Introduction: Theoretical Understandings of Deep Neural Networks
报告人：张绍群 （南京大学）
时间：2022年 3月31日，4月1日， 20：00-22：00
地点：3月31日：腾讯会议：398 468 181
4月01日：腾讯会议：485 495 873
摘要：Recent years have witnessed a tremendous hot wave of deep neural networks. While the deep neural network is successful in various applications, it is not yet well understood theoretically. This presentation first introduces some theoretical understandings of deep neural networks, especially from the perspectives of approximation, optimization, and generalization. Based on the introduction, I review our recent results on this topic, including neural kernel representation, approximation capacity, and optimization dynamics. Finally, I provide a discussion on the dynamics and algorithmic powers of the spiking neural network, which is known as the 
-generation neural network due to its temporal modeling behavior.
报告人简介：张绍群是南京大学计算机科学与技术系的在读博士生。其本科和硕士毕业于四川大学数学学院。他的研究兴趣包括机器学习和数据挖掘。他曾在人工智能顶级期刊 (JMLR、NCJ等) 上发表论文，并多次担任期刊 (Nature、AIJ、TPAMI、MLJ等) 以及会议 (ICLR、ICML、IJCAI、NeurIPS等) 审稿人。
