报告题目：Geometrization of Deep Representation Learning

（深度表示学习的几何解释）

报告时间：2024年5月14日（周二）14：00

报告地点：C1-437

主讲人：朱志辉博士

报告摘要：

In the past decade, the revival of deep neural networks has led to dramatic success in numerous applications, ranging from computer vision to natural language processing, scientific discovery, and beyond. Nevertheless, the practice of deep networks has been shrouded in mystery as our theoretical understanding of the success of deep learning remains elusive. In this talk, we will focus on the geometric properties of the representations learned by well-trained deep learning models. We will first provide a geometric analysis to shed light on the neural collapse phenomenon. Then, we will leverage these findings to improve network design and training, as well as to understand the transferability of deep neural networks.

主讲人简介：

朱志辉博士，俄亥俄州立大学计算机科学与工程系助理教授。2012年，毕业于浙江工业大学（健行学院）获得通信工程学士学位，2017年在科罗拉多矿业学院获得电气工程博士学位。2018年至2019年，约翰斯·霍普金斯大学数据科学数学研究所博士后研究员。IEEE信号处理协会, 机器学习信号处理委员会(MLSP TC) 当选成员，Transactions on Machine Learning Research执行编辑，NeurIPS和ICML的区域主席。