Cmpt 310 Project Proposal
Intelligent Game Playing: The Challenge is Connect Four
Ian Hay ithay@sfu.ca 903014378
Mike Joe mrjoe@sfu.ca 963015744
Jamaluddin Muhammad jmuhamma@sfu.ca 200025281
Application Area
Our
project application area will be to develop an adaptive agent to play Connect
Four against an opponent. Connect Four is a children’s game developed by Milton
Bradley. The game board consists of seven slots each six deep. Players take
turns dropping colored pieces down the slots so in an attempt to link four
either vertically, horizontally or diagonally.
The
agent will play this game using the minimax algorithm. What makes our approach
different is that the evaluation function used to compute min and max will be
computed using a Neural Network. This will allow the agent to learn and adapt.
It may even be possible to run two instances of this program against one
another so that each instance learns. One of the first stages of our project
will be to determine what the inputs to the neural network should be. We will
also be building our own neural network as opposed to using one.
The
ultimate criteria to test our agent is its ability to learn and win. Starting
with a basic knowledge of the game, if it can learn from either playing a
replica of itself or a human player, and develop a moderate competency, then it
will be successful. Also, a human player should not be able to repeat a win by
making the same moves.
To
quantitatively measure its success, we will keep a count of the number of
pieces placed on the board. As the program improves, it will defeat its
opponents’ attacks and should end up placing more pieces.
The
place to buy the original game:
http://www.etoys.com/prod/toy/50061261
Alternative
online Connect Four games:
http://www.stud.ifi.uio.no/~lmariusg/4parad/4parad.html
http://www.theebest.com/games/connect4/connect4.htm
Neural
Network links
http://www.emsl.pnl.gov:2080/proj/neuron/neural/systems/shareware.html
http://www.ci.tuwien.ac.at/docs/services/nnfaq/FAQ.html