id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
cord-232238-aicird98	Ferrario, Andrea	A Series of Unfortunate Counterfactual Events: the Role of Time in Counterfactual Explanations	2020-10-09		.txt	text/plain	6857	329	39	However, at the basis of any discussion on post-hoc explanations lies the assumption that the machine learning model whose outcomes have to be explained remains "stable" or does not change, in a given time frame of interest [2, 9, 19] . This time delay may lead to the emergence of unfavorable cases-called "unfortunate counterfactual events" (UCE) in these notes-where the retraining of the machine learning model invalidates the efforts of an individual who successfully implemented the scenario originally recommended by a feasible, actionable and possibly sparse counterfactual explanation. As noted in Section 3, the degree of certainty of counterfactual scenarios is computed as result of the machine learning model retraining, i.e., only after the generation of the corresponding counterfactual explanation (at time 0 ). In Table 1 we enumerate all possible cases that emerge from the change in time of data points, machine learning models and their outcomes, when considering the implementation of counterfactual scenarios.	./cache/cord-232238-aicird98.txt	./txt/cord-232238-aicird98.txt