Advanced Networks Colloquium: Rajesh Sundaresan, "Recursive Distributional Equations"
Wednesday, May 8, 2013
2168 A.V. Williams Bldg.
The Advanced Networks Colloquium
Recursive distributional equations, endogeny, and belief propagation via three examples
Coordinated Science Lab
University of Illinois at Urbana-Champaign
In this talk, I will introduce the notion of a recursive distributional equation (RDE) via three combinatorial optimization problems -- matching, edge-cover, and traveling salesman -- in the random mean field setting. I will then highlight a property called endogeny, introduced by Aldous and Bandyopadhyay, for capturing the notion of correlation decay. On the three examples, I will then indicate how the RDE and endogeny could be useful in coming up with and establishing validity of belief propagation equations.
The approach works for the matching and edge-cover problems, but endogeny and validity of belief propagation remain open for the traveling salesman problem. The talk will be based on joint work with Mustafa Khandwawala on the edge-cover problem.
Rajesh Sundaresan is currently a visiting scholar at the Coordinated Science Laboratory, University of Illinois at Urbana-Champaign. He is an associate professor at the ECE department of the Indian Institute of Science, Bangalore. He is visiting the Coordinated Science Laboratory on an Indo-US Science and Technology Forum Fellowship. He received his Ph.D. in Electrical Engineering from Princeton University in 1999, designed wireless modems at Qualcomm Incorporated from 1999 to 2005, and joined the faculty of the Indian Institute of Science in 2005. His research interests are in the areas of information theory and networks.