About

Teaching Computers to Play Video Games: An Introduction to Unsupervised Machine Learning using Atari and Super Mario Bros.

Google's DeepMind division made history last year by developing an algorithm to learn and master 49 classic Atari games, using only the pixels on the screen and the score. Machine learning is one of the most exciting and nascent fields in computer science today, particularly "unsupervised" learning, where humans essentially just set up the the system and let the machine figure out the rest on its own. Remember Space Invaders, Pong, and Breakout? After a bit of time with each of these games, the algorithm learns to play better than a human can! Most fascinating of all, these algorithms may mimic how we ourselves reason and learn, so the exploration of machine learning is akin to the exploration of our own minds.

This talk will explain how Google's "Deep-Q Learning" algorithm works and how it could be applied in other areas. Beyond Atari, we will demo two machine learning approaches to learning Super Mario Bros: a genetic algorithm and a refinement to Deep-Q Learning to play this much more complex game. This talk does not require any background knowledge of machine learning or programming.

This is a free event for the community. If you'd like further information please contact Nathan Abell at n_abell@yahoo.com