Abstract
First-order typed model counting extends first-order model counting by the ability to distinguish between different types of models. In this paper, we exploit this benefit in order to calculate weighted conditional impacts (WCIs) which play a central role in nonmonotonic reasoning based on conditionals. More precisely, WCIs store information about the verification and the falsification of conditionals with respect to a possible worlds semantics, and therefore serve as sufficient statistics for maximum entropy (ME) distributions as models of probabilistic conditional knowledge bases. Formally, we annotate formulas with algebraic types that encode concisely all structural information needed to compute WCIs, while allowing for a systematic and efficient counting of models. In this way, our approach to typed model counting for ME-reasoning integrates both structural and counting aspects in the same framework.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
Notes
- 1.
Non-conditional statements of the form “\(A\in \mathsf {FOL}\) holds with probability p” can be accommodated into \(\mathcal {R}\) by adding the conditional \((A|\top )[p]\) to \(\mathcal {R}\).
- 2.
We usually omit the operation symbol “\(\cdot \)”.
- 3.
The operation \(\circ \) shall be the one that binds weakest.
- 4.
We write edge labels in dashed frames and omit the label “\(\mathbf {1_\mathcal {S}}\)”.
- 5.
Note that \(\mathsf {sd}\text {-}\mathsf {DNNF}^{\mathcal {S}}\)-circuits principally represent grounded formulas, and so are \(A,B,C\in \mathsf {FOL}^\mathcal {S}\) in the following definitions. The power of \(\mathsf {sd}\text {-}\mathsf {DNNF}^{\mathcal {S}}\)-circuits lies in the capability of consolidating isomorphic formulas via set conjunction respectively set disjunction.
- 6.
In \(\mathsf {TMC}(A)\) resp. \(\mathsf {TMC}(B)\), those structured interpretations are considered that are restricted to the ground atoms in A resp. B.
- 7.
Here, \(\forall _{X\in \mathsf {var}(r_i)} X.\phi _i(r_i)=\forall X_1. (\ldots \forall X_m.\phi _i(r_i))\) where \(\mathsf {var}(r_i)=\{X_1,\ldots ,X_m\}\). Note that \(\mathsf {var}(r_i)=\mathsf {var}(\phi _i(r_i))\) for \(i=1,\ldots ,n\), which can be proved easily.
References
Van den Broeck, G.: Lifted inference and learning in statistical relational models. Ph.D. thesis, KU Leuven (2013)
Van den Broeck, G., Taghipour, N., Meert, W., Davis, J., De Raedt, L.: Lifted probabilistic inference by first-order knowledge compilation. In: Proceedings of 22nd IJCAI Conference, IJCAI/AAAI, pp. 2178–2185 (2011)
Darroch, J.N., Ratcliff, D.: Generalized iterative scaling for log-linear models. Ann. Math. Stat. 43, 1470–1480 (1972). Institute of Mathematical Statistics
Finthammer, M., Beierle, C.: A two-level approach to maximum entropy model computation for relational probabilistic logic based on weighted conditional impacts. In: Straccia, U., Calì, A. (eds.) SUM 2014. LNCS, vol. 8720, pp. 162–175. Springer, Cham (2014). doi:10.1007/978-3-319-11508-5_14
Finthammer, M., Beierle, C.: Towards a more efficient computation of weighted conditional impacts for relational probabilistic knowledge bases under maximum entropy semantics. In: Hölldobler, S., Krötzsch, M., Peñaloza, R., Rudolph, S. (eds.) KI 2015. LNCS, vol. 9324, pp. 72–86. Springer, Cham (2015). doi:10.1007/978-3-319-24489-1_6
Getoor, L., Taskar, B. (eds.): Introduction to Statistical Relational Learning. MIT Press, Cambridge (2007)
Kern-Isberner, G.: A note on conditional logics and entropy. Int. J. Approx. Reason. 19(3–4), 231–246 (1998)
Kern-Isberner, G.: Conditionals in Nonmonotonic Reasoning and Belief Revision. Springer, Heidelberg (2001)
Kern-Isberner, G., Lukasiewicz, T.: Combining probabilistic logic programming with the power of maximum entropy. Artif. Intell. 157(1–2), 139–202 (2004)
Kern-Isberner, G., Thimm, M.: Novel semantical approaches to relational probabilistic conditionals. In: Proceedings of 12th KR Conference. AAAI Press (2010)
Kimmig, A., Van den Broeck, G., de Raedt, L.: Algebraic model counting. Int. J. Appl. Logic 22(C), 46–62 (2017)
Paris, J.B.: The Uncertain Reasoner’s Companion - A Mathematical Perspective. Cambridge University Press, Cambridge (1994)
Paris, J.B.: Common sense and maximum entropy. Synthese 117(1), 75–93 (1999)
Poole, D.: First-order probabilistic inference. In: Proceedings of 18th IJCAI Conference, pp. 985–991. Morgan Kaufmann (2003)
Richardson, M., Domingos, P.M.: Markov logic networks. Mach. Learn. 62(1–2), 107–136 (2006)
Shore, J.E., Johnson, R.W.: Axiomatic derivation of the principle of maximum entropy and the principle of minimum cross-entropy. IEEE Trans. Inf. Theor. 26(1), 26–37 (1980)
Wilhelm, M., Kern-Isberner, G.: Typed model counting and its application to probabilistic conditional reasoning at maximum entropy. In: Proceedings of 30th FLAIRS Conference, pp. 748–753. AAAI Press (2017)
Acknowledgements
This research was supported by the German National Science Foundation (DFG) Research Unit FOR 1513 on Hybrid Reasoning for Intelligent Systems.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2017 Springer International Publishing AG
About this paper
Cite this paper
Wilhelm, M., Finthammer, M., Kern-Isberner, G., Beierle, C. (2017). First-Order Typed Model Counting for Probabilistic Conditional Reasoning at Maximum Entropy. In: Moral, S., Pivert, O., Sánchez, D., Marín, N. (eds) Scalable Uncertainty Management. SUM 2017. Lecture Notes in Computer Science(), vol 10564. Springer, Cham. https://doi.org/10.1007/978-3-319-67582-4_19
Download citation
DOI: https://doi.org/10.1007/978-3-319-67582-4_19
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-67581-7
Online ISBN: 978-3-319-67582-4
eBook Packages: Computer ScienceComputer Science (R0)