Introduction to Artificial Inteligence

Assignment 5

Programming assignment - learning

In this assignment, you will implement some learning algorithm (of your choice) for improving some aspect of your RISK playing agent. You are given a free hand to use whatever you wish as the representation of the performance element, be it the evaluation function, move action rules, or other heuristic.

In addition, you may use any representation you find useful (or one you have already used for a previous exercise!), such as weighted sum of features (e.g. for an evaluation function), neural network, set of rules, etc. Use any of the algorithms you learned in order to train your performance element so as to achieve SOME DEGREE of improved play. For example, use some form of hill climbing or simulated annealing to train the weights in the "sum of wighted features" evaluation function.

Deliverables:

  1. Your program.
  2. Documentation on what your program does - specify which aspect of your agent was improved, how it was represented, and what learning algorithm you used to train it.
  3. Report on a set of experiments - how good your agent was before the training, and how it improved and in what manner.

Deadline: February 4, 2004.