INRIA Lille - Nord Europe, SequeL team (Sequential Learning)
Research interests:
- Compressed Learning, random projections
- Bandits, Experts, and online learing
- Optimistic planning, tree search
- bandits in metric spaces
- bandits with infinitely many arms
- Reinforcement Learning (RL) and approximate dynamic programming (DP):
- Analysis of RL and DP with Lp norms
- Sample complexity bounds
- RL and DP with function approximation
- Reinforcement Learning in continuous time
- Numerical approximations of Hamilton-Jacobi-Bellman equations
- Link to the theory of viscosity solutions
- Variable resolution discretizations
- Numerical approximations of Hamilton-Jacobi-Bellman equations
- Policy gradient
- Sensitivity analysis in continuous time
- Sensitivity analysis in POMDPs via particle filters
- Variance reduction techniques for value function and policy gradient estimation
Publications
Projects and activities:
- PASCAL2 site INRIA Lille, since October 2009.
- European project COMPLACS (Composing Learning for Artificial Cognitive Systems) 2011-2015
- ANR EXPLO-RA (EXPLOration - EXPLOitation for efficient Resource Allocation. Applications to optimization, control, learning, and games) 2009-2011
- ANR CO-ADAPT (Brain computer co-adaptation for better interfaces), 2010 - 2013.
- PASCAL 2 Pump Priming Programme Sparse Reinforcement Learning in High Dimensions, 2010 - 2011
- Associated Team with RLAI University of Alberta, 2009 - 2010, 2011
- ARC CODA: Contrôle Optimal d'un Digesteur Anaérobie, 2007 - 2008
- Associated researcher with CREA (Centre de Recherche en Epistémologie Appliquée), Ecole Polytechnique, from 2007.
Organization of scientific events
- INRIA Workshop on Statistical Learning. December 5, 6, 2011
- Machine Learning Summer School 2011 in Bordeaux. Slides of my Introduction to Reinforcement Learning: Part1, Part2, Part3
- ICML 2011 tutorial on bandits: Introduction to Bandits: Algorithms and Theory (with Jean-Yves Audibert). Slides: Part1, Part2
- ICML 2009 workshop On-line Learning with Limited Feedback (Sponsored by PASCAL 2). See Videolectures
- European Workshop on Reinforcement Learning, 2008. A post selection of 21 papers have been published by Springer in this LNCS Volume.
- Co-chair of ADPRL 2007 (IEEE Symposium on Approximate Dynamic Programming and Reinforcement Learning), celebrating the 50th anniversary of Richard Bellman's pioneering work on Dynamic Programming in 1957. April 1-5, 2007, Hawaii, USA.
- ICML/COLT 2006 Workshop Kernel Machines and Reinforcement Learning, June 29, 2006, Pittsburgh, USA.
Teaching (Master Maths Vision Apprentissage ENS Cachan)
PhD Students:
- Sébastien Bubeck, now assistant professor at Princeton. Prix Gilles Khan 2010, Prix Jacques Neveu 2010, Prix IA 2011
- Odalric-Ambrym Maillard (co-supervized with Philippe Berthet), now postdoc with Peter Auer
- Jean-François Hren
- Alexandra Carpentier
- Pierre-Arnaud Coquelin, now CEO of Vekia
- Robin Jaulmes
- Emilie Kauffmann (co-supervized with Aurélien Garivier and Olivier Cappé)
- Amir Sani (co-supervized with Alessandro Lazaric)
Contact:
Address:Rémi Munos, SEQUEL project, INRIA Lille - Nord Europe,
40 avenue Halley, 59650 Villeneuve d'Ascq, FRANCE
Email: remi (dot) munos (at) inria (dot) fr
Tel: (0 or 33)3 59 57 79 06
Fax: (0 or 33)3 59 57 78 50