A0730
Title: Deep neural networks estimation for non-IID data
Authors: Chao Zheng - University of Southampton (United Kingdom) [presenting]
Abstract: The theoretical development of deep neural networks has been heavily investigated in recent years, and most works have been focused on the setting of independent data. The theoretical properties of deep feedforward ReLU networks are studied in modeling non-linear non-IID mixing sequences, which include a wide range of time series models. Non-asymptotic generalization bounds are presented for the estimation error, and it is shown that it is related to the data's dependence structure and the DNN architecture.
