BIMSA >
Seminar on Control Theory and Nonlinear Filtering
Recurrent Neural Networks and Universal Approximation of Bayesian Filters
Recurrent Neural Networks and Universal Approximation of Bayesian Filters
Organizer
Speaker
Yangtianze Tao
Time
Tuesday, November 8, 2022 9:00 PM - 9:30 PM
Venue
Online
Abstract
The Bayesian optimal filtering problem is to estimate some conditional statistics of a latent time-series signal from an observation sequence. Classical approaches often rely on the use of assumed or estimated transition and observation models. Using a general architecture for recurrent neural networks and attempting to directly learn a recursive mapping from observational inputs to the desired estimator statistics is an alternative strategy. In this presentation, we shall talk about the following topics. 1.The approximation capabilities of this framework and the approximation error bounds for filtering in general non-compact domains. 2.The strong time-uniform approximation error bounds that guarantee good long-time performance.