Attribution-Based Explanations for Markov Decision Processes
A new paper on arXiv introduces attribution techniques for Markov Decision Processes (MDPs), extending explainable AI to sequential decision-making. The authors formalize what attributions should represent in MDPs, focusing on importance scores for individual states and execution paths. They leverage strategy synthesis to compute these scores efficiently despite non-determinism. The approach is evaluated on five case studies, demonstrating its utility in providing interpretable insights into sequential decision logic.
Key facts
- Paper published on arXiv with ID 2605.09780
- Introduces attribution techniques for Markov Decision Processes
- Formalizes attributions for states and execution paths
- Uses strategy synthesis for efficient computation
- Evaluated on five case studies
- Addresses gap in explainable AI for sequential decision-making
Entities
Institutions
- arXiv