Saikat Chatterjee: Finding truth under noise using time-series and dynamical systems

Abstract:

Let us consider a scenario where we would like to know a time-series from its noisy observations – a case in pursuit of true signal / data under noise. We also wish to forecast what happens next time given observations up to the current time. In statistical signal processing and machine learning, it is called Bayesian state estimation and prediction of stochastic dynamical systems. The famous (and perhaps most useful) method is called Kalman Filter – a classical method. The Kalman Filter requires the model of the true dynamics for the time-series. In this seminar, we will discuss AI-aided Kalman Filters, that do not need to know the model of true dynamics, but still can track and predict the time-series!