On the Effectiveness of Low Frequency Perturbations

Authors: Yash Sharma, Gavin Weiguang Ding, Marcus A. Brubaker

IJCAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental By systematically controlling the frequency components of the perturbation, evaluating against the top-placing defense submissions in the Neur IPS 2017 competition, we empirically show that performance improvements in both the whitebox and black-box transfer settings are yielded only when low frequency components are preserved.
Researcher Affiliation Industry Yash Sharma , Gavin Weiguang Ding and Marcus A. Brubaker Borealis AI
Pseudocode No No structured pseudocode or algorithm blocks were found in the paper.
Open Source Code No The paper refers to supplementary material via an arXiv link to its own appendix but does not contain an explicit statement about releasing code for the described methodology or a direct link to a code repository.
Open Datasets Yes Testing against state-of-the-art Image Net [Deng et al., 2009] defense methods... To evaluate the effectiveness of perturbations under different frequency constraints, we test against models and defenses from the Neur IPS 2017 Adversarial Attacks and Defences Competition [Kurakin et al., 2018]... We also show that these results do not transfer to the significantly lower-dimensional CIFAR-10 dataset
Dataset Splits No The paper evaluates attacks on pre-trained models and mentions using 1000 test examples, but does not specify the train/validation/test splits for the datasets these models were originally trained on, as the authors did not perform the original training.
Hardware Specification No The paper does not provide specific details regarding the hardware used for running the experiments, such as GPU or CPU models.
Software Dependencies No The paper does not provide specific versions for software dependencies, such as programming languages or libraries, used in the experiments.
Experiment Setup Yes We test on ϵ = 16/255 (competition distortion bound) and iterations = [1, 10] for the non-targeted case; ϵ = 32/255 and iterations = 10 for the targeted case. ... For each mask type, we test n = [256, 128, 64, 32] with d = 299. For DCT Rand, we average results over 3 random seeds.