Decentralized Control of Markov Teams and Games with Partial Information


Eitan Altman


The National Institute for Research in
Computer Science and Automation (INRIA)


Thursday, 20 December 2018, 11:30 to 12:30


  • A-201 (STCS Seminar Room)


Abstract:  We consider in this talk distributed control by agents that have a common clock. The information available to agents at time t may vary from one agent to another. We show how to transform such problems into equivalent Markov Decision Processes with full information. 

We study in particular the 1 step delay sharing information structure for problems of constrained Markov teams as well as for stochastic games. Both cases present extra complexity due to the fact that optimal policies require randomization. We present applications to power control problems.