We revisit a game from the literature that characterizes the probabilistic bisimilarity distances of a labelled Markov chain. We illustrate how an optimal policy of the game can explain these distances. Like the games that characterize bisimilarity and probabilistic bisimilarity, the game is played on pairs of states and matches transitions of those states. To obtain more convincing and interpretable explanations than those provided by generic optimal policies, we restrict to optimal policies that delay reaching observably inequivalent state pairs for as long as possible (called 1-maximal) while quickly reaching equivalent ones (called 0-minimal). We present iterative algorithms that compute 1-maximal and 0-minimal policies and prove an exponential lower bound for the number of iterations of the algorithm that computes 1-maximal policies.
@InProceedings{vlasman_et_al:LIPIcs.CONCUR.2025.36, author = {Vlasman, Emily and Nanah Ji, Anto and Worrell, James and van Breugel, Franck}, title = {{Explainability is a Game for Probabilistic Bisimilarity Distances}}, booktitle = {36th International Conference on Concurrency Theory (CONCUR 2025)}, pages = {36:1--36:20}, series = {Leibniz International Proceedings in Informatics (LIPIcs)}, ISBN = {978-3-95977-389-8}, ISSN = {1868-8969}, year = {2025}, volume = {348}, editor = {Bouyer, Patricia and van de Pol, Jaco}, publisher = {Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik}, address = {Dagstuhl, Germany}, URL = {https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.CONCUR.2025.36}, URN = {urn:nbn:de:0030-drops-239861}, doi = {10.4230/LIPIcs.CONCUR.2025.36}, annote = {Keywords: probabilistic bisimilarity distance, labelled Markov chain, game, policy, explainability} }
Feedback for Dagstuhl Publishing