Dr. Maxim Raginsky, Assistant Research Professor
Consider a scenario where a large number of sensors are deployed at random over a given geographical region, where they will sense some phenomenon of interest, such as the concentration of a poisonous chemical or pressure of a gas. They will then transmit their measurements via a wireless channel to a fusion center, whose task is to fit a function of location to the sensor observations in the least-squares sense in order to acquire a good model of the phenomenon being sensed. Such a network will typically be on a tight power budget, so the sensors will have to compress their observations before passing them on to the fusion center. I will present a novel scheme which is (a) low-complexity, requiring the sensors to carry out simple operations such as uniform scalar quantization with additive dither, as well as simple message passing, and (b) universal, in the sense that it is applicable to measurements, provided some mild regularity conditions are met. The scheme comes with solid theoretical guarantees on the average number of bits sent out by the network and on the mean-squared error incurred by the fusion center in estimating the regression function. I will also present results of numerical simulations, which validate the theory.
Maxim Raginsky, received the B.S. and M.S. degrees in 2000 and the Ph.D. in 2002 from Northwestern University, Evanston, Illinois in Electrical Engineering. From 2002 to 2004 he was a Postdoctoral Researcher at the Center for Photonic Communication and Computing at Northwestern University where he pursued work on quantum cryptography and quantum communication and information theory. From 2004 to 2007 he was a Beckman Foundation Postdoctoral Fellow at the University of Illinois in Urbana-Champaign, where he carried out research on information theory, statistical learning and computational neuroscience. Starting Fall 2007, he will be joining the Department of Electrical and Computer Engineering at Duke University as a Research Scientist. His interests include statistical signal processing, information theory, statistical learning and nonparametric estimation, and wireless sensor networks. In particular, he is interested in design and analysis of low-complexity, scalable universal algorithms for nonparametric learning and inference in communication networks under bandwidth constraints, as well as in the theory and practice of robust statistical inference with limited information.