Inspiration
This project was inspired by the desire to teach a computer to do things that even we aren't very good at.
What it does
Plays an advanced and difficult form of tic tac toe, and learns without supervision.
How we built it
Combining two machine learning algorithms to provide an answer in every case. The two algoritms are Monte Carlo Trees, and Artificial Neural Networks.
Challenges we ran into
The main challenge was discovering our own set of learning algorithms that allow for normal neural network training when you don't know exactly what went wrong. So we had to come up with our own. Because of the high complexity and search space of the game, storing the monte carlo tree in memory forced us to add more memory to the system.
Accomplishments that we're proud of
We mapped out almost a million states of game, and had the AI play over 40,000 matches.
What we learned
How to use machine learning libraries, and how to get data such that we can successfully teach an AI to play a complex game.
What's next for Ultic AI
The next step would be to try out the system on an even more complex game, such as Go, and to give it better hardware so it can become a master.
Log in or sign up for Devpost to join the conversation.