Classified by Research TopicSorted by DateClassified by Publication Type

Efficient Certified Reasoning for Binarized Neural Networks

Efficient Certified Reasoning for Binarized Neural Networks.
Jiong Yang, Yong Kiam Tan, Mate Soos, Magnus O. Myreen and Kuldeep S. Meel.
In Proceedings of the International Conference on Theory and Applications of Satisfiability Testing (SAT), pp. 32:1–32:22, 2025.

Download

[PDF] 

Abstract

Neural networks have emerged as essential components in safety-critical applications - these use cases demand complex, yet trustworthy computations. Binarized Neural Networks (BNNs) are a type of neural network where each neuron is constrained to a Boolean value; they are particularly well-suited for safety-critical tasks because they retain much of the computational capacities of full-scale (floating-point or quantized) deep neural networks, but remain compatible with satisfiability solvers for qualitative verification and with model counters for quantitative reasoning. However, existing methods for BNN analysis suffer from either limited scalability or susceptibility to soundness errors, which hinders their applicability in real-world scenarios. In this work, we present a scalable and trustworthy approach for both qualitative and quantitative verification of BNNs. Our approach introduces a native representation of BNN constraints in a custom-designed solver for qualitative reasoning, and in an approximate model counter for quantitative reasoning. We further develop specialized proof generation and checking pipelines with native support for BNN constraint reasoning, ensuring trustworthiness for all of our verification results. Empirical evaluations on a BNN robustness verification benchmark suite demonstrate that our certified solving approach achieves a 9x speedup over prior certified CNF and PB-based approaches, and our certified counting approach achieves a 218x speedup over the existing CNF-based baseline. In terms of coverage, our pipeline produces fully certified results for 99% and 86% of the qualitative and quantitative reasoning queries on BNNs, respectively. This is in sharp contrast to the best existing baselines which can fully certify only 62% and 4% of the queries, respectively.

BibTeX

@inproceedings{YTSMM25,
  title={Efficient Certified Reasoning for Binarized Neural Networks},
  author={
    Yang, Jiong and Tan, Yong Kiam and Soos, Mate and Myreen, Magnus O. and
    Meel, Kuldeep S.
  },
  booktitle=SAT,
  pages={32:1--32:22},
  year={2025},
  bib2html_rescat={Formal Methods 4 ML},
  bib2html_pubtype={Refereed Conference},
  bib2html_dl_pdf=
    {https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.SAT.2025.32},
  abstract={
    Neural networks have emerged as essential components in safety-critical
    applications - these use cases demand complex, yet trustworthy computations.
    Binarized Neural Networks (BNNs) are a type of neural network where each
    neuron is constrained to a Boolean value; they are particularly well-suited
    for safety-critical tasks because they retain much of the computational
    capacities of full-scale (floating-point or quantized) deep neural networks,
    but remain compatible with satisfiability solvers for qualitative
    verification and with model counters for quantitative reasoning. However,
    existing methods for BNN analysis suffer from either limited scalability or
    susceptibility to soundness errors, which hinders their applicability in
    real-world scenarios. In this work, we present a scalable and trustworthy
    approach for both qualitative and quantitative verification of BNNs. Our
    approach introduces a native representation of BNN constraints in a
    custom-designed solver for qualitative reasoning, and in an approximate
    model counter for quantitative reasoning. We further develop specialized
    proof generation and checking pipelines with native support for BNN
    constraint reasoning, ensuring trustworthiness for all of our verification
    results. Empirical evaluations on a BNN robustness verification benchmark
    suite demonstrate that our certified solving approach achieves a 9x speedup
    over prior certified CNF and PB-based approaches, and our certified counting
    approach achieves a 218x speedup over the existing CNF-based baseline. In
    terms of coverage, our pipeline produces fully certified results for 99% and
    86% of the qualitative and quantitative reasoning queries on BNNs,
    respectively. This is in sharp contrast to the best existing baselines which
    can fully certify only 62% and 4% of the queries, respectively.
  },
}

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