Emergent Solutions to High Dimensional Reinforcement Tasks with Tangled Program Graphs
Algorithms that learn through environmental interaction and delayed rewards, or reinforcement learning, increasingly face the challenge of scaling to dynamic, high-dimensional, and partially observable environments. Significant attention is being paid to frameworks from deep learning, which scale to high-dimensional data by decomposing the task through multi-layered neural networks. While effective, the representation is complex and computationally demanding. In this work we propose a framework based on genetic programming, which adaptively scales the complexity of policies through interaction with the task. We make a direct comparison with a recently proposed deep reinforcement learning framework in the challenging Atari video game environment. Results indicate that the proposed approach matches the quality of deep learning while being a minimum of three orders of magnitude simpler with respect to model complexity.
Stephen Kelly is a PhD candidate in the Faculty of Computer Science at Dalhousie University. His thesis research is concerned with machine learning in video games, with particular emphasis on building non-person characters with modular genetic programming. Prior to entering the PhD program, Stephen received a Master of Computer Science degree from Dalhousie and a Bachelor of Fine Arts from the Nova Scotia College of Art and Design. He has maintained an art practice throughout graduate studies at Dalhousie, crossing art and science within public installations and ongoing research projects in art and artificial life.
Slonim Conference Room (#430), Goldberg Computer Science Building