报告人： Prof. Jian-Guo Liu (Duke University)
时间： 2018年9月21日（星期五）10:30—11:30
地点： 苏州大学本部精正楼二楼报告厅
报告摘要：
Inthis talk I will present some mathematical questions in machine learning that Ihave worked with various people including some undergraduate students. Thepropose of this talk is to show various mathematical tools can be used toeffectively study and understand machine learning. I will not go throughtechnique details, rather I give an overview of these questions and howmathematics is used.
(1)Rigorous justification of the small jump approximation of the stochasticgradient descent (SGD) and online principal component analysis (PCA).
(2)Shape estimates on the escape time for SGD to escape from unstable stationarypoints including both saddle points and local maximums.
(3)Online learning in optical tomography by stochastic gradient descent (SGD).
(4)A modified Levy jump-diffusion model based on market sentiment memory foronline jump prediction.
(5)Data clustering based on Langevin annealing with a self-consistent potential.
(6)Learning interacting particle systems: diffusion parameter estimation foraggregation equations.
报告人简介：
Jian-GuoLiu earned the BS and MS at Fudan University in 1982 and 1985 respectively, andthe PhD at University of California, Los Angeles, in 1990. He is currently aProfessor at the departments of Physics and Mathematics at Duke University. Hewas a Courant Instructor at NYU, an assistant Professor at Temple University, anAssociate Professor and then Professor in the Department of Mathematics andInstitute for Physical Science and Technology at University of Maryland. Hisresearch interests include nonlinear partial differential equations, kinetictheory, collective dynamics, numerical methods for incompressible viscous flow.He is a fellow of AMS. He published about 200 journal papers and gave more than300 invited talks, colloquia and seminars.
欢迎参加！