计算与应用数学拔尖博士生系列论坛——Finite element methods and deep neural networks
主 题: 计算与应用数学拔尖博士生系列论坛——Finite element methods and deep neural networks
报告人: Juncai He (Peking University)
时 间: 2018-05-11 12:00-13:30
地 点: Room 1418, Sciences Building No. 1
12:00-12:30 lunch; 12:30-13:30 Talk
Abstract: Firstly, we will talk about some regularity and discrete compactness results in finite element methods especially for finite element exterior calculus framework. The analysis relies on a new Hodge mapping and its approximation property. Recently, deep learning methods achieved unprecedented and state-of-art results in various machine learning and artificial intelligence tasks. However, there are still many questions for us to understand why deep learning works. We are going to explore some connection between finite element methods and deep neural networks with ReLU activation function.