AdvAD: Exploring Non-Parametric Diffusion for Imperceptible Adversarial Attacks

Authors: Jin Li, Ziqiang He, Anwei Luo, Jian-Fang Hu, Z. Jane Wang, Xiangui Kang

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate the effectiveness of the proposed Adv AD and Adv AD-X. Compared with state-of-the-art imperceptible attacks, Adv AD achieves an average of 99.9% (+17.3%) ASR with 1.34 (-0.97) l2 distance, 49.74 (+4.76) PSNR and 0.9971 (+0.0043) SSIM against four prevalent DNNs with three different architectures on the Image Net-compatible dataset. Code is available at https://github.com/Xiangui Kang/Adv AD.
Researcher Affiliation Academia 1Guangdong Key Lab of Information Security, School of Computer Science and Engineering, Sun Yat-Sen University 2Electrical and Computer Engineering Dept, University of British Columbia
Pseudocode Yes Algorithm 1 Adv AD
Open Source Code Yes Code is available at https://github.com/Xiangui Kang/Adv AD.
Open Datasets Yes In line with prior studies [15, 19, 32, 40], our experiments are conducted on the Image Net-compatible Dataset 1, containing 1,000 images of Image Net [41] classes with size of 299 299, and the images are resized to standard input size of 224 224 in all experiments. 1https://github.com/cleverhans-lab/cleverhans/tree/master/cleverhans_v3.1.0/examples/nips17_ adversarial_competition/dataset
Dataset Splits No The paper mentions using an 'Image Net-compatible Dataset' and evaluating on 'test' data, but it does not provide specific details on training, validation, and test dataset splits, or refer to any standard predefined splits with explicit percentages or sample counts for validation.
Hardware Specification Yes The reported running times are obtained using a RTX 3090 GPU on a same machine.
Software Dependencies No The paper mentions 'Python environment requirements' as part of its open-source code documentation, but it does not specify concrete versions for software dependencies or libraries (e.g., Python, PyTorch, TensorFlow, CUDA versions).
Experiment Setup Yes For our proposed Adv AD and Adv AD-X, we set ΞΎ = 8/255 and T = 1000 for all experiments unless specifically mentioned.