Graduate Student Seminar | Avneet Kaur, System estimation using a recurrent neural network

Candidate

Avneet Kaur  | University of Waterloo

Title

System estimation using a recurrent neural network

Abstract

While it is well-known that a  Kalman filter is an  optimal state estimator of linear systems (given Gaussian noises), optimal estimation of non-linear systems of ordinary differential equations is an open problem. In this talk we will discuss some first steps on a pathway to optimal state estimation using neural nets. We propose to use a recurrent neural network(RNN) to estimate the state of a system. Our codes will be tested on a number of examples, including the Lorenz attractor. The performance of the neural net based estimator will be compared to the popular  extended Kalman filter(EKF).