CSCI 3154 - Introduction to AI with Gaming Applications

Fall 2012

Last update: 29 Oct, 2012
Tues, Thurs; CS bldg. Rm 127; 8.30 - 10.00am

Following is a collection of resources provided during the course.
In any one year we will cover a representative sample of the topics listed!

Final Exam

Date: Friday Dec 7, 10.00am
Location: Rm 1009, Kenneth Rowe (Management) Building
Exam Front Page
Outline of marking scheme, text, pre-requisit etc..

Assignments

Assignment late policy.

Assignment 1 - released Sept 11, Due Sept 25 + 1 day ; midnight.
- Ant War results
Assignment 2 - released Sept 27, Due Oct 11; end of day (email to mheywood@cs.dal.ca; pdf format).
Assignment 3 - released Oct 11, Due Oct 25; end of day (email to mheywood@cs.dal.ca; pdf format).
- Membership functions
Assignment 4 - released Oct 30, Due Nov 8; end of day (email to mheywood@cs.dal.ca; pdf format).
- Generic answers
Assignment 5 - released Nov 15, Due Nov 29; end of day (see assignment text for submission instructions!)
- Matlab script - reading training / test partitions and saving classifier labels.
- Training partition - With labels
- Test partition - No labels

Presentations

Presentation groups

A sample from following notes provide the basis for the material covered in the course.

Computer Games and the Role of AI

Computer Game Genres;
- Some game design considerations... and some AI design considerations...
Role of AI;
- Video games as AI's killer application...
- The TravelerDilemma: An example of a 'common sense' reasoning problem.
- The AI Complexity-believability tradeoff: Case study from Halo 2 by Damian Isla
Generic Artificial Intelligence.

Basic Reactive Behaviours

Decision Trees and State Machines;
- To tree or to state (machine)...

Exhaustive Search algorithms

Blind Search
Game Trees and Search: Turn based strategy games.
- Links to animated tutorials: Yosen Lin, Bruce Rosen and Mikael Boden
A* and path finding
Goal orientated decision making: Tactical decision making in F.E.A.R.
- The 'F.E.A.R.' model from Monolith Productions and MIT Media Lab by Jeff Orkin.

Rule Based Methods

Neural Networks and Decision Making

Supervised learning: the general picture.
Perceptron Feedforward model: gradient based learning and Back Propagation.
Cascade Correlation and Multi-Layer Perceptron: gradient based non-linear models.
- MATLAB Neural Network Script: Illustrates perceptron and MLP (with and without Conjugate gradient) on simple 'non-linear' data set;
  - Matlab script: Case of simple 6 pattern data set
  - Matlab script: Case of reading in a training and test partition
    - Pocker training partition
    - Pocker test partition
Temporal difference method: Reinforcement learning under gradient decent.

Evolutionary Computation and Games

Some Terminology: Evolutionary computation and some basic terminology.
Canonical Genetic Algorithm: the base case GA by hand.
Genetic Algorithm to Genetic Program: GP base case.
GP benchmark: Simple Ant.
- little GP: Local copy of lilGP (care of Michigan State, GARAGe lab)
Case studies:
- Case study 1: NERO first person shooter;
  - The NERO game engine and homepage
- Case study 2: GP backgammon-chess-robocode;
- Case study 3: GP AntWar - GECCO'07 competition winner;
Schema Theory: or how solutions are composed under GAs.
- QA on GA schema.

AI Gaming Resources

AI and games specifically

Game development in general plus AI

Author: Malcolm Heywood