Bootstrapped Q-learning with Context Relevant Observation Pruning to Generalize in Text-based Games

TL;DR

This paper introduces CREST, a method that prunes irrelevant tokens from observations in text-based games, enabling reinforcement learning agents to better generalize to unseen games with fewer training examples.

Contribution

The paper proposes CREST, a novel observation pruning technique combined with bootstrapped Q-learning, to enhance generalization in text-based game environments.

Findings

Improved generalization on unseen TextWorld games.

Achieved 10x-20x fewer training games compared to prior methods.

Required fewer training episodes while maintaining performance.

Abstract

We show that Reinforcement Learning (RL) methods for solving Text-Based Games (TBGs) often fail to generalize on unseen games, especially in small data regimes. To address this issue, we propose Context Relevant Episodic State Truncation (CREST) for irrelevant token removal in observation text for improved generalization. Our method first trains a base model using Q-learning, which typically overfits the training games. The base model's action token distribution is used to perform observation pruning that removes irrelevant tokens. A second bootstrapped model is then retrained on the pruned observation text. Our bootstrapped agent shows improved generalization in solving unseen TextWorld games, using 10x-20x fewer training games compared to previous state-of-the-art methods despite requiring less number of training episodes.