ELSC Media

Sanjoy Mitter: Information and Entropy Flow in Estimation and Control

Information, Control, and Learning: The Ingredients of Intelligent Behavior.

September 2016, Jerusalem


There are two great theories in the Systems, Communications and Control field: Information Theory and Stochastic Control. In this talk, I discuss the role of Information in Estimation Theory and Stochastic Control. There are ``dynamical’’ analogues of source coding and rate distortion theory, as well as channel coding in stochastic control theory, where the concepts of directed information and transfer entropy play a central role. I give an information-theoretic view of Kalman filtering and its nonlinear generalizations. I illustrate the role of directed information and transfer entropy in stochastic control by considering the problem of extracting the maximal amount of work from a noisy, electrical circuit acted upon by a Maxwell Demon. Joint work with Nigel Newton (University of Essex, UK), Henrik Sandberg (KTH, Sweden), JeanCharles Delvenne (UCLouvaine, Belgium) and Takashi Tanaka (KTH, Sweden)

See video