System Identification
This MIT OpenCourseWare graduate course covers how to build mathematical models of dynamic systems directly from observed input-output data. Topics include time series, state-space, and input-output model structures, questions of parametrization and identifiability, and non-parametric estimation methods. The core of the course develops prediction error methods for parameter estimation, covering convergence, consistency, and asymptotic distribution, and connects these to maximum likelihood estimation and recursive estimation techniques including Kalman filters. Later sessions address structure determination, order estimation using the Akaike criterion, and robustness under bounded but unknown noise. Materials include lecture notes and problem sets drawn from MIT's graduate curriculum in electrical engineering and computer science. The course assumes prior background in linear systems and probability, and it suits students who want the statistical theory behind building predictive models from real measured data rather than idealized equations.