COGS 202: Neural Computation & Computational Models of Cognition

Instructor: Emo Todorov

TA: Brad Aimone TA Sessions TBD

Lectures: Wednesday 3:30-6:30, CSB 003

Updated Info (June 7^th, 2007)

Course Evaluations: http://www-cogsci.ucsd.edu/~bwalton/course_eval.html

Homeworks: If you have not turned in your homework 2 (or even homework 1), please do so ASAP. The quarter is ending very soon…

Final Projects: our tentative time is class time (3:30-6:30), Wednesday June 13^th. If the room is not available, we will adjust our plans.

Because there are numerous presentations, plan on 10 minutes. It would be nice if you could leave a minute or two for a couple of questions.

Also, please add comments to your slides for Emo to see what you did and e-mail him. As with last week’s presentations, e-mail me your slides prior to the talk so we can speed transition times.

Suggested Reading: Jamie suggested that I put up Jon Shlens' thorough description of PCA up for everyone. Shlens' PCA Review: This may help develop an intuition for this tool.

Schedule for rest of quarter:

Homework #2 : Due on May 30^th. I recommend that you work on it earlier and start working on your paper presentation, as we will also have a 3^rd homework before the end of the quarter.

Paper presentation (June 6^th):

In groups of two, present a paper that uses a computational model to either analyze a system or interpret a dataset. Email myself and Emo with the paper that you are planning to present. Recommended sources of papers are: Nature, Science, Nature Neuroscience, Journal of Neural Computation, Neural Networks, IEEE and CoSyne meeting abstracts. Plan on about a ten minute presentation of the paper, with potential for questions afterwards.

Final Project (Finals week sometime)

Either individually or in groups of two, pick a computationally interesting project and do something with it. Discuss desired options with Emo. Groups should try more ambitious projects…

Here is the homework (Assignment #1). This involves calculating network gradients using both backpropagation and finite difference method. Please e-mail me your code by next Monday (April 30^th).

Slides regarding backpropagation from Emo

This also has a very nice illustration of backprop by Miro.

http://cogsci.ucsd.edu/~menev/nat_comp/assignments/backpropagation/backpropagation.pdf

-----------

Grading: ~4 MATLAB assignments, 1 paper presentation, and final project

Readings: Papers posted on website

Topics to be covered:

Mathematical foundations

- Linear algebra
- Multivariate calculus
- Probability theory
- Numerical optimization

Learning

- Backpropagation in feedforward and recurrent networks
- Hebbian learning, associative memory, map formation
- Reinforcement learning and neural coding of reward
- Unsupervised learning and dimensionality reduction

Sensory systems

- Bayesian inference models of behavioral and neural data
- Neural codes and natural scene statistics
- Estimation over time (Kalman filters, Hidden Markov Models)

Motor systems

- Optimal control models of behavioral and neural data
- Motor adaptation and internal model formation
- Pattern generators and locomotion

Guest lectures (dates TBD)

- Jeff Elman: computational models of language
- Rik Belew: Artificial Intelligence, computational complexity

Assigned Readings:

· Week 1:

1. Introduction to Linear Algebra Concepts

2. 2^nd Linear algebra & Matlab overview

3. Matlab Review

4. Sample Matlab code

· Week 3:

1. Slides regarding backpropagation from Emo

2. Assignment #1 (Due April 30^th)

· Week 7:

1. Assignment #2 (Due May 30^th)

· Week 9:

1. Assignment #3

· Week 10:

1. Paper presentations

Final projects / presentations

Other websites for information

Emo's Winter Course: Natural Computation (COGS 118A) : Contains lecture notes, readings and additional materials