"The Exploration-Exploitation Dilemma on Structured Environments" , IRIT conference

on the December 16, 2014

12:30-1:30 pm
Manufacture de Tabacs
ME303 Room

Filipo Studzinski Perotto will talk about "The Exploration-Exploitation Dilemma on Structured Environments"

Abstract : Balancing exploratory and exploitative behavior is an essential dilemma faced by adaptive agents. The problem of finding a good trade-off between exploration (learn new things) and exploitation (act optimally based on the current knowledge) has been largely studied for Markov Decision Processes (MDP), but it is relatively new for Factored MDPs. In this presentation we present a strategy to solve the exploration-exploitation dilemma coupled with a learning mechanism designed to learn the FMDP parameters. The solution consists in explicitly creating two different policies, each one designed for exploring or exploiting.
Updated on the December 10, 2014