Français Anglais
Accueil Annuaire Plan du site
Home > Research results > Dissertations & habilitations
Research results
Ph.D de

Group : Learning and Optimization

Around the Use of Gradients in Machine Learning

Starts on 01/09/2014
Advisor : OLLIVIER, Yann

Funding :
Affiliation : Université Paris-Sud
Laboratory : LRI - AO

Defended on 14/12/2017, committee :
Directeur de thèse :
- Yann Ollivier, chercheur à Facebook

Rapporteurs :
- Sébastien Bubeck, chercheur à Microsoft
- Emmanuel Trélat, professeur à l'UPMC

Examinateur :
- Éric Moulines, professeur à l'École Polytechnique

Research activities :

Abstract :
The main result is a local convergence theorem for the classical dynamical system online optimisation algorithm called RTRL, in a non linear setting. The RTRL works on line, but must maintain in memory a huge amount of information, which makes it unfit to train even moderately large learning models. The NBT algorithm turns it by replacing these informations by a non biased, low dimension, random approximation. We also prove the convergence, with probability arbitrarily close to one, of this algorithm to the local optimum reached by the RTRL algorithm.