9.39/9.390 Computational Laboratory in Cognitive Science
Spring Semester, 1999
Tuesdays and Thursdays, 9-10:30am, E10-013
Welcome to the home page of the course 9.39/9.390 Computational Laboratory in Cognitive Science, offered in the Spring of 1999 at MIT. For questions regarding the course, contact Prof. Gert Cauwenberghs, email@example.com, E25-201.
This laboratory course is about models of computation in neural systems, and their implementation in software and hardware. The focus is on machine learning in complex environments, including unsupervised and reinforcement learning. The course is heavily project oriented, and students will implement algorithms in Matlab code, and program a microrobot (Khepera, K-Team) to "learn" to navigate towards a goal while avoiding obstacles. This tiny (5 cm diameter) mobile robot has primitive proximity and light sensors, and the challenge is to learn internal state representations without absolute position encoding, and learn state-action pairs from discrete, delayed reward and punishment.
I assume students are familiar with programming in C or C++. Experience with the Matlab language is also useful. The first week will include a very brief introduction to Matlab. Undergraduates take the course as 9.39, and graduate students as 9.390.
Lectures cover learning models and algorithms with examples of Matlab code. Homework assignments include writing Matlab programs. There is a midterm, but no final exam. Instead, students will prepare a report and in-class presentation for the final project. The project is to be conducted in small groups (2-3 students), and involves programming code for the Khepera microrobot and testing the learning performance in an actual environment. Students are welcome to suggest other projects related to neural computation, including VLSI hardware design.
The textbook for the lectures is An Introduction to Natural Computation (Dana Ballard, MIT Press). You can find copies of the book and also the student version of Matlab (including manual) at the MIT COOP. Advanced topics such as Neuromorphic Analog VLSI Systems will be documented with class notes.