On the Effects of Fairness to Adversarial Vulnerability

Authors: Cuong Tran, Keyu Zhu, Pascal Van Hentenryck, Ferdinando Fioretto

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on non-linear models and different architectures validate the theoretical findings.This section empirically validates the theoretical insights discussed earlier, extending them to more complex architectures, datasets, and loss functions.
Researcher Affiliation Academia Cuong Tran1 , Keyu Zhu2 , Pascal Van Hentenryck2 and Ferdinando Fioretto1 1University of Virginia 2Georgia Institute of Technology
Pseudocode No The paper describes methods in prose but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement or link for open-sourcing the code for the described methodology.
Open Datasets Yes Datasets. The experiments of this section focus on three vision datasets: UTK-Face [Zhang et al., 2017], FMNIST [Xiao et al., 2017] and CIFAR-10 [Krizhevsky et al., 2009].
Dataset Splits No The paper mentions using UTK-Face, FMNIST, and CIFAR-10 datasets and refers to 'standard labels' for some, but does not provide specific percentages or counts for train/validation/test splits.
Hardware Specification No The paper does not provide any specific hardware details (e.g., GPU or CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions specific software like 'Torchattacks' by reference but does not provide specific version numbers for software dependencies used in their own experimental setup.
Experiment Setup Yes Models trained on the UTK-Face data use a learning rate of 1e 3 and 70 epochs. Those trained on FMNIST and CIFAR, use a learning rate of 1e 1 and 200 epochs, as suggested in previous work [Xu et al., 2021a].