Events

PhD Defense: "A Deterministic Annealing Framework for Global Optimization of Delay-Constrained Communication and Control Strategies"

Mustafa Said Mehmetoglu

August 26th (Friday), 11:30am
Harold Frank Hall, Room 1132 (CS Conf. Rm.)


This dissertation is concerned with the problem of global optimization of delay constrained communication and control strategies. Specifically, we are interested in obtaining optimal encoder decoder functions that map between the source space and the channel space to minimize a given cost function. The cost surfaces associated with these problems are highly complex and riddled with local minima, rendering gradient descent based methods ineffective. We propose and develop a powerful non-convex optimization method based on the concept of deterministic annealing (DA) — which is derived from information theoretic principles with analogies to statistical physics, and was successfully employed in several problems including vector quantization, classification and regression. DA has several useful properties including reduced sensitivity to initialization and strong potential to avoid poor local minima. We develop DA-based optimization methods for the following fundamental communication problems: the Wyner-Ziv setting where only a decoder has access to side information, the distributed setting where independent encoders transmit over independent channels to a central decoder, and analog multiple descriptions setting which is an extension of the well known source coding problem of multiple descriptions. We present comparative numerical results that show strict superiority of the proposed method over gradient descent based optimization methods as well as prior approaches in literature. We give a detailed analysis of the highly non-trivial structure of obtained mappings.

We also study the related problem of global optimization of controller mappings in decentralized stochastic control problems including Witsenhausen’s celebrated 1968 counter-example. It is well-known that most decentralized control problems do not admit closed-form solutions and require numerical optimization. We develop an optimization method for a class of decentralized stochastic control problems. We present comparative numerical results for two test problems that show strict superiority of the proposed method over prior approaches in literature, and analyze the structure of obtained controller functions.

About Mustafa Said Mehmetoglu:

photo of mustafa said mehmetogluMustafa Said Mehmetoglu obtained the B.S. degree in 2011 from the Department of Electrical and Electronics Engineering, Bilkent University, Turkey, and the M.S. degree in 2013 from the Department of Electrical and Computer Engineering, University of California at Santa Barbara, where he is currently working toward the Ph.D. degree. His research interests span signal processing, communications, optimization theory and information theory. He is currently working on novel low-delay joint source-channel coding approaches in communications. He received several fellowships for his undergraduate studies due to ranking 19th in the university entrance exam nationwide, and received the UCSB Dissertation Fellowship in 2015.

Hosted by: Professor Kenneth Rose