Deep Approximation via Deep Learning

  • A+

:沈佐伟(新加坡国立大学)
:2024-12-12 16:30
:海韵园实验楼S103

报告人:沈佐伟新加坡国立大学

 间:2024121216:30

 点:海韵园实验楼S103

内容摘要:

The primary task of many applications is approximating/estimating a function  through samples drawn from a probability distribution on the input space. The deep approximation is to approximate a function by compositions of many layers of simple functions, that can be viewed as a series of nested feature extractors. The key idea of deep learning network is to convert layers of compositions to layers of tuneable parameters that can be adjusted through a learning process, so that it achieves a good approximation with respect to the input data. In this talk, we shall discuss mathematical theory behind this new approach and approximation rate of deep network; we will also show how this new approach  differs from the classic approximation theory, and how this new theory can be used to understand and design deep learning networks.

人简介

沈佐伟是新加坡国立大学杰出讲席教授,数学科学研究所所长。他的研究方向为数据科学的数学基础,包括逼近与小波理论、图像处理与压缩感知、计算机视觉和机器学习等领域。沈教授是发展中国家科学院院士、新加坡科学院院士,同时还是美国工业与应用数学协会和美国数学学会的会士。他曾应邀在2010年国际数学家大会(ICM)2015年国际工业与应用数学大会(ICIAM)做特邀报告。

 

联系人:谭绍滨