reddi, S. J., Kale, S., & Kumar, S. (2019) · arXiv · doi:10.48550/arxiv.1904.09237 · Cited by 1,771 · cited 2×
“Onconvexproblems,Adam-type algorithms,e.g.,AdamandAMSGrad[19],enjoytheregret O ( √ T ) under the online learning setting with training iteration number T .”
BackgroundMethodAn Introduction to Numerical Analysis
Süli, E., & Mayers, D. F. (2003) · Cambridge University Press eBooks · doi:10.1017/cbo9780511801181 · Cited by 1,255
Empirical Analysis of the Hessian of Over-Parametrized Neural NetworksPreprint
Sagun, L., Evci, U., Güney, V. U., Dauphin, Y., & Bottou, L. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1706.04454 · Cited by 486
Stochastic Gradient Descent as Approximate Bayesian Inference
Mandt, S., Hoffman, M. D., & Blei, D. M. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1704.04289 · Cited by 369
Eigenvalues of the Hessian in Deep Learning: Singularity and BeyondPreprint
Sagun, L., Bottou, L., & LeCun, Y. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1611.07476 · Cited by 329
Lower bounds for non-convex stochastic optimization
Arjevani, Y., Carmon, Y., Duchi, J. C., Foster, D. J., Srebro, N., & Woodworth, B. (2022) · Mathematical Programming · doi:10.1007/s10107-022-01822-7 · Cited by 244
PyHessian: Neural Networks Through the Lens of the Hessian
Yao, Z., Gholami, A., Keutzer, K., & Mahoney, M. W. (2020) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata50022.2020.9378171 · Cited by 227
Highly Efficient Salient Object Detection with 100K Parameters
Gao, S.-H., Tan, Y.-Q., Cheng, M.-M., Lu, C., Chen, Y., & Yan, S. (2020) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-58539-6_42 · Cited by 220
Self-citationAdan: Adaptive Nesterov Momentum Algorithm for Faster Optimizing Deep Models
Xie, X., Zhou, P., Li, H., Lin, Z., & Yan, S. (2024) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2024.3423382 · Cited by 220
Self-citationThree Factors Influencing Minima in SGD
Jastrzȩbski, S., Kenton, Z., Arpit, D., Ballas, N., Fischer, A., Bengio, Y., & Storkey, A. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1711.04623 · Cited by 217
L2 Regularization versus Batch and Weight Normalization
van Laarhoven, T. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1706.05350 · Cited by 188
McAllester, D. A. (1999) · Machine Learning · doi:10.1023/a:1007618624809 · Cited by 162
An Investigation into Neural Net Optimization via Hessian Eigenvalue Density
Ghorbani, B., Krishnan, S., & Xiao, Y. (2019) · arXiv · doi:10.48550/arxiv.1901.10159 · Cited by 144 · cited 3×
“Meanwhile, generally, we have x of Hessian matrix in deep networks are much smaller than one which is well observed in many works, e.g., fully connected networks, AlexNet, VGG and ResNet [49], [59], [60], [61] and our experimental results on ResNet50 and ViT-small in Fig.”
BackgroundResultAlgorithm-Dependent Generalization Bounds for Multi-Task Learning
Liu, T., Tao, D., Song, M., & Maybank, S. J. (2016) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2016.2544314 · Cited by 131
Mugs: A Multi-Granular Self-Supervised Learning FrameworkPreprint
Zhou, P., Zhou, Y., Si, C., Yu, W., Ng, T. K., & Yan, S. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2203.14415 · Cited by 75
Self-citationSankar, A. R., Khasbage, Y., Vigneswaran, R., & N Balasubramanian, V. (2021) · AAAI Publications (The Association for the Advancement of Artificial Intelligence (AAAI)) · doi:10.1609/aaai.v35i11.17142 · Cited by 26
Understanding Decoupled and Early Weight Decay
Björck, J., Weinberger, K. Q., & Gomes, C. P. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i8.16837 · Cited by 25
FixNorm: Dissecting Weight Decay for Training Deep Neural NetworksPreprint
Zhou, Y., Sun, Y., & Zhong, Z. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.15345 · Cited by 4
Zhou, P., Yuan, X., Lin, Z., & Hoi, S. C. H. (2021) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2021.3087328 · Cited by 3
Self-citation
Metadata
- Authors
- Zhou, P., Xie, X., Lin, Z., & Yan, S.
- Venue
- IEEE Transactions on Pattern Analysis and Machine Intelligence
- Published
- 2024-09
- Volume
- 46
- Issue
- 9
- Pages
- 6486-6493
- ISSN
- 0162-8828 (ISSN-L)
- 2160-9292
- 1939-3539
- OpenAlex
- W4393241094
- Cited by
- 329
- References
- 30
- FWCI
- 50.15
- Type
- journal article
- Language
- en
- Open access
- closed
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.