Enhanced Face Recognition using Message-Passing Algorithm | On How Searching is the New Packing

Title: Enhanced Face Recognition using Message-Passing Algorithm

Abstract: The goal of this project is to develop theories and algorithms to facilitate large-scale, unconstrained identity discovery, using images containing faces where the individuals belong to various social networks. By exploiting the contextual information from multiple photos and other resources, a delicate social and affiliation network structure can be exposed, allowing us to leverage such information and enhance the accuracy of face recognition.

This talk will introduce some background on affiliation networks, related problems in face recognition, and how affiliation networks relate to face recognition. Then we present an explicit real-world example, followed by message-passing algorithms we developed based on an analogy with the theory of Tanner graphs in coding theory and iterative decoding through belief propagation. We conclude with some challenges we face as regards such message-passing algorithms.

**** Yana
Title: On How Searching is the New Packing

Abstract: In this talk I will set up
the problem of communicating over Multiple Access Channels with feedback. I will begin by introducing the channel coding problem and the classical packing-type solutions to this problem using Binary Symmetric and Additive White Gaussian Noise Channels as examples. I will then contrast how coding over these channels in the presence of feedback can also be interpreted as a search problem using recently discovered Posterior Matching principle. Finally, I will introduce the challenges and potential directions for characterizing capacity region of an arbitrary Multiple Access Channel with feedback using ideas of Posterior Matching.