IJCAI 2023 is the 32nd International Joint Conference on Artificial Intelligence and is taking place in Macao, SAR from 19 to 25 August 2023.
Deep learning is increasingly used to learn policies for planning problems, yet policies represented by neural networks are difficult to interpret, verify and trust. Existing formal approaches to post hoc explanations provide concise reasons for a single decision made by an ML model. However, understanding planning policies requires explaining sequences of decisions.
In the paper that is being presented, the authors formulate the problem of finding explanations for the sequence of decisions recommended by a learned policy in a given state. They show that, under certain assumptions, a minimal explanation for a sequence can be computed by solving a number of single decision explanation problems which is linear in the length of the sequence. The authors present experimental results of the implementation of this approach for ASNet policies for classical planning domains.