Abstract—In recent years, with the development of deep
learning technology, neural networks play an increasingly
important role in more and more fields. However, research
shows that neural networks are vulnerable to the attack of
adversarial examples. The purpose of this paper is to study the
principle of adversarial examples generation and propose a new
method of generating adversarial examples. Compared with
existed methods, our method achieves better deception rate and
perturbs less pixels of images. During an epoch in batch
dimension iteration, multiple pixels are perturbed while
Manhattan-Distance constraints are added to them. Our
algorithm performs well in experiments. Compared with
Carlini-Wagner method, only 60 more dimensions are
perturbed, which indicates that the computation cost of our
algorithm is completely acceptable. Besides, compared with
FGSM algorithm, the deception rate increases by 12% while the
generation times of them are almost same.
Index Terms—Adversarial attacks, deep learning,
adversarial examples, distance constraints.
D. Liu is with the Key Laboratory of Measurement and Control of CSE,
Ministry of Education, Southeast University, Nanjing 210096, China (e-mail:
11973053@qq.com).
Z. Song is with the College of Electronic and Optical Engineering,
Nanjing University of Posts and Telecommunications, Nanjing 210023,
China (e-mail: szian@hotmail.com).
S. Ren is with the Oversea Education College, Nanjing University of
Posts and Telecommunications, Nanjing 210023, China (e-mail:
Rensy1121@outlook.com).
S. Xia is with the Key Laboratory of Measurement and Control of CSE,
Ministry of Education, Southeast University, Nanjing 210096, China, and
also with the School of Automation, Southeast University, Nanjing 210096,
China (Corresponding author; e-mail: xia081@gmail.com).
Cite: Dahui Liu, Zihan Song, Siyu Ren, and Siyu Xia, "Adversarial Attacks on Neural Network with Batch Dimensions Perturbation and Manhattan-Distance Constraints," International Journal of Machine Learning and Computing vol. 12, no. 1, pp. 17-22, 2022.
Copyright © 2022 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).