Optimal Learning-Rate Schedules under Functional Scaling Laws: Power Decay and Warmup-Stable-Decay
This work provides a theoretical foundation for optimizing learning-rate schedules in machine learning, particularly for large-scale training like LLMs, though it is incremental by building on prior scaling law frameworks.
The paper tackles the problem of determining optimal learning-rate schedules for training machine learning models under a functional scaling law framework, revealing a phase transition between power decay and warmup-stable-decay regimes based on task difficulty, and applies this to achieve minimax-optimal rates in kernel regression.
We study optimal learning-rate schedules (LRSs) under the functional scaling law (FSL) framework introduced in Li et al. (2025), which accurately models the loss dynamics of both linear regression and large language model (LLM) pre-training. Within FSL, loss dynamics are governed by two exponents: a source exponent $s>0$ controlling the rate of signal learning, and a capacity exponent $β>1$ determining the rate of noise forgetting. Focusing on a fixed training horizon $N$, we derive the optimal LRSs and reveal a sharp phase transition. In the easy-task regime $s \ge 1 - 1/β$, the optimal schedule follows a power decay to zero, $η^*(z) = η_{\mathrm{peak}}(1 - z/N)^{2β- 1}$, where the peak learning rate scales as $η_{\mathrm{peak}} \eqsim N^{-ν}$ for an explicit exponent $ν= ν(s,β)$. In contrast, in the hard-task regime $s < 1 - 1/β$, the optimal LRS exhibits a warmup-stable-decay (WSD) (Hu et al. (2024)) structure: it maintains the largest admissible learning rate for most of training and decays only near the end, with the decay phase occupying a vanishing fraction of the horizon. We further analyze optimal shape-fixed schedules, where only the peak learning rate is tuned -- a strategy widely adopted in practiceand characterize their strengths and intrinsic limitations. This yields a principled evaluation of commonly used schedules such as cosine and linear decay. Finally, we apply the power-decay LRS to one-pass stochastic gradient descent (SGD) for kernel regression and show the last iterate attains the exact minimax-optimal rate, eliminating the logarithmic suboptimality present in prior analyses. Numerical experiments corroborate our theoretical predictions.