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

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.

arXiv commentAccepted to EMNLP 2020

References in corpus (9)

• Distributed Representations of Words and Phrases and their Compositionality
• Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation
• Parameter Space Noise for Exploration
• A Deep Reinforcement Learning Chatbot
• TextWorld: A Learning Environment for Text-based Games
• Learning Dynamic Belief Graphs to Generalize on Text-Based Games
• Counting to Explore and Generalize in Text-based Games
• What can you do with a rock? Affordance extraction via word embeddings
• Playing Text-Adventure Games with Graph-Based Deep Reinforcement Learning