6.435 -- System Identification
Course Description: Mathematical models of systems from observations of their behavior. Time series, state-space, and input-output models. Model structures, parametrization, and identifiability. Non-parametric methods. Prediction error methods for parameter estimation, convergence, consistency, andasymptotic distribution. Relations to maximum likelihood estimation. Recursive estimation; relation to Kalman filters; structure determination; order estimation; Akaike criterion; and bounded but unknown noise models. Robustness and practical issues. Alternate years.

This class is at the Graduate level
Course Website: http://web.mit.edu/6.435/www/index.html
An example of a syllabus: 6_435_Syllabus.pdf
Instructor: M. A. Dahleh, S. K. Mitter
Prerequisites: 6.241, 6.432

Students's Comments on the Class

From a PhD student in the ORC
I would only recommend this course to someone who is considering doing research in Machine Learning and has taken some introductory courses to the field. The course covered some really theoretical concepts underlying the field of Machine Learning. For anyone who is just interested in doing applied Machine Learning research, this course could be overkill. It did include what I wanted to learn, but the course is still evolving so the organization and presentation of the material could have been improved.

Even though the material was extremely challenging, the course was structured more like a seminar, with a few challenging problem sets and one final project.

Back to Classes