Classified by Research TopicSorted by DateClassified by Publication Type

Engineering an Efficient Probabilistic Exact Model Counter

Engineering an Efficient Probabilistic Exact Model Counter.
Mate Soos, and Kuldeep S. Meel.
In Proceedings of International Conference on Computer-Aided Verification (CAV), pp. 72–91, 2025.

Download

[PDF] 

Abstract

Given a formula F , the problem of model counting, also known as #SAT, is to compute the number of satisfying assignments of F . While model counting has emerged as a crucial primitive in diverse domains from quantitative information flow analysis to neural network verification, scalability remains a fundamental challenge despite advances in both exact and approximate counting techniques. We present $$\textsfGanak2$$ Ganak 2 , a novel framework that achieves substantial performance improvements through three key technical innovations: (1) refined residual formula processing incorporating SAT-specific techniques while maintaining seamless state transitions, (2) dual independent set framework maintaining distinct SAT-eligibility and decision sets, and (3) chronological backtracking specifically adapted to model counting. Our empirical evaluation on 1600 previous model counting competition instances demonstrates that $$\textsfGanak2$$ Ganak 2 successfully computes counts for 1121 instances within the one hour time limit, compared to 1032 instances by the prior state of the art approach, representing an 8.7% improvement. This progress is especially remarkable considering the extensive development and refinement of model counting tools over the years, driven by yearly competitive evaluation in the field.

BibTeX

@inproceedings{SM25a,
  title={Engineering an Efficient Probabilistic Exact Model Counter},
  author={Soos, Mate and Meel, Kuldeep S.},
  booktitle=CAV,
  pages={72--91},
  year={2025},
  bib2html_rescat={Counting, Solver Engineering},
  bib2html_pubtype={Refereed Conference},
  bib2html_dl_pdf=
    {https://link.springer.com/chapter/10.1007/978-3-031-98682-6_5},
  abstract={
    Given a formula F , the problem of model counting, also known as #SAT, is to
    compute the number of satisfying assignments of F . While model counting has
    emerged as a crucial primitive in diverse domains from quantitative
    information flow analysis to neural network verification, scalability
    remains a fundamental challenge despite advances in both exact and
    approximate counting techniques. We present $$\textsf{Ganak2}$$ Ganak 2 , a
    novel framework that achieves substantial performance improvements through
    three key technical innovations: (1) refined residual formula processing
    incorporating SAT-specific techniques while maintaining seamless state
    transitions, (2) dual independent set framework maintaining distinct
    SAT-eligibility and decision sets, and (3) chronological backtracking
    specifically adapted to model counting. Our empirical evaluation on 1600
    previous model counting competition instances demonstrates that
    $$\textsf{Ganak2}$$ Ganak 2 successfully computes counts for 1121 instances
    within the one hour time limit, compared to 1032 instances by the prior
    state of the art approach, representing an 8.7% improvement. This progress
    is especially remarkable considering the extensive development and
    refinement of model counting tools over the years, driven by yearly
    competitive evaluation in the field.
  },
}

Generated by bib2html.pl (written by Patrick Riley with layout from Sanjit A. Seshia ) on Tue Apr 28, 2026 01:27:21