Deep Neural Networks and the Curse of Dimensionality
组织者
演讲者
孙泽钜
时间
2023年07月10日 14:30 至 15:00
地点
数学系理科楼A-203
摘要
In this talk, we will discuss the approximation capability of a deep neural network in high dimensions. Especially, we would like to deal with the question: in what kind of circumstances can a neural network overcomes the curse of dimensionality. A partial answer to this question is that the neural network is necessary to be deep enough, and there exists a counter-example for shallow neural networks. This talk is based on the recent work of P. Grohs et al.: arXiv:2103.04488 and A. Riekert: arXiv:2304.05790.