Location:S16-05-96, Computer Lab 4, Faculty of Science
Time:03:00pm - 04:00pm
THIS SEMINAR IS CANCELLED. WE APOLOGISED FOR ANY INCONVENIENCE.
THANK YOU FOR YOUR KIND UNDERSTANDING.
Big datasets are often accompanied by high-dimensional statistical inferences arising from modern engineering problems such as natural language processing, computer vision, genetics, and neuroscience. High-dimensionality has significantly challenged traditional statistical theory in that classical methods can break down drastically due to either high computational cost or low statistical accuracy. Can one devise new strategies and develop new algorithms to improve both statistical and computational efficiencies?
In this talk, I will present a principled way of designing and analyzing algorithms using approximation-theoretic methods. To overcome the aforementioned challenges, the main ingredient is the choice of complexity parameters to balance the approximation error and the estimation error, which leads to efficient algorithms with provable optimality guarantees. More fundamentally, via duality between best approximation and moment matching, tight lower bounds can be naturally derived. I will discuss successful applications in property estimation, learning mixture models, and training overparametrized neural networks as examples to show the power of the proposed principle in solving challenging problems in a variety of engineering fields.
Pengkun Yang is from the Department of Electrical Engineering at Princeton University. He is a Postdoctoral Research Associate advised by Professor Yuxin Chen. His research interests include statistical inference, learning, optimization and systems. He received a Ph.D. degree (2018) and a master’s degree (2016) from the Department of Electrical and Computer Engineering at University of Illinois at Urbana-Champaign, and a B.E. degree (2013) from the Department of Electronic Engineering at Tsinghua University. He is a recipient of Shun Lien Chuang Memorial Award for Excellence in Graduate Education in 2018, and a recipient of Jack Keil Wolf ISIT Student Paper Award at the 2015 IEEE International Symposium on Information Theory (semi-plenary talk).