Contrastive Explanations for Reinforcement Learning via Embedded Self Predictions

Abstract

We investigate a deep reinforcement learning (RL) architecture that supports explaining why a learned agent prefers one action over another. The key idea is to learn action-values that are directly represented via human-understandable properties of expected futures. This is realized via the embedded self-prediction (ESP)model, which learns said properties in terms of human provided features. Action preferences can then be explained by contrasting the future properties predicted for each action. To address cases where there are a large number of features, we develop a novel method for computing minimal sufficient explanations from anESP. Our case studies in three domains, including a complex strategy game, show that ESP models can be effectively learned and support insightful explanations.

arXiv commentPublished (Oral) at ICLR 2021

References in corpus (7)

• Continuous control with deep reinforcement learning
• Axiomatic Attribution for Deep Networks
• Hybrid Reward Architecture for Reinforcement Learning
• Visualizing and Understanding Atari Agents
• General Value Function Networks
• Counterfactual States for Atari Agents via Generative Deep Learning
• Transparency and Explanation in Deep Reinforcement Learning Neural Networks

Cited by in corpus (1)

• Bridging the Gap: Providing Post-Hoc Symbolic Explanations for Sequential Decision-Making Problems with Inscrutable Representations