In this section
@inproceedings{Kobialka:2025,
author = {Kobialka, P and Gerlach, L and Leofante, F and Abraham, E and Silvia, Lizeth TT and Broch, Johnsen E},
publisher = {IJCAI Organization},
title = {Counterfactual strategies for Markov decision processes},
year = {2025}
}
TY - CPAPER
AB - Counterfactuals are widely used in AI to explain how minimal changes to a model’s input can cause a different output. However, established methods for computing counterfactuals focus on one-step decision-making, and are not applicable for sequential decision-making. This paper fills this gap by introducing counterfactuals for Markov decision processes (MDPs), i.e., discrete-time Markov models with non-determinism. During MDP execution, a strategy decides which of the enabled actions (with known probabilistic effects) to execute next. Given an initial strategy that reaches an undesired outcome with a probability above some limit, we identify minimal changes to the initial strategyto reduce that probability below the limit. We encode such counterfactual strategies as solutions to non-linear optimization problems, and further extend this encoding to synthesize diverse counterfactual strategies. We evaluate our approach on four real-world datasets and demonstrate its suitability for providing algorithmic recourse in sophisticated sequential decision-making tasks.
AU - Kobialka,P
AU - Gerlach,L
AU - Leofante,F
AU - Abraham,E
AU - Silvia,Lizeth TT
AU - Broch,Johnsen E
PB - IJCAI Organization
PY - 2025///
TI - Counterfactual strategies for Markov decision processes
ER -
