Towards Interpretable Reinforcement Learning Using Attention Augmented Agents

Abstract

Inspired by recent work in attention models for image captioning and question answering, we present a soft attention model for the reinforcement learning domain. This model uses a soft, top-down attention mechanism to create a bottleneck in the agent, forcing it to focus on task-relevant information by sequentially querying its view of the environment. The output of the attention mechanism allows direct observation of the information used by the agent to select its actions, enabling easier interpretation of this model than of traditional models. We analyze different strategies that the agents learn and show that a handful of strategies arise repeatedly across different games. We also show that the model learns to query separately about space and content (`where' vs. `what'). We demonstrate that an agent using this mechanism can achieve performance competitive with state-of-the-art models on ATARI tasks while still being interpretable.

References in corpus (13)

• Show, Attend and Tell: Neural Image Caption Generation with Visual Attention
• Teaching Machines to Read and Comprehend
• Dueling Network Architectures for Deep Reinforcement Learning
• Recurrent Models of Visual Attention
• Multiple Object Recognition with Visual Attention
• End-To-End Memory Networks
• IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
• Residual Attention Network for Image Classification
• Deep Attention Recurrent Q-Network
• Multi-focus Attention Network for Efficient Deep Reinforcement Learning
• Neural Attention Models for Sequence Classification: Analysis and Application to Key Term Extraction and Dialogue Act Detection
• A spatiotemporal model with visual attention for video classification
• CRAM: Clued Recurrent Attention Model

Cited by in corpus (5)

• Towards falsifiable interpretability research
• Automatic Discovery of Interpretable Planning Strategies
• The Sensory Neuron as a Transformer: Permutation-Invariant Neural Networks for Reinforcement Learning
• Interpretable Imitation Learning with Dynamic Causal Relations
• Verifiability and Predictability: Interpreting Utilities of Network Architectures for Point Cloud Processing