Skip to yearly menu bar Skip to main content


Workshop

Learning How Not to Act in Text-based Games

Matan Haroush · Tom Zahavy · Daniel Mankowitz · Shie Mannor

East Meeting Level 8 + 15 #16

Wed 2 May, 4:30 p.m. PDT

Large actions spaces impede an agent's ability to learn, especially when many of the actions are redundant or irrelevant. This is especially prevalent in text-based domains. We present the action-elimination architecture which combines the generalization power of Deep Reinforcement Learning and the natural language capabilities of NLP architectures to eliminate unnecessary actions and solves quests in the text-based game of Zork, significantly outperforming the baseline agents.

Live content is unavailable. Log in and register to view live content