Speed learning on the fly - Archive ouverte HAL
Pré-Publication, Document De Travail Année : 2015

Speed learning on the fly

Résumé

The practical performance of online stochastic gradient descent algorithms is highly dependent on the chosen step size, which must be tediously hand-tuned in many applications. The same is true for more advanced variants of stochastic gradients, such as SAGA, SVRG, or AdaGrad. Here we propose to adapt the step size by performing a gradient descent on the step size itself, viewing the whole performance of the learning trajectory as a function of step size. Importantly, this adaptation can be computed online at little cost, without having to iterate backward passes over the full data.

Dates et versions

hal-01228955 , version 1 (15-11-2015)

Identifiants

Citer

Pierre-Yves Massé, Yann Ollivier. Speed learning on the fly. 2015. ⟨hal-01228955⟩
118 Consultations
0 Téléchargements

Altmetric

Partager

More