报告题目：Learning dynamics near singularties in feedforward neural networks
报 告 人：郭伟立 副教授
The feedforward neural networks have been widely used in many fields. However, there exist singular behaviors in the learning processes, such as the learning process often becomes very slow, is easy to converge to the local minima, and the learning dynamics sometimes trap in the so-called plateaus. In this report, the theoretical and numerical analysis near singularities in typical feedforward neural networks, such as the multilayer perceptrons and RBF networks, are investigated and the mechanism of these singular behaviors is introduced.
Weili Guo was born in Jining, China, in 1987. He received the Ph.D. degree in School of Automation, Southeast University, China in Mar 2015. From May 2015 to Oct 2018, he was a postdoctoral fellow in School of Instrument Science and Engineering, Southeast University (From Dec 2016 to Dec 2017, he was a postdoctoral fellow in School of Computer Science and Engineering, Nanyang Technological University, Singapore). Since Oct 2018, he is currently Associate Professor in School of Computer Science and Engineering, Nanjing University of Science and Technology, China. His research interests include singular learning dynamics of neural networks, deep learning and machine learning.