Advbox: a toolbox to generate adversarial examples that fool neural networks

View on arXiv ← Back to list

Authors: Dou Goodman, Hao Xin, Wang Yang, Wu Yuesheng, Xiong Junfeng, Zhang Huan

Published: 2020-01-13 08:11:27+00:00

AI Summary

AdvBox is a toolbox for generating adversarial examples that fool neural networks across multiple deep learning frameworks. It supports various attack scenarios, including face recognition attacks, stealth T-shirt attacks, and deepfake detection, and benchmarks the robustness of machine learning models.

Abstract

In recent years, neural networks have been extensively deployed for computer vision tasks, particularly visual classification problems, where new algorithms reported to achieve or even surpass the human performance. Recent studies have shown that they are all vulnerable to the attack of adversarial examples. Small and often imperceptible perturbations to the input images are sufficient to fool the most powerful neural networks. emph{Advbox} is a toolbox to generate adversarial examples that fool neural networks in PaddlePaddle, PyTorch, Caffe2, MxNet, Keras, TensorFlow and it can benchmark the robustness of machine learning models. Compared to previous work, our platform supports black box attacks on Machine-Learning-as-a-service, as well as more attack scenarios, such as Face Recognition Attack, Stealth T-shirt, and DeepFake Face Detect. The code is licensed under the Apache 2.0 and is openly available at https://github.com/advboxes/AdvBox. Advbox now supports Python 3.

Key findings

AdvBox successfully generates adversarial examples for various attack scenarios including face recognition, object detection, and deepfake detection. The toolbox's versatility allows for comprehensive evaluation of model robustness and facilitates research into adversarial attacks and defenses.

Approach

AdvBox generates adversarial examples using several popular attack algorithms (FGSM, BIM, DeepFool, JSMA, CW, PGD) and provides interfaces to multiple deep learning frameworks. It also includes defense algorithms and robustness evaluation methods, enabling comprehensive testing of model vulnerability.

Datasets

UNKNOWN

Model(s)

FaceNet (for face recognition attacks), and unspecified models for deepfake detection and object detection (stealth T-shirt attacks). The toolbox supports various frameworks, including TensorFlow, PyTorch, PaddlePaddle, Caffe2, MxNet, and Keras.

Author countries

China

← Previous