Introduction to Artificial Inteligence

Programming Assignment - Learning for BackGammon

In this assignment, you will train your backgammon playing program in order to improve its (static) heuristic evaluation function. Use the representation of your choice for the evaluation function (neural networks, Bayes networks, decision tree, weighted sum of features, etc.). Use training from multiple games against 1) A random opponent and 2) Your own program to learn the evaluation fuction. You may use your existing evaluation function from previous assignments as the initial, "untrained" evaluation function, in fact - you are expected to do that, if possble!

Deliverables:

1. Your program.
2. Information on features used, representation of the evaluation function, algorithms.
3. Report on a set of experiments - with initial untrained evaluation function, interim steps, and the most up-to-date results. As a minimum, supply score "average point gain per game" against a random opponent.

Deadline: January 16, 2002.