Short Term Course on "Nonlinear Estimation for Engineers" 01-05 February 2020
The problem of estimating the unobserved states of a system from observed data often arises in many branches of science, ranging from tracking the location of an object from radar-based observations to estimating the volatility from observed prices of financial securities. Filtering refers to any method for obtaining such state estimates, recursively in time, by combining model predictions with noisy observations. If the future state depends linearly on the present state, a state estimator which is optimal in certain sense is known after its inventor as Kalman filter (KF) is popular in engineering, finance and econometrics since 1970s. While the use of KF is well-understood for linear models, nonlinear models are often needed to describe the observed dynamics adequately. Exact nonlinear filtering is often impossible and various Bayesian approximations exist to solve the filtering problem. Over the last two decades, significant advances have been made in theory and applications of nonlinear filtering. The proposed course will guide the participants through the theory and practice of linear and nonlinear filtering. Theory, including the state of the art developments in filtering as applied in different branches of engineering will be taught through class-room based lectures. The participants will have an opportunity to implement the filtering algorithms they learn in the class on prototype simulations in a high level programming language such as Matlab, in supervised tutorial session.