Pith. sign in

REVIEW 9 cited by

Optimal Linear Decay Learning Rate Schedules and Further Refinements

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 2310.07831 v2 pith:JZS2MP3C submitted 2023-10-11 cs.LG cs.AIstat.ML

Optimal Linear Decay Learning Rate Schedules and Further Refinements

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

Learning rate schedules used in practice bear little resemblance to those recommended by theory. We close much of this theory/practice gap, and as a consequence are able to derive new problem-adaptive learning rate schedules. Our main technical contribution is a refined analysis of learning rate schedules for a wide class of optimization algorithms (including SGD). When considering only worst-case analysis, our theory predicts that the optimal choice is the linear decay schedule where the step-size is set proportional to 1 - t/T, where t is the current iteration and T is the total number of steps. To go beyond this worst-case analysis, we use the observed gradient norms to derive schedules refined for any particular task. These refined schedules exhibit learning rate warm-up and rapid learning rate annealing near the end of training. Ours is the first systematic approach to automatically yield both of these properties. We perform the most comprehensive evaluation of learning rate schedules to date, evaluating across 10 diverse deep learning problems, a series of LLMs, and a suite of logistic regression problems. We validate that overall, the linear-decay schedule outperforms all commonly used default schedules including cosine annealing. Our adaptive schedule refinement method gives further improvements.

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. A Defense of the Quadratic Model

    cs.LG 2026-07 conditional novelty 7.0

    Local Taylor-expanded quadratic models reproduce a 150M-parameter LLM's validation loss for up to 10% of training late in the run, and LLM pretraining operates within a factor of 2 of a stochastic or deterministic edg...

  2. Last Iterate Convergence of AdaGrad-Norm for Convex Non-Smooth Optimization

    math.OC 2026-04 unverdicted novelty 7.0

    AdaGrad-Norm last iterate achieves O(1/N^{1/4}) suboptimality for convex non-smooth problems, with tight lower bounds.

  3. Theory of Optimal Learning Rate Schedules and Scaling Laws for a Random Feature Model

    cond-mat.dis-nn 2026-02 unverdicted novelty 7.0

    In a random feature model, optimal SGD learning-rate schedules are polynomial decay in the easy phase and warmup-stable-decay in the hard phase, outperforming constant or simple power-law schedules and transferring di...

  4. Taking the Road Less Scheduled with Adaptive Polyak Steps

    cs.LG 2025-11 unverdicted novelty 7.0

    Polyak-style step sizes for Schedule-Free SGD and Adam achieve O(1/sqrt(t)) anytime last-iterate rates for convex Lipschitz problems using per-iteration loss and gradient information.

  5. WSqD: A Horizon-Free Learning Rate Schedule for Large Model Training

    cs.LG 2026-07 conditional novelty 6.0

    WSqD uses a horizon-independent shifted inverse-square-root base plus linear cooldown to get optimal last-iterate rates in convex stochastic optimization and match or beat tuned WSD on LLM pretraining with one reused ...

  6. Convergence of Continual Learning in Homogeneous Deep Networks

    cs.LG 2026-06 unverdicted novelty 6.0

    Continual classification in homogeneous models is sequential projections onto margin sets, with local linear convergence under regularity properties for random and cyclic tasks, extended to regression.

  7. Sharp asymptotic theory for Q-learning with LDTZ learning rate and its generalization

    stat.ML 2026-04 unverdicted novelty 6.0

    Q-learning with PD2Z/LD2Z step sizes admits sharp non-asymptotic bounds, a tail Polyak–Ruppert CLT, and a time-uniform Gaussian approximation, establishing a best-of-both-worlds rate-and-bias tradeoff.

  8. Step-Size Stability in Stochastic Optimization: A Theoretical Perspective

    math.OC 2026-02 conditional novelty 6.0

    The stability index δ_t, the variance term in a two-term suboptimality bound, is provably smaller and saturates in the step-size cap for SPS, NGN, and proximal point, explaining their observed learning-rate robustness...

  9. A Closed-Form Upper Bound for Admissible Learning-Rate Steps in Belief-Space Dynamics

    cs.LG 2026-05 unverdicted novelty 5.0

    A closed-form formula gives the maximum admissible learning-rate step size for belief-space updates to ensure contractivity under KL/Bregman geometry.