Robust Filtering – Novel Statistical Learning and Inference Algorithms with Applications

Abstract: Filtering or online state estimation of dynamical systems is a key task with applications in several fields. However, it is common to have gross errors in the measurements which violate the basic assumption of the standard filtering setup i.e. apriori knowledge of the system model.   Hence, the concept of robust filtering is useful which can handle different kinds of gross errors including outliers and biases in the data. Our proposed robust state estimators are based on modeling the data anomalies in the generative structure of the state-space models and subsequently performing joint state and parameter estimation using tools from Bayesian theory. In this talk, I ll first discuss the case of handling outliers in filtering. Then I ll talk how we extend these results to a general estimation setup before a discussion on mitigating the effect of biased observations in filtering. Lastly, I plan to discuss how we handle the occurrence of both outliers and biases in observations. I intend to discuss performance evaluation mostly in terms of estimation accuracy and computational strain in applications including target tracking, indoor positioning, spatial perception problems.

Bio: Aamir is a visiting research scholar working with Professor Melkior Ornik in the LEADCAT group. He is a graduating PhD student in Electrical Engineering at the Lahore University of Management Sciences (LUMS), Pakistan where his research has been focused on information engineering. In particular, he has been interested in devising different mathematical algorithms using tools from statistical signal processing, mostly from a Bayesian perspective. The developed techniques have been successfully tested in different problems including target tracking, indoor positioning, spatial perception problem etc.