Running summarizing tools on a new article
Check the paper below, then start its synthesis.
Multi-agent Sequential Learning via Gradient Ascent Credit Assignment
Oussama Sabri (University Medical Center Hamburg-Eppendorf), Luc Lehéricy (Centre National de la Recherche Scientifique, Université Côte d'Azur), Alexandre Muzy (Centre National de la Recherche Scientifique, Université Côte d'Azur)
Abstract: We consider a multi-agent learning problem with two layers: first, split up the tasks among the agents, and then have them solve their allocated task.A set of algorithms is proposed based on a policy gradient method with different reward functions and models of policy parameterizations.The algorithms are proved to converge.Their performances are assessed and compared on several synthetic data sets.The proposed algorithms prove to be more efficient than a naive multi-armed bandit algorithm.