Provable Preimage Under-Approximation for Neural Networks

被引:3
|
作者
Zhang, Xiyue [1 ]
Wang, Benjie [1 ]
Kwiatkowska, Marta [1 ]
机构
[1] Univ Oxford, Dept Comp Sci, Oxford, England
关键词
VERIFICATION;
D O I
10.1007/978-3-031-57256-2_1
中图分类号
TP31 [计算机软件];
学科分类号
081202 ; 0835 ;
摘要
Neural network verification mainly focuses on local robustness properties, which can be checked by bounding the image (set of outputs) of a given input set. However, often it is important to know whether a given property holds globally for the input domain, and if not then for what proportion of the input the property is true. To analyze such properties requires computing preimage abstractions of neural networks. In this work, we propose an efficient anytime algorithm for generating symbolic under-approximations of the preimage of any polyhedron output set for neural networks. Our algorithm combines a novel technique for cheaply computing polytope preimage under-approximations using linear relaxation, with a carefully-designed refinement procedure that iteratively partitions the input region into subregions using input and ReLU splitting in order to improve the approximation. Empirically, we validate the efficacy of our method across a range of domains, including a high-dimensional MNIST classification task beyond the reach of existing preimage computation methods. Finally, as use cases, we showcase the application to quantitative verification and robustness analysis. We present a sound and complete algorithm for the former, which exploits our disjoint union of polytopes representation to provide formal guarantees. For the latter, we find that our method can provide useful quantitative information even when standard verifiers cannot verify a robustness property.
引用
收藏
页码:3 / 23
页数:21
相关论文
共 50 条
  • [1] Under-approximation of Reachability in Multivalued Asynchronous Networks
    Folschette, Maxime
    Pauleve, Loie
    Magnin, Morgan
    Roux, Olivier
    ELECTRONIC NOTES IN THEORETICAL COMPUTER SCIENCE, 2013, 299 : 33 - 51
  • [2] Provable approximation properties for deep neural networks
    Shaham, Uri
    Cloninger, Alexander
    Coifman, Ronald R.
    APPLIED AND COMPUTATIONAL HARMONIC ANALYSIS, 2018, 44 (03) : 537 - 557
  • [3] PREDICATE ABSTRACTION WITH UNDER-APPROXIMATION REFINEMENT
    Pasareanu, Corina S.
    Pelanek, Radek
    Visser, Willem
    LOGICAL METHODS IN COMPUTER SCIENCE, 2007, 3 (01)
  • [4] Under-Approximation Refinement for Classical Planning
    Heusner, Manuel
    Wehrle, Martin
    Pommerening, Florian
    Helmert, Malte
    TWENTY-FOURTH INTERNATIONAL CONFERENCE ON AUTOMATED PLANNING AND SCHEDULING, 2014, : 365 - 369
  • [5] UTOPIC: Under-Approximation Through Optimal Control
    Doncel, Josu
    Gast, Nicolas
    Tribastone, Mirco
    Tschaikowski, Max
    Vandin, Andrea
    QUANTITATIVE EVALUATION OF SYSTEMS (QEST 2019), 2019, 11785 : 277 - 291
  • [6] Better under-approximation of programs by hiding variables
    Ball, Thomas
    Kupferman, Orna
    VERIFICATION, MODEL CHECKING, AND ABSTRACT INTERPRETATION, PROCEEDINGS, 2007, 4349 : 314 - +
  • [7] Under-Approximation for Scalable Bug Detection (Keynote)
    Raad, Azalea
    PROCEEDINGS OF THE 13TH ACM SIGPLAN INTERNATIONAL CONFERENCE ON CERTIFIED PROGRAMS AND PROOFS, CPP 2024, 2024, : 1 - 1
  • [8] Provable Second Preimage Resistance Revisited
    Bouillaguet, Charles
    Vayssiere, Bastien
    SELECTED AREAS IN CRYPTOGRAPHY - SAC 2013, 2014, 8282 : 513 - 532
  • [9] Effective Bit-Width and Under-Approximation
    Brummayer, Robert
    Biere, Armin
    COMPUTER AIDED SYSTEMS THEORY - EUROCAST 2009, 2009, 5717 : 304 - 311
  • [10] Under-Approximation Generation Driven by Relevance Predicates and Variants
    Julliand, J.
    Kouchnarenko, O.
    Masson, P-A
    Voiron, G.
    TESTS AND PROOFS, TAP 2018, 2018, 10889 : 63 - 82