Fall 2002 schedule: Tuesday and Thursday 10:30-12 in NE20-461.
Organization of synaptic connectivity as the basis of neural computation and learning. Perceptrons. Dynamical theories of recurrent networks: amplifiers, attractors, and hybrid computation. Backpropagation and Hebbian learning. Models of perception, motor control, memory, and neural development.
|
Prof. Sebastian Seung seung@mit.edu E25-435a, x2-1693 |
TA: Russ Tedrake russt@ai.mit.edu E25-214, x2-2691 Office hours by appointment |
The subject will focus on basic mathematical concepts for understanding nonlinearity and feedback in neural networks, with examples drawn from both neurobiology and computer science. Most of the subject is devoted to recurrent networks, because recurrent feedback loops dominate the synaptic connectivity of the brain. There will be some discussion of statistical pattern recognition, but less than in the past, because this perspective is now covered in 6.893 Machine Learning and Neural Networks. Instead the connections to dynamical systems theory will be emphasized.
Modern research in theoretical neuroscience can be divided into three categories: cellular biophysics, network dynamics, and statistical analysis of neurobiological data. This subject is about the dynamics of networks, but excludes the biophysics of single neurons, which will be taught in 9.29 Introduction to Computational Neuroscience.