Machine learning using neural networks is a very powerful methodology which has demonstrated utility in many different situations. In this talk I will show how work in the mathematical discipline called topological data analysis can be used to (1) lessen the amount of data needed in order to be able to learn and (2) make the computations more transparent. We will work primarily with image and video data.

Profile

Gunnar Carlsson received his Ph.D. from Stanford University in 1973, and has taught at University of Chicago, University of California (San Diego), Princeton University, and since 1991 at Stanford University.  His early work was in algebraic topology and homotopy theory, and includes proofs of Segal's Burnside ring conjecture, Sullivan's fixed point conjecture, and many cases of the Novikov's higher signature conjecture.  Since the late 1990's, he has also worked on topological approaches to data analysis, machine learning, and deep learning.  He was a founder of the data analytics company Ayasdi Inc., and is a founder of the Deep Learning startup BlueLightAI INc.

Advisory board: Guowei WEI, Stephan YAU
Organizers: Haibao DUAN (AMSS), Yong LIN (YMSC), Jianzhong PAN (AMSS), Jie WU (BISMA),
Associate organizers: Fei HAN (NUS), Kelin XIA (NTU), Chao ZHOU (NUS)