Add to Google Calendar
Date:  Fri, March 23, 2018
Time:  11:30am
Location:  Holmes 244
Speaker:  Yuanzhang Xiao

Prior to joining HCAC, Yuanzhang Xiao was a postdoctoral researcher in the Department Electrical Engineering and Computer Science at Northwestern University. He's the recipient of the 2013-2014 Dissertation Year Fellowship from UCLA. He is broadly interested in any decision making problems, especially problems that can be solved by optimization, reinforcement learning, and game theory. Dr. Xiao has worked on a variety of applications, such as wireless communications, power systems, and socio-technological networks, decision making problems (in particular, optimization, reinforcement learning, and game theory); wireless communications, power systems, and socio-technological networks.

Deep Learning is Not (Entirely) Alchemy

Deep learning has achieved tremendous success in a variety of applications, such as computer vision, speech recognition, autonomous vehicle, and health. However, a long-lasting criticism about deep learning, ever since its inception, is the lack of theoretical understanding. One particular issue with deep learning is that the training procedure solves a nonconvex problem. Therefore, we cannot guarantee the optimality of the training and cannot analyze the gap from the optimum easily. In this talk, I will show how to optimally train some neural networks with certain architectures, using theory of convex relaxation. I will also describe other ongoing research activities.