Pulkit Grover - Communication and Control through an Information-scientific Lens
Tuesday, January 24, 2012 2:30 PM - 3:30 PM EST
by Pulkit Grover, Postdoctoral Researcher, Stanford University
Besides being communicated, information is also acquired, processed, protected, managed, and, most importantly, used. In this talk, I will discuss some example systems where information is dealt with in more than one such way.
I will first focus on a problem of communicating and processing information: minimizing power consumed in transmission and processing (encoding/decoding). I will first present new fundamental limits on *total* (transmit + processing) power consumption. These limits show that traditional bounded-transmit-power techniques can be highly inefficient from a total power perspective as the target error probability is lowered. Using information-theoretic code-design techniques and experimental decoder designs, I will present new joint designs of code/decoder pairs that are significantly more power-efficient.
Next, I will talk about problems of communicating and using information. I will discuss how the traditional control-theoretic doctrine of certainty-equivalence and the traditional information-theoretic doctrine of Shannon-separation fail to address such toy problems. I will then focus on the simplest such toy problem -- the Witsenhausen counterexample -- which has remained unsolved for the last 40 years. By combining tools from information theory, control, and large deviations, I will present the first provably approximately-optimal solutions to the Witsenhausen counterexample. Time remaining, I will discuss applications of this result to signaling in economics, systems biology, cyber-physical systems, and interference cancelation.
Joint work with (theory) Andrea Goldsmith, Anant Sahai; (experimental) Karthik Ganesan, Yang Wen, Jan Rabaey.
Dr. Grover says, "There are two ... papers that people can read the introductions of (and if interested, the results too)"
"The first paper presents the entirety of the picture of minimizing total power. The strongest results appear in a later paper, but this paper presents the vision very well."
"The second paper talks about control + communication (usage/transmission of information). This is a magazine paper, so it is not mathematical and is a significantly easier read."
"I want to convey (and get feedback on) the info-scientific nature of the problems, not the techniques, so my talk will stay at a high level."