Counterfactual Explanations for Reinforcement Learning Agents

Counterfactual Explanations for Reinforcement Learning Agents

연구 분야: Artificial Intelligence

학회: AAMAS '23: Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems

Reinforcement learning (RL) algorithms often use neural networks to represent agent's policy, making them difficult to interpret. Counterfactual explanations are human-friendly explanations which offer users actionable advice on how to change their features to obtain a desired output from a black-box model. However, methods for generating counterfactuals in RL ignore the stochastic and sequential nature of RL tasks, and can generate counterfactuals which are difficult to obtain, affecting user effort and trust. My dissertation focuses on developing methods that take into account the complexities of RL framework and provide counterfactual explanations that are easy to reach and confidently produce the desired output