In many modern communication systems, optimal dynamic control is the primary tool for optimizing a long run objective function. To name some important cases, consider optimal PHY configuration selection prior to each wireless transmission, e.g. as in wireless relay system; best coding decision in Network Coding (NC) system; optimal scheduling and pricing in cloud computing and others. In this talk, we will address several problems in modern communication, in the aspect of optimal dynamic control. In particular, we will define a unified Markov decision process (MDP) or Semi-Markov decision process (SMDP) based framework for PHY configuration selection problem, NC problem and problems related to cloud computing. For all problems, we will show the general conditions for existence of special structure of the optimal policies, and will overview the proof methodology. For the NC case, we will discuss the treatment of particularly large state-space by means of Abstract MDP.
Joint work with Asaf Cohen, Omer Gurewitz, Dennis Goeckel, Daniel Menasche, Israel Cidon, Rami Atar.
Mark Shifrin has been serving as a postdoctoral researcher in Ben Gurion University, Israel where he was a recipient of Kreitman scholarship. He participated in Horizon 2020 European program during his latest research projects. Mark received his B.Sc., M.Sc. and Ph.D. degrees from the Department of Electrical Engineering, Technion. His research interests include application of stochastic modeling and control to various scenarios in modern communication by means of MDP, SMDP, treatment of large state space by Machine Learning esp. Reinforcement learning, Abstract MDP, theoretical aspects of MDP, applications, including analysis of optimal policies. Among this he treated a variety of applications in Cloud Computing, Information Theory, Physical layer configuration.