|
|
|
No opportunities right now....
The PhD of Matthieu ZIMMER is co-supervised with Yann Boniface.
Matthieu is interested in Reinforcement Learning with continuous state and action spaces. He is looking to solve them with minimal a-priori and being learning data efficient. He developped an original neuronal actor-critic system and he is now working on a developmental approach to guide the exploration of high dimensionnal sensori-motor spaces.
I have co-supervised the PhD of Raghav ARAS with François Charpillet.
The work of Raghav was on Decentralized Partially observable Markov Decision Processes (Dec-POMDPs). He developped an original approach by using mathematical programming in order to tackle this complex class of problems (NEXP complexity).
Web page : http://members.loria.fr/olivier.buffet
Olivier defended is PhD in september 2003 and, after a post-doc in Caberra (Australia), was hired as a research fellow at the Loria by the INRIA. I co-supervised his master and his PhD.
With Olivier, we worked on using Reinforcement Learning to build reactive Multi-Agent Sytems. The main contribution is an incremental approach of learning inspired by the notion of shaping in psychology.
Another nice achievement of Olivier was a distributed reinforcement learning algorithm where an individual agent learns to combine basic behavior in order to achieve the complex behavior needed to solve a cooperative problem.
With Frédérick Garcia (INRA - Toulouse), Abdel-Illah Mouaddib (Université CAEN) and Olivier Sigaud (LIP6 - Paris), we have founded a french working group. Our aim is to promote exhanges and discussions on planification, decision and learning. We organize regular conferences.
Workgroup web page : www.loria.fr/projets/PDMIA.
Last conference : JFPDA'16
This Workshop used to be held every two years. It gathers the european community on reinforcement learning. Some people cross the atlantic ocean to attend.
We organized the 6th edition in Nancy in september 2003. Past events were held in Bruxelles (1994), Milan (1995), Rennes (1997), Lugano (1999) et Utrecht (2001).
Web page : EWRL
The european project Ozone (IST) deals with ambiant intelligence. Ozone aims at delivering a generic toolbox to help building applications that use various interactions modalities according to the user's context (environment, available media, current behavor, preferences). It is a prospective work with many industrial and academic partners. Prrof of concepts were validatedby demonstrators. We work on a intelligent module to select the best interation modality with a given user.
Proteus was a european project (ITEA) focussed on industrial e-maintenance. I was co-advisor of a work package on the use of artificial intelligence tools integrated in the generic framework designed in Proteus. Our applications were mainly fault diagnosis tasks.
 |