License: Creative Commons Attribution 4.0 International license (CC BY 4.0)
When quoting this document, please refer to the following
DOI: 10.4230/LIPIcs.CP.2021.35
URN: urn:nbn:de:0030-drops-153261
URL: https://drops.dagstuhl.de/opus/volltexte/2021/15326/
Go to the corresponding LIPIcs Volume Portal


Korikov, Anton ; Beck, J. Christopher

Counterfactual Explanations via Inverse Constraint Programming

pdf-format:
LIPIcs-CP-2021-35.pdf (1 MB)


Abstract

It is increasingly recognized that automated decision making systems cannot be black boxes: users require insight into the reasons that decisions are made. Explainable AI (XAI) has developed a number of approaches to this challenge, including the framework of counterfactual explanations where an explanation takes the form of the minimal change to the world required for a user’s desired decisions to be made. Building on recent work, we show that for a user query specifying an assignment to a subset of variables, a counterfactual explanation can be found using inverse optimization. Thus, we develop inverse constraint programming (CP): to our knowledge, the first definition and treatment of inverse optimization in constraint programming. We modify a cutting plane algorithm for inverse mixed-integer programming (MIP), resulting in both pure and hybrid inverse CP algorithms. We evaluate the performance of these algorithms in generating counterfactual explanations for two combinatorial optimization problems: the 0-1 knapsack problem and single machine scheduling with release dates. Our numerical experiments show that a MIP-CP hybrid approach extended with a novel early stopping criteria can substantially out-perform a MIP approach particularly when CP is the state of the art for the underlying optimization problem.

BibTeX - Entry

@InProceedings{korikov_et_al:LIPIcs.CP.2021.35,
  author =	{Korikov, Anton and Beck, J. Christopher},
  title =	{{Counterfactual Explanations via Inverse Constraint Programming}},
  booktitle =	{27th International Conference on Principles and Practice of Constraint Programming (CP 2021)},
  pages =	{35:1--35:16},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-211-2},
  ISSN =	{1868-8969},
  year =	{2021},
  volume =	{210},
  editor =	{Michel, Laurent D.},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/opus/volltexte/2021/15326},
  URN =		{urn:nbn:de:0030-drops-153261},
  doi =		{10.4230/LIPIcs.CP.2021.35},
  annote =	{Keywords: Explanation, Inverse Optimization, Scheduling}
}

Keywords: Explanation, Inverse Optimization, Scheduling
Collection: 27th International Conference on Principles and Practice of Constraint Programming (CP 2021)
Issue Date: 2021
Date of publication: 15.10.2021


DROPS-Home | Fulltext Search | Imprint | Privacy Published by LZI