Abstract
Recent years have witnessed a renewed interest in Boolean functions in explaining binary classifiers in the field of explainable AI (XAI). The standard approach to Boolean functions is based on propositional logic. We present a modal language of a ceteris paribus nature which supports reasoning about binary classifiers and their properties. We study a family of classifier models, axiomatize it and show completeness of our axiomatics. Moreover, we prove that satisfiability checking for our modal language relative to such a class of models is NP-complete. We leverage the language to formalize counterfactual conditional as well as a variety of notions of explanation including abductive, contrastive and counterfactual explanations, and biases. Finally, we present two extensions of our language: a dynamic extension by the notion of assignment enabling classifier change and an epistemic extension in which the classifier’s uncertainty about the actual input can be represented.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
So the classifier we model here is slightly more expressive than Boolean classifier. Introducing decision atoms \(\mathsf {t}(x), \mathsf {t}(y), \dots \) below allows us to encode more than two decision values (classifications). Sometime we also use binary/Boolean classifier in this more general sense. Notice that we cannot use the term psuedo-Boolean, since in Boolean function it means \( Val = \mathfrak {R}\) [5], but we need our \( Val \) staying finite.
- 2.
In fact it appears to be a prime implicant, when we formally introduce this notion.
- 3.
- 4.
There are other options besides measuring distance by cardinality, e.g., distance in sense of subset relation as [2]. We will consider them in further research.
- 5.
A similar approach of ceteris paribus is [11]. They also refine Lewis’ semantics for counterfactual by selecting the closest worlds according to not only the actual world and antecedent, but also a set of formulas where they note as \(\Gamma \). The main technical difference is that they allow any counterfactual-free formula as a member of \(\Gamma \), while in our setting X only contains atomic formulas.
- 6.
A remarkable fact is that not all \(\Rightarrow _X\) satisfy the strong centering condition, which says that the actual world is the only closest world when the antecedent is already true here. To see it, consider a toy classifier model (C, s) such that \(S = \{s, s',s'',s''' \}\) with \(s = \{p, q\}\), \(s' = \{p \}\), \(s'' = \{q\}\), \(s''' = \emptyset \). We have \( closest _{C}(s{,}p{,}\emptyset ) = \{s,s'\}\), rather than \( closest _{C}(s{,}p{,}\emptyset ) = \{s\}\). All the rest of conditions in \(\mathsf {VC}\) are satisfied regardless of what X is.
- 7.
The symbol \(\triangle \) denotes symmetric difference.
- 8.
For the significance of actionablility in XAI, see e.g. [26].
- 9.
Notice that \(\mathsf {cn}_{Y{,} Atm \setminus Dec }\) is just another expression of \(\widehat{s}\) where \(s = Y\).
References
Biran, O., Cotton, C.: Explanation and justification in machine learning: a survey. In: IJCAI 2017 Workshop on Explainable AI (XAI), vol. 8, no. 1, pp. 8–13 (2017)
Borgida, A.: Language features for flexible handling of exceptions in information systems. ACM Trans. Database Syst. (TODS) 10(4), 565–603 (1985)
Caridroit, T., Lagniez, J.-M., Le Berre, D., de Lima, T., Montmirail, V.: A SAT-based approach for solving the modal logic S5-satisfiability problem. In: Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence (AAAI 2017), pp. 3864–3870. AAAI Press (2017)
Charrier, T., Herzig, A., Lorini, E., Maffre, F., Schwarzentruber, F.: Building epistemic logic from observations and public announcements. In: Proceedings of the Fifteenth International Conference on Principles of Knowledge Representation and Reasoning (KR 2016), pp. 268–277. AAAI Press (2016)
Crama, Y., Hammer, P.L.: Boolean Functions: Theory, Algorithms, and Applications. Cambridge University Press, Cambridge (2011)
Dalal, M.: Investigations into a theory of knowledge base revision: preliminary report. In: Proceedings of the Seventh National Conference on Artificial Intelligence, vol. 2, pp. 475–479. Citeseer (1988)
Darwiche, A., Hirth, A.: On the reasons behind decisions. In: 24th European Conference on Artificial Intelligence, ECAI 2020. Frontiers in Artificial Intelligence and Applications, vol. 325, pp. 712–720. IOS Press (2020)
Dhurandhar, A., et al.: Explanations based on the missing: towards contrastive explanations with pertinent negatives. In: Advances in Neural Information Processing Systems, pp. 592–603 (2018)
Dretske, F: Meaningful perception. An Invitation to Cognitive Science: Visual Cognition, pp. 331–352 (1995)
Fagin, R., Moses, Y., Halpern, J.Y., Vardi, M.Y.: Reasoning about Knowledge. MIT Press, Cambridge (1995)
Girard, P., Triplett, M.A.: Ceteris paribus logic in counterfactual reasoning. In: TARK 2015, pp. 176–193 (2016)
Grossi, D., Lorini, E., Schwarzentruber, F.: The ceteris paribus structure of logics of game forms. J. Artif. Intell. Res. 53, 91–126 (2015)
Hempel, C.G., Oppenheim, P.: Studies in the logic of explanation. Philos. Sci. 15(2), 135–175 (1948)
Herzig, A., Lorini, E.: A modal logic of perceptual belief. In: Lihoreau, F., Rebuschi, M. (eds.) Epistemology, Context, and Formalism. SL, vol. 369, pp. 197–211. Springer, Cham (2014). https://doi.org/10.1007/978-3-319-02943-6_12
Herzig, A., Lorini, E., Maffre, F.: A poor man’s epistemic logic based on propositional assignment and higher-order observation. In: van der Hoek, W., Holliday, W.H., Wang, W. (eds.) LORI 2015. LNCS, vol. 9394, pp. 156–168. Springer, Heidelberg (2015). https://doi.org/10.1007/978-3-662-48561-3_13
Ignatiev, A., Cooper, M.C., Siala, M., Hebrard, E., Marques-Silva, J.: Towards formal fairness in machine learning. In: Simonis, H. (ed.) CP 2020. LNCS, vol. 12333, pp. 846–867. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-58475-7_49
Ignatiev, A., Narodytska, N., Asher, N., Marques-Silva, J.: From contrastive to abductive explanations and back again. In: Baldoni, M., Bandini, S. (eds.) AIxIA 2020. LNCS (LNAI), vol. 12414, pp. 335–355. Springer, Cham (2021). https://doi.org/10.1007/978-3-030-77091-4_21
Ignatiev, A., Narodytska, N., Marques-Silva, J.: Abduction-based explanations for machine learning models. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, pp. 1511–1519 (2019)
Kment, B.: Counterfactuals and explanation. Mind 115(458), 261–310 (2006)
Lewis, D.: Counterfactuals. Harvard University Press, Cambridge (1973)
Martens, D., Provost, F.: Explaining data-driven document classifications. MIS Q. 38(1), 73–100 (2014)
Mittelstadt, B., Russell, C., Wachter, S.: Explaining explanations in AI. In: Proceedings of the Conference on Fairness, Accountability, and Transparency, pp. 279–288 (2019)
Mothilal, R.K., Sharma, A., Tan, C.: Explaining machine learning classifiers through diverse counterfactual explanations. In: Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency, pp. 607–617 (2020)
Quine, W.V.: A way to simplify truth functions. Am. Math. Mon. 62(9), 627–631 (1955)
Shi, W., Shih, A., Darwiche, A., Choi, A.: On tractable representations of binary neural networks. arXiv preprint arXiv:2004.02082 (2020)
Sokol, K., Flach, P.A.: Counterfactual explanations of machine learning predictions: opportunities and challenges for AI safety. In: SafeAI@ AAAI (2019)
Van Benthem, J., Van Eijck, J., Kooi, B.: Logics of communication and change. Inf. Comput. 204(11), 1620–1662 (2006)
van der Hoek, W., Iliev, P., Wooldridge, M.J.: A logic of revelation and concealment. In: Proceedings of the International Conference on Autonomous Agents and Multiagent Systems, (AAMAS 2012), pp. 1115–1122. IFAAMAS (2012)
Van Der Hoek, W., Troquard, N., Wooldridge, M.J.: Knowledge and control. In: Proceedings of the 10th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2021), pp. 719–726. IFAAMAS (2011)
van Ditmarsch, H., van Der Hoek, W., Kooi, B.: Dynamic Epistemic Logic. Synthese Library, vol. 337. Springer, Heidelberg (2007)
van Ditmarsch, H.P., van der Hoek, W., Kooi, B.P.: Dynamic epistemic logic with assignment. In: Proceedings of the 4th International Joint Conference on Autonomous Agents and Multiagent Systems (AAMAS 2005), pp. 141–148. ACM (2005)
Verma, S., Dickerson, J., Hines, K.: Counterfactual explanations for machine learning: a review. arXiv preprint arXiv:2010.10596 (2020)
Acknowledgements
Support from the ANR-3IA Artificial and Natural Intelligence Toulouse Institute is gratefully acknowledged.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2021 Springer Nature Switzerland AG
About this paper
Cite this paper
Liu, X., Lorini, E. (2021). A Logic for Binary Classifiers and Their Explanation. In: Baroni, P., Benzmüller, C., Wáng, Y.N. (eds) Logic and Argumentation. CLAR 2021. Lecture Notes in Computer Science(), vol 13040. Springer, Cham. https://doi.org/10.1007/978-3-030-89391-0_17
Download citation
DOI: https://doi.org/10.1007/978-3-030-89391-0_17
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-89390-3
Online ISBN: 978-3-030-89391-0
eBook Packages: Computer ScienceComputer Science (R0)