David Tao is a Singaporean undergraduate CS student currently studying at McGill University, and also a research intern at Microsoft Research Montreal. His current research resides in the intersection between language and reinforcement learning.
Learning with text-based games
The exponential nature of action and state spaces make text-based dialogue systems difficult to solve for most machine algorithms. This is especially true for dynamic action spaces that condition on state. To solve this problem we propose an architecture to learn our action space and a control policy to solve text-based games - a simplistic model for dialogue systems. We use games generated by the TextWorld framework and show promising results on a carrot-cooking game.