Second-order step-size tuning of SGD for non-convex optimization - Archive ouverte HAL Access content directly
Journal Articles Neural Processing Letters Year : 2022

Second-order step-size tuning of SGD for non-convex optimization


In view of a direct and simple improvement of vanilla SGD, this paper presents a fine-tuning of its step-sizes in the mini-batch case. For doing so, one estimates curvature, based on a local quadratic model and using only noisy gradient approximations. One obtains a new stochastic first-order method (Step-Tuned SGD), enhanced by second-order information, which can be seen as a stochastic version of the classical Barzilai-Borwein method. Our theoretical results ensure almost sure convergence to the critical set and we provide convergence rates. Experiments on deep residual network training illustrate the favorable properties of our approach. For such networks we observe, during training, both a sudden drop of the loss and an improvement of test accuracy at medium stages, yielding better results than SGD, RMSprop, or ADAM.
Fichier principal
Vignette du fichier
2103.03570v2.pdf (1.47 Mo) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03161775 , version 1 (08-03-2021)
hal-03161775 , version 2 (23-11-2021)



Camille Castera, Cédric Févotte, Jérôme Bolte, Edouard Pauwels. Second-order step-size tuning of SGD for non-convex optimization. Neural Processing Letters, 2022, pp.1--26. ⟨10.1007/s11063-021-10705-5⟩. ⟨hal-03161775v2⟩
152 View
174 Download



Gmail Facebook Twitter LinkedIn More