 |
Center for Nonlinear Studies |
DATE CORRECTION: In this talk, I will first present on a numerical global optimization framework for discrete-time decentralized stochastic control problems. It is well-understood that such decentralized control problems, including the celebrated 1968 Witsenhausen’s counterexample, do not admit linear optimal solutions. Moreover, these problems are in general highly non-convex, with a cost surface riddled with local minima, hence any simple optimization strategy based on gradient descent converges to poor local minima. We propose a generally applicable non-convex numerical optimization method based on the concept of deterministic annealing – which is derived from information theoretic principles and was successfully employed in several problems including vector quantization, classification and regression. We will present comparative numerical results for two test problems that show strict superiority of the proposed method over prior approaches in the optimization literature. In the second part of the talk (if time permits), I will present my recent work at UIUC, on strategic communication.Strategic communication problems are radically different from the conventional communication paradigms in information theory since they involve different objectives for the encoder and the decoder, which are aware of this mismatch and act accordingly. This leads, in our setting, to a hierarchical communication game, where the transmitter announces an encoding strategy with full commitment, and its distortion measure depends on a private information sequence whose realization is available at the transmitter. The receiver decides on its decoding strategy that minimizes its own distortion based on the announced encoding map and the statistics. We will analyze the equilibria of such games for several settings and will discuss the implications of the results in the broader context of the game-theoretic analysis of smart grid networks.
Host: Michael Chertkov