| Lecturer Profile |
Chung Chan received his B.Sc., M.Eng. and Ph.D. from the Electrical Engineering and Computer Science Department at the Massachusetts Institute of Technology (MIT) in 2004, 2005 and 2010 respectively. He is currently a Research Assistant Professor at the Institute of Network Coding, the Chinese University of Hong Kong. His research area is information theory, with applications to network coding, multi-terminal source coding and security problems that involve high-dimensional statistics. He is currently working on an efficiently computable multi-terminal information theory, with potential applications to machine learning such as data clustering and feature selection. |
| Lecture Abstract |
We propose a new information-theoretic paradigm, called info-clustering, to address the data clustering problems in machine learning. Other approaches have been considered in the literature, such as the mutual information relevance networks (MIRN) in gene clustering, and the integrated information theory (IIT) in the study of brain connectome in neuroscience. However, they lack theoretical justification and efficient non-trivial implementation.
Our proposed paradigm is based on a measure of multivariate mutual information (MMI). The measure was established from our study of the multi-terminal secret-key agreement and network coding problems. While traditional information-theoretic expressions are often difficult to compute, the MMI can be computed in strongly polynomial-time. It is also possible to develop more efficient and tunable algorithms under various simplifying assumptions. In fact, the clustering by MIRN is a special case of info-clustering, and the IIT can also be modified and justified properly under the proposed paradigm. The theoretical underpinning of the MMI is a highly non-trivial mathematical structure that provides a new dimension to multi-terminal information theory. |
