Carleton University - School of Computer Science Honours Project
Summer 2019
Connect-4 AI using Neural Networks
ABSTRACT
This project seeks to create an AI capable of playing the game Connect-4 using neural networks to learn
the game. After researching previous works in the field of neural network based game AI, the type of
network chosen was a policy-based neural network. This type of network was researched further, and a
neural network using this method was successfully created. After much testing, it was determined that
the neural network created using this type does not perform adequately as an AI. The network gets
caught in a local minima state, where it has learned a strategy far better than any similar variations, so it
never learns to try strategies that are much better.