Stochastic continuous depth networks are mathematical models for analyzing the behavior of residual neural networks when many stochastic network layers are stacked. In the limit of many layers of infinitesimal changes they can be described by a stochastic differential equation. Such continuous-depth networks share the expressiveness of deep neural networks and can be trained efficiently through back-propagation. In the talk I discuss recurrent stochastic continuous depth networks for input and output data in time domain.
Moritz Schauer is an Associate Senior Lecturer at the Department of Mathematical Sciences.
Online, register to receive the link
10 June, 2022, 14:00
10 June, 2022, 15:00