Title: Deep Representative Learning

Speaker: Jiao Yuling

Abstract: It is believe that success of deep learning depends on its automatic data representation abilities. But few theoretical works to explore this. In this talk, we present a statistical framework to achieve a good data representation that enjoys information preservation, low dimensionality and disentanglement. At the population level, we formulate the ideal representation learning task as finding a nonlinear repesentaion map that minimizes the sum of losses characterizing conditional independence and disentanglement. We estimate the target map at the sample level nonparametrically with deep neural networks. We derive a bound on the excess risk of the deep nonparametric estimator. The proposed method is validated via comprehensive numerical experiments and real data analysis in the context of regression and classification.

Introduction:

Jiao Yuling, Associate Professor of Wuhan University, graduated from School of Mathematics and Statistics, Wuhan University in 2014. He mainly studies statistical learning, inverse problems and other aspects. He has presided over the general program and youth program of National Natural Science Foundation of China, the general program of Natural Science Foundation of Hubei Province, and the program of Key Laboratory of Advanced Theory and Application in Statistics and Data Science, Ministry of Education. Besides, he has published over 40 papers in journals and conferences including SIAM Journal on Numerical Analysis, SIAM Journal on Scientific Computing, Applied and Computational Harmonic Analysis, Statistical Science, Journal of Machine Learning Research, ICML, Inverse Problems, IEEE Transactions on Signal Processing, Statistica Sinica and Science China.

Inviter: Yan Xiaodong

Conference Time: 15:00-16:00 on September 29 (Tuesday)

Tencent Conference, 749 485 513