MagNet and "Efficient Defenses Against Adversarial Attacks" are Not Robust to Adversarial Examples

MagNet and "Efficient Defenses..." were recently proposed as a defense to adversarial examples. We find that we can construct adversarial examples that defeat these defenses with only a slight increase in distortion.

Health & medicine Security & robustness cs.LG cs.AI cs.CR Adversarial Robustness in Machine Learning Forensic and Genetic Research Ion-surface interactions and analysis Computer Science Mathematics

View graph Source

Paper

References (12)

01Efficient Defenses Against Adversarial AttacksValentina Zantedeschi, Maria-Irina Nicolae, Ambrish Rawat2017 · AISec@CCS · 316 citations In Library

02APE-GAN: Adversarial Perturbation Elimination with GANGuoqing Jin, Shiwei Shen, Dongming Zhang et al.2017 · IEEE International Conference on Acoustics, Speech, and Signal Processing · 247 citations In Library

03Towards Deep Learning Models Resistant to Adversarial AttacksA. Ma̧dry, Aleksandar Makelov, Ludwig Schmidt et al.2017 · International Conference on Learning Representations · 15k citations In Library

04MagNet: A Two-Pronged Defense against Adversarial ExamplesDongyu Meng, Hao Chen2017 · Conference on Computer and Communications Security · 1.3k citations In Library

05Delving into Transferable Adversarial Examples and Black-box AttacksYan-Po Liu, Xinyun Chen, Chang Liu et al.2016 · International Conference on Learning Representations · 2.0k citations In Library

06Towards Evaluating the Robustness of Neural NetworksNicholas Carlini, D. Wagner2016 · IEEE Symposium on Security and Privacy · 10k citations In Library

07Explaining and Harnessing Adversarial ExamplesI. Goodfellow, Jonathon Shlens, Christian Szegedy2014 · International Conference on Learning Representations · 23k citations In Library

08Intriguing properties of neural networksChristian Szegedy, Wojciech Zaremba, I. Sutskever et al.2013 · International Conference on Learning Representations · 17k citations In Library

09Rectified Linear Units Improve Restricted Boltzmann MachinesVinod Nair, Geoffrey E. Hinton2010 · International Conference on Machine Learning · 19k citations

10GENERATIVE ADVERSARIAL NETSIndividualized Treat, Jinsung Yoon2018 · 43k citations

11Learning Multiple Layers of Features from Tiny ImagesA. Krizhevsky2009 · 43k citations

12Gradient-based learning applied to document recognitionYann LeCun, L. Bottou, Yoshua Bengio et al.1998 · Proceedings of the IEEE · 62k citations

Similar papers

On the Limitation of MagNet Defense Against L1-Based Adversarial ExamplesPei-Hsuan Lu, Pin‐Yu Chen, Kang-Cheng Chen et al.2018 In Library

On the Limitation of MagNet Defense against $L_1$-based Adversarial ExamplesPei-Hsuan Lu, Pin‐Yu Chen, Kangcheng Chen et al.2018 · arXiv (Cornell University) · 21 citations In Library

MagNet: a Two-Pronged Defense against Adversarial ExamplesDongyu Meng, Hao Chen2017 · Conference on Computer and Communications Security · 1.3k citations In Library

MagNet with Randomization Defense against Adversarial ExamplesFangwei Wang, Guofang Chai, Qingru Li et al.2020 · International Conference on Networking and Network Applications · 3 citations In Library

PosterLeo Hyun Park, Jaewoo Park, Soochang Chung et al.2022 · Proceedings of the 2022 ACM SIGSAC Conference on Computer and Communications Security · 3 citations In Library

MagNetDongyu Meng, Hao Chen2017 In Library

Provably Minimally-Distorted Adversarial ExamplesNicholas Carlini, Guy Katz, Clark Barrett et al.2017 · arXiv (Cornell University) · 91 citations In Library

Ensemble MagNet:A fast training network against adversarial examplesZebin Zhuang2023 · International Conference on Algorithms, Microchips and Network Applications · 0 citations In Library

Adversarial Examples Are Not Easily DetectedNicholas Carlini, David Wagner2017 In Library

Markov Chain Based Efficient Defense Against Adversarial Examples in Computer VisionYue Zhou, Xiaofang Hu, Lidan Wang et al.2018 · IEEE Access · 6 citations In Library

Patents near this paper

Automatic Machine Learning Vulnerability Identification and RetrainingUS 11,818,163 · PAYPAL, INC. · 2023 Using multimodal model consistency to detect adversarial attacksUS 11,675,896 · INTERNATIONAL BUSINESS MACHINES CORPORATION · 2023 METHOD AND MACHINE LEARNING SYSTEM FOR DETECTING ADVERSARIAL EXAMPLESUS 11,501,206 · NXP B.V. · 2022 Systems and Methods for Improved Adversarial Training of Machine-Learned ModelsUS 11,494,667 · GOOGLE LLC · 2022 METHOD AND DATA PROCESSING SYSTEM FOR MAKING MACHINE LEARNING MODEL MORE RESISTENT TO ADVERSARIAL EXAMPLESUS 11,410,078 · NXP B.V. · 2022

Source of record

arXiv