Integrating Prior Knowledge in Post-hoc Explanations

In the field of explainable artificial intelligence (XAI), post-hoc interpretability methods aim at explaining to a user predictions trained decision model. Integrating prior knowledge into such aims improving explanation understandability and allowing for personalised explanations adapted each user. this paper, we propose define cost function that explicitly integrates objectives: present general framework optimization problem methods, show can thus be integrated any method by adding compatibility term in function. We instantiate proposed formalization case counterfactual new called Knowledge Integration Counterfactual Explanation (KICE) optimize it. The paper performs an experimental study on several benchmark data sets characterize instances generated KICE, as compared reference methods.