Learning to detect an oddball target

Rajesh Sundaresan -
Data Science

Date: -
Location: Eurecom

Title: Learning to detect an oddball target Abstract: This talk is motivated by a visual neuroscience problem where individual subjects are asked to identify an oddball target image among many distractors, but are not told the nature of the oddball and distractors images. The key feature here is that the oddball and distractor images are unknown a priori and have to be learnt along the way. We propose an information theoretic lower bound on the expected search time, and provide an asymptotically optimal strategy, where the asymptotics is as the probability of erroneous detection vanishes. Such problems are of interest to neuroscientists that work on quantifying the "perceptual distance" between objects in order to understand how objects are represented in our brains. I will briefly describe the connection at the beginning of the talk. This is joint work with Nidhin Koshy Vaidhiyan. Biography: Rajesh Sundaresan is an associate professor at the ECE department of the Indian Institute of Science, Bangalore. He received his B.Tech in 1994 (IIT Madras), M.A. in 1996 (Princeton University), Ph.D. in 1999 (Princeton University). He worked on modem design and development from 1999-2005 (at Qualcomm Inc.). Since 2005, he has been at the Indian Institute Science. His interests are in communication, computation, and control over networks.