EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

3562. Multistage counterfactual decisions

Invited abstract in session MD-27: Mathematical Optimization for Counterfactual Explanations, stream Mathematical Optimization for XAI.

Monday, 14:30-16:00
Room: 047 (building: 208)

Authors (first author is the speaker)

1. Emilio CARRIZOSA
IMUS - Instituto de Matemáticas de la Universidad de Sevilla
2. Jasone Ramírez-Ayerbe
Université de Montréal
3. Dolores Romero Morales
Copenhagen Business School

Abstract

Counterfactual Analysis is a powerful tool for Explainable Artificial Intelligence. In Supervised Classification, this means associating with each record a so-called counterfactual decision, i.e., an instance close to the record (closeness measured by an appropriate and context-dependent dissimilarity) and whose probability of being classified by a given classifier in the positive (desired) class is high. In other words, the counterfactual analysis is a reference for a given individual: how he/she should minimally change to improve the probability of being classified in the positive class.
While the literature has focused on the case of finding one counterfactual, we propose in the talk how to address the problem in which a sequence of k decisions, gradually increasing the probabilities of positive, and implying an acceptable cost at each stage.
A mathematical optimization model is considered and theoretical properties are derived.

Keywords

Status: accepted

Back to the list of papers