It is well known that relationships between data points (i.e., context) in structured data can be exploited to obtain better recognition performance. In our recent work, we have explored a different, but related, problem: how can these inter relationships be used to efficiently learn and continuously update a recognition model, with minimal human labeling effort. Towards this goal, we have proposed an active learning framework to select an optimal subset of data points for manual labeling by exploiting the relationships between them, which will be the focus of the first part of the talk. We will show how information theoretic measures, using ideas of entropy, mutual information and typicality, can be used to identify the optimal subsets. In the second part, we will demonstrate how context can be exploited in a camera network for target reidentification, even as the camera network topology can change over time. This is a critical step towards addressing open world dynamic camera network scenarios, which is only starting to receive interest in the research community.

Prof. Amit Roy-Chowdhury