Pith. sign in

REVIEW 9 cited by

When and Why Momentum Accelerates SGD:An Empirical Study

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 2306.09000 v1 pith:L7IYTOVG submitted 2023-06-15 cs.LG cs.CV

When and Why Momentum Accelerates SGD:An Empirical Study

classification cs.LG cs.CV
keywords momentumsgdmlearningwhenbatchsameabruptsharpening
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

Momentum has become a crucial component in deep learning optimizers, necessitating a comprehensive understanding of when and why it accelerates stochastic gradient descent (SGD). To address the question of ''when'', we establish a meaningful comparison framework that examines the performance of SGD with Momentum (SGDM) under the \emph{effective learning rates} $\eta_{ef}$, a notion unifying the influence of momentum coefficient $\mu$ and batch size $b$ over learning rate $\eta$. In the comparison of SGDM and SGD with the same effective learning rate and the same batch size, we observe a consistent pattern: when $\eta_{ef}$ is small, SGDM and SGD experience almost the same empirical training losses; when $\eta_{ef}$ surpasses a certain threshold, SGDM begins to perform better than SGD. Furthermore, we observe that the advantage of SGDM over SGD becomes more pronounced with a larger batch size. For the question of ``why'', we find that the momentum acceleration is closely related to \emph{abrupt sharpening} which is to describe a sudden jump of the directional Hessian along the update direction. Specifically, the misalignment between SGD and SGDM happens at the same moment that SGD experiences abrupt sharpening and converges slower. Momentum improves the performance of SGDM by preventing or deferring the occurrence of abrupt sharpening. Together, this study unveils the interplay between momentum, learning rates, and batch sizes, thus improving our understanding of momentum acceleration.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 9 Pith papers

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

  1. Momentum Further Constrains Sharpness at the Edge of Stochastic Stability

    cs.LG 2026-04 unverdicted novelty 7.0

    Momentum SGD exhibits two distinct EoSS regimes for batch sharpness, stabilizing at 2(1-β)/η for small batches and 2(1+β)/η for large batches, aligning with linear stability thresholds.

  2. On the Provable Suboptimality of Momentum SGD in Nonstationary Stochastic Optimization

    stat.ML 2026-01 conditional novelty 7.0

    Momentum SGD incurs a provable drift-amplification penalty in nonstationary stochastic optimization that makes it worse than vanilla SGD in drift-dominated regimes, confirmed by finite-time upper bounds and minimax lo...

  3. Overcoming the Weakest-Link Effect in LLM-Driven Program Optimization via Heterogeneous Edit Recombination

    cs.LG 2026-07 conditional novelty 6.0

    HERO optimizes programs by generating atomic edits without score feedback and selecting the highest-scoring subset of those edits, avoiding the 'weakest-link' failure of accepting or rejecting whole edit bundles.

  4. Compute Efficiency and Serial Runtime Tradeoffs for Stochastic Momentum Methods

    cs.LG 2026-06 unverdicted novelty 6.0

    Lower bounds establish that heavy-ball momentum extends the compute-efficient batch-size window by sqrt(kappa) over SGD in linear regression, with accelerated SGD showing spectrum-dependent CE-serial runtime tradeoffs.

  5. Perfect Parallelization in Mini-Batch SGD with Classical Momentum Acceleration

    cs.LG 2026-05 unverdicted novelty 6.0

    Classical momentum acceleration in mini-batch SGD for quadratics is proportional to batch size up to saturation, enabling perfect parallelization under minimal noise assumptions.

  6. Instruction-as-State: Environment-Guided and State-Conditioned Semantic Understanding for Embodied Navigation

    cs.CV 2026-04 unverdicted novelty 6.0

    Instruction understanding is reframed as an evolving Instruction-as-State variable conditioned on perceptual state and realized via the S-EGIU coarse-to-fine framework, reporting a +2.68% SPL gain on REVERIE Test Unseen.

  7. On the Provable Suboptimality of Momentum SGD in Nonstationary Stochastic Optimization

    stat.ML 2026-01 conditional novelty 6.0

    Momentum SGD pays a provable tracking penalty under distribution shift — lag growing as (1−β)⁻¹ and tracking floors worse than vanilla SGD in drift-dominated regimes.

  8. MGUP: A Momentum-Gradient Alignment Update Policy for Stochastic Optimization

    cs.LG 2026-06 unverdicted novelty 5.0

    MGUP augments momentum optimizers with selective larger steps on a fixed proportion of parameters per iteration, claiming convergence guarantees for MGUP-AdamW and superior empirical performance on pretraining and fin...

  9. The Essence of Balance for Self-Improving Agents in Vision-and-Language Navigation

    cs.CV 2026-04 unverdicted novelty 5.0

    SDB balances behavioral diversity and learning stability in VLN self-improvement by expanding decisions into latent hypotheses, performing reliability-aware aggregation, and applying a regularizer, yielding gains such...