Generating Distributional Adversarial Examples to Evade Statistical Detectors

Authors: Yigitcan Kaya, Muhammad Bilal Zafar, Sergul Aydore, Nathalie Rauschmayr, Krishnaram Kenthapadi

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our attack on image classification tasks with CNNs, where adversarial attacks are most well-studied. Our main focus in this section is evaluating SIA against DDs. We experiment on two popular datasets: CIFAR-10 (Krizhevsky et al., 2009) and Tiny-Image Net (Deng et al., 2009).
Researcher Affiliation Collaboration 1University of Maryland College Park 2Amazon Web Services 3Fiddler AI.
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not include an unambiguous statement about releasing its own source code for the described methodology, nor does it provide a direct link to a code repository.
Open Datasets Yes We experiment on two popular datasets: CIFAR-10 (Krizhevsky et al., 2009) and Tiny-Image Net (Deng et al., 2009).
Dataset Splits No The paper mentions a 'training set' and 'holdout set' but does not explicitly provide details about a distinct 'validation set' split or its size for the models trained.
Hardware Specification No The paper mentions using a 'commodity GPU' but does not specify the exact GPU model, CPU, or other detailed hardware specifications used for running experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., Python, PyTorch/TensorFlow, CUDA versions) used for its implementation.
Experiment Setup Yes Unless specified otherwise, we set ε = 0.03, following prior work (Madry et al., 2018). We perform 200 PGD iterations, use a scheduler that periodically reduces the step size, and craft 250 AEs at once as a batch.