CS 420: Advanced Topics in Machine Intelligence

Spring 2000: Natural Computation

Grades are available here

Instructor:
Bruce MacLennan
Phone: 974-5067
Office: 110-A Ayres
Hours: TBA or make an appointment
Email: maclennan@cs.utk.edu

Teaching Assistant:
None

Classes: 12:40-1:55 Tues. & Thurs. in Ayres 102

Directory of Handouts, Labs, etc.

This page: http://www.cs.utk.edu/~mclennan/Classes/420-S00

Information

Description

CS 420 covers advanced topics in machine intelligence with an emphasis on faculty research. In the Spring semester of 2000 the topic will be natural computation.

This course will be based on a new text, Dana H. Ballard's An Introduction to Natural Computation (MIT Press, 1997).

Natural computation is the study of the computational processes in natural systems, expecially the learning algorithms found in nature. These algorithms operate on at least three timescales: (1) learning to react (neural networks), (2) learning during a lifetime (reinforcement learning), and (3) learning across generations (genetic algorithms & genetic programming). This course will present these three approaches to learning and the essential ideas they have in common.

Prerequisites

Basic calculus through differential equations (Mat 231, 241), linear algebra (Mat 251), probability and statistics (Mat 222). Ability to program in Lisp is not necessary for this AI course, since neural network programs use simple data structures. However, you should be able to program competently in some programming language.

Grading

Subject to change! I currently anticipate that there will be one homework assignment per week, mostly exercises from the book. A few of these (probably 3 or 4) will be small programming projects, which you may implement in any reasonable programming language. Generally I will assign homework on Thursday to be turned in the following Tuesday, but you will have more time for the programming projects.

At this time I do not anticipate having a final exam, so your grade will be based on your homework. Of course, all homework should be your own work.

Text

Dana H. Ballard, An Introduction to Natural Computation (MIT Press, 1997)

Tentative List of Topics

  1. Natural Computation (overview)
  2. Fitness (probability distributions, information theory etc.)
  3. Programs (search etc.)
  4. Data (compression, representation etc.)
  5. Dynamics (linear and nonlinear systems)
  6. Optimization (minimization, optimal control)
  7. Content-addressable Memories (Hopfield, Kanerva, RBF etc.)
  8. Supervised Learning (perceptrons, back-propagation etc.)
  9. Unsupervised Learning (PCA, competitive learning etc.)
  10. Markov Models (regular chains, nonregular chains, hidden M.M.)
  11. Reinforcement Learning (Markov decision process, Q-learning etc.)
  12. Genetic Algorithms (schemata, coevolution)
  13. Genetic Programming (operators, programming, modules)
We will do about one topic per week.
Return to MacLennan's home page

Send mail to Bruce MacLennan / MacLennan@cs.utk.edu
Last updated: Fri May 12 10:15:50 EDT 2000