{"id":"arxiv/2610.16518","url":"https://arxiv.org/abs/2610.16518","tags":["cs.LG","math.OC","stat.ML"],"score":0,"title":"Asymptotic Risk of Learning-Rate Schedules for Least-Squares Regression with Discretely Distributed Features","domain":"cs.LG","source":"arxiv","authors":["Noé Colomban","Francis Bach","Fabian Schaipp"],"reviews":0,"subject":"cs.LG","summary":"We study the last iterate of stochastic gradient descent (SGD) on streaming least squares with discretely distributed features (Slock's model), which decouples the dynamics across eigendirections. Under power-law source and capacity conditions, we derive the exact finite-time risk and asymptotic equivalents of constant and Warmup-Stable-Decay (WSD) schedules. With a tuned base learning-rate, the resulting scaling law decays as $Θ(T^{-q})$, where $T$ is the horizon, and $q$ is independent of the cooldown length $c \\in (0,1]$ but is strictly smaller for the constant schedule: any cooldown of…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-10-12","reactions":{}}