Pith. sign in

REVIEW 5 cited by

Understanding the Generalization of Adam in Learning Neural Networks with Proper Regularization

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2108.11371 v1 pith:YJRD63SD submitted 2021-08-25 cs.LG math.OCstat.ML

classification cs.LGmath.OCstat.ML
keywords adamlearningregularizationconvergedeepdifferentgeneralizationgradient
verification ladder T0 review T1 audit T2 compute T3 formal

Signed reviews

No signed human review yet.

0 comments
read the original abstract

Adaptive gradient methods such as Adam have gained increasing popularity in deep learning optimization. However, it has been observed that compared with (stochastic) gradient descent, Adam can converge to a different solution with a significantly worse test error in many deep learning applications such as image classification, even with a fine-tuned regularization. In this paper, we provide a theoretical explanation for this phenomenon: we show that in the nonconvex setting of learning over-parameterized two-layer convolutional neural networks starting from the same random initialization, for a class of data distributions (inspired from image data), Adam and gradient descent (GD) can converge to different global solutions of the training objective with provably different generalization errors, even with weight decay regularization. In contrast, we show that if the training objective is convex, and the weight decay regularization is employed, any optimization algorithms including Adam and GD will converge to the same solution if the training is successful. This suggests that the inferior generalization performance of Adam is fundamentally tied to the nonconvex landscape of deep learning optimization.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 5 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Representations Shape Weak-to-Strong Generalization: Theoretical Insights and Empirical Predictions

    cs.LG 2025-02 conditional novelty 8.0 of 10

    Weak-to-strong performance is governed by the overlap between the weak model's unlearnable error space and the strong model's principal-representation space, quantified by ||P_s(I-P_w)||.

  2. Gluon: Making Muon & Scion Great Again! (Bridging Theory and Practice of LMO-based Optimizers for LLMs)

    cs.LG 2025-05 conditional novelty 6.0 of 10

    The paper derives convergence rates and adaptive per-layer step sizes for Muon and Scion under a layer-wise (L0,L1)-smoothness assumption, and reports that these step sizes approximately match tuned values on NanoGPT.

  3. GradAlign for Training-free Model Performance Inference

    cs.LG 2024-11 conditional novelty 6.0 of 10

    GradAlign scores architectures by per-sample gradient agreement at initialization and beats most existing training-free NAS metrics on NAS-Bench-101, NAS-Bench-201, and NDS.

  4. A Method for Enhancing Generalization of Adam by Multiple Integrations

    cs.LG 2024-12 reject novelty 4.0 of 10

    MIAdam adds an n-th order integral term to Adam's gradient update for the first ζ steps, then switches to Adam, and is reported to improve test accuracy and label-noise robustness.

  5. Learning Causality for Modern Machine Learning

    cs.LG 2025-06 conditional novelty 2.0 of 10

    A thesis compiling six papers that use causal invariance to improve graph neural networks' out-of-distribution generalization, interpretability, and robustness.

Pith tools