Chordal sparsity for SDP-based neural network verification

被引:0
|
作者
Xue, Anton [1 ]
Lindemann, Lars [2 ]
Alur, Rajeev [1 ]
机构
[1] Univ Penn, Dept Comp & Informat Sci, Philadelphia, PA 19104 USA
[2] Univ Southern Calif, Thomas Lord Dept Comp Sci, Los Angeles, CA USA
基金
美国国家科学基金会;
关键词
Neural networks; Convex optimization; Safety verification; SAFETY;
D O I
10.1016/j.automatica.2023.111487
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Neural networks are central to many emerging technologies, but verifying their correctness remains a major challenge. It is known that network outputs can be sensitive and fragile to even small input perturbations, thereby increasing the risk of unpredictable and undesirable behavior. Fast and accurate verification of neural networks is therefore critical to their widespread adoption, and in recent years various methods have been developed as a response to this problem. In this paper, we focus on improving semidefinite programming (SDP) based techniques for neural network verification. Such techniques offer the power of expressing complex geometric constraints while retaining a convex problem formulation, but scalability remains a major issue in practice. Our starting point is the DeepSDP framework proposed by Fazlyab et al., which uses quadratic constraints to abstract the verification problem into a large-scale SDP. However, solving this SDP quickly becomes intractable when the network grows. Our key observation is that by leveraging chordal sparsity, we can decompose the primary computational bottleneck of DeepSDP - a large linear matrix inequality (LMI) - into an equivalent collection of smaller LMIs. We call our chordally sparse optimization program Chordal-DeepSDP and prove that its construction is identically expressive as that of DeepSDP. Moreover, we show that additional analysis of Chordal-DeepSDP allows us to further rewrite its collection of LMIs in a second level of decomposition that we call Chordal-DeepSDP-2 - which results in another significant computational gain. Finally, we provide numerical experiments on real networks of learned cart-pole dynamics, showcasing the computational advantage of Chordal-DeepSDP and Chordal-DeepSDP-2 over DeepSDP. (c) 2023 Elsevier Ltd. All rights reserved.
引用
收藏
页数:9
相关论文
共 50 条
  • [21] Computational Experience with a SDP-Based Algorithm for Maximum Cut with Limited Unbalance
    Galbiati, Giulia
    Gualandi, Stefano
    Maffioli, Francesco
    NETWORKS, 2010, 55 (03) : 247 - 255
  • [22] On Completeness of SDP-Based Barrier Certificate Synthesis over Unbounded Domains
    Wu, Hao
    Feng, Shenghua
    Gang, Ting
    Wang, Jie
    Xia, Bican
    Zhan, Naijun
    FORMAL METHODS, PT II, FM 2024, 2025, 14934 : 248 - 266
  • [23] PEP analysis of SDP-based non-coherent signal detection
    Stojnic, Mihailo
    Hassibi, Babak
    Vikalo, Haris
    2007 IEEE INTERNATIONAL SYMPOSIUM ON INFORMATION THEORY PROCEEDINGS, VOLS 1-7, 2007, : 2916 - 2920
  • [24] Chordal Sparsity for Lipschitz Constant Estimation of Deep Neural Networks
    Xue, Anton
    Lindemann, Lars
    Robey, Alexander
    Hassani, Hamed
    Pappas, George J.
    Alur, Rajeev
    2022 IEEE 61ST CONFERENCE ON DECISION AND CONTROL (CDC), 2022, : 3389 - 3396
  • [25] SDP-based approximation of stabilising solutions for periodic matrix Riccati differential equations
    Gusev, Sergei V.
    Shiriaev, Anton S.
    Freidovich, Leonid B.
    INTERNATIONAL JOURNAL OF CONTROL, 2016, 89 (07) : 1396 - 1405
  • [26] Hidden in Plain Sight. SDP-Based Covert Channel for Botnet Communication
    Tsiatsikas, Zisis
    Anagnostopoulos, Marios
    Kambourakis, Georgios
    Lambrou, Sozon
    Geneiatakis, Dimitris
    TRUST, PRIVACY AND SECURITY IN DIGITAL BUSINESS, 2015, 9264 : 48 - 59
  • [27] SDP-based Joint Sensor and Controller Design for Information-regularized Optimal LQG Control
    Tanaka, Takashi
    Sandberg, Henrik
    2015 54TH IEEE CONFERENCE ON DECISION AND CONTROL (CDC), 2015, : 4486 - 4491
  • [28] SDP-based branch-and-bound for non-convex quadratic integer optimization
    Buchheim, Christoph
    Montenegro, Maribel
    Wiegele, Angelika
    JOURNAL OF GLOBAL OPTIMIZATION, 2019, 73 (03) : 485 - 514
  • [29] SDP-Based Optimal Power Flow With Steady-State Voltage Stability Constraints
    Wang, Chong
    Cui, Bai
    Wang, Zhaoyu
    Gu, Chenghong
    IEEE TRANSACTIONS ON SMART GRID, 2019, 10 (04) : 4637 - 4647
  • [30] Deep Neural Network Acceleration Method Based on Sparsity
    He, Ming
    Zhao, Haiwu
    Wang, Guozhong
    Chen, Yu
    Zhu, Linlin
    Gao, Yuan
    DIGITAL TV AND MULTIMEDIA COMMUNICATION, 2019, 1009 : 133 - 145