Dynamic Slicing for Deep Neural Networks

被引:19
|
作者
Zhang, Ziqi [1 ]
Li, Yuanchun [2 ]
Guo, Yao [1 ]
Chen, Xiangqun [1 ]
Liu, Yunxin [2 ]
机构
[1] Peking Univ, Dept Comp Sci, Key Lab High Confidence Software Technol MOE, Beijing, Peoples R China
[2] Microsoft Res, Beijing, Peoples R China
基金
中国国家自然科学基金;
关键词
Program slicing; deep neural networks; dynamic slicing; data flow analysis;
D O I
10.1145/3368089.3409676
中图分类号
TP31 [计算机软件];
学科分类号
081202 ; 0835 ;
摘要
Program slicing has been widely applied in a variety of software engineering tasks. However, existing program slicing techniques only deal with traditional programs that are constructed with instructions and variables, rather than neural networks that are composed of neurons and synapses. In this paper, we propose NNSlicer, the first approach for slicing deep neural networks based on data flow analysis. Our method understands the reaction of each neuron to an input based on the difference between its behavior activated by the input and the average behavior over the whole dataset. Then we quantify the neuron contributions to the slicing criterion by recursively backtracking from the output neurons, and calculate the slice as the neurons and the synapses with larger contributions. We demonstrate the usefulness and effectiveness of NNSlicer with three applications, including adversarial input detection, model pruning, and selective model protection. In all applications, NNSlicer significantly outperforms other baselines that do not rely on data flow analysis.
引用
收藏
页码:838 / 850
页数:13
相关论文
共 50 条
  • [31] Cell dynamic morphology classification using deep convolutional neural networks
    Li, Heng
    Pang, Fengqian
    Shi, Yonggang
    Liu, Zhiwen
    [J]. CYTOMETRY PART A, 2018, 93A (06) : 628 - 638
  • [32] Accelerating the characterization of dynamic DNA origami devices with deep neural networks
    Yuchen Wang
    Xin Jin
    Carlos Castro
    [J]. Scientific Reports, 13
  • [33] DISCO: Dynamic and Invariant Sensitive Channel Obfuscation for deep neural networks
    Singh, Abhishek
    Chopra, Ayush
    Garza, Ethan
    Zhang, Emily
    Vepakomma, Praneeth
    Sharma, Vivek
    Raskar, Ramesh
    [J]. 2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021, 2021, : 12120 - 12130
  • [34] SuperNeurons: Dynamic GPU Memory Management for Training Deep Neural Networks
    [J]. 2018, Association for Computing Machinery, 2 Penn Plaza, Suite 701, New York, NY 10121-0701, United States (53):
  • [35] SuperNeurons: Dynamic GPU Memory Management for Training Deep Neural Networks
    Wang, Linnan
    Ye, Jinmian
    Zhao, Yiyang
    Wu, Wei
    Li, Ang
    Song, Shuaiwen Leon
    Xu, Zenglin
    Kraska, Tim
    [J]. ACM SIGPLAN NOTICES, 2018, 53 (01) : 41 - 53
  • [36] High dynamic range ultrasound beamforming using deep neural networks
    Luchies, Adam
    Byram, Brett
    [J]. MEDICAL IMAGING 2019: ULTRASONIC IMAGING AND TOMOGRAPHY, 2019, 10955
  • [37] A Dynamic Pruning Method on Multiple Sparse Structures in Deep Neural Networks
    Hu, Jie
    Lin, Peng
    Zhang, Huajun
    Lan, Zining
    Chen, Wenxin
    Xie, Kailiang
    Chen, Siyun
    Wang, Hao
    Chang, Sheng
    [J]. IEEE ACCESS, 2023, 11 : 38448 - 38457
  • [38] Transferable Dynamic Molecular Charge Assignment Using Deep Neural Networks
    Nebgen, Benjamin
    Lubbers, Nicholas
    Smith, Justin S.
    Sifain, Andrew E.
    Lokhov, Andrey
    Isayev, Olexandr
    Roitberg, Adrian E.
    Barros, Kipton
    Tretiak, Sergei
    [J]. JOURNAL OF CHEMICAL THEORY AND COMPUTATION, 2018, 14 (09) : 4687 - 4698
  • [39] Dynamic Energy Optimization in Chip Multiprocessors Using Deep Neural Networks
    Moghaddam, Milad Ghorbani
    Guan, Wenkai
    Ababei, Cristinel
    [J]. IEEE TRANSACTIONS ON MULTI-SCALE COMPUTING SYSTEMS, 2018, 4 (04): : 649 - 661
  • [40] Accelerating the characterization of dynamic DNA origami devices with deep neural networks
    Wang, Yuchen
    Jin, Xin
    Castro, Carlos
    [J]. SCIENTIFIC REPORTS, 2023, 13 (01)