Transferable Adversarial Attack based on Integrated Gradients

Authors: Yi Huang, Adams Wai-Kin Kong

ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate that TAIG outperforms the state-of-the-art methods.
Researcher Affiliation Academia Yi Huang, Adams Wai-Kin Kong School of Computer Science and Engineering Nanyang Technological University 50 Nanyang Avenue, Singapore 639798 {yi.huangy,adamskong}@ntu.edu.sg
Pseudocode No The paper describes the algorithm using equations and narrative, but does not include a formally structured pseudocode block or an 'Algorithm' label.
Open Source Code Yes The code will available at https://github.com/yihuang2016/TAIG.
Open Datasets Yes The experiments are conducted on Image Net (Russakovsky et al., 2015) validation set.
Dataset Splits No The paper uses pre-trained models and the ImageNet validation set as source data for generating and evaluating adversarial examples, but it does not specify train/validation/test splits for its own experimental setup or attack algorithm training.
Hardware Specification Yes All the experiments are performed on two NVIDIA Ge Force RTX 3090 with the main code implemented using Py Torch.
Software Dependencies No The paper mentions 'Py Torch' but does not specify a version number or other software dependencies with specific versions.
Experiment Setup Yes Res Net50 (He et al., 2016) is selected as a surrogate model to generate adversarial examples to compare with the state-of-the-art methods. ... By following Lin BP, the maximum allowable perturbations are set as ε = 0.03, 0.05, 0.1. ... Thirty sampling points are used to estimate TAIG-S. For TAIG-R, the number of turning point E is set to 30 and τ is set equal to ε. ... for Lin BP, we keep the default setting where the number of iterations is 300 and the step size is 1/255 for all different ε. ... TAIG-S and TAIG-R are run 20, 50, and 100 iterations with the same step size as Lin BP for ε = 0.03, 0.05, 0.1, respectively.