Exploring and Exploiting Decision Boundary Dynamics for Adversarial Robustness

Authors: Yuancheng Xu, Yanchao Sun, Micah Goldblum, Tom Goldstein, Furong Huang

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on the CIFAR-10 and Tiny-Image Net datasets verify that Dy ART alleviates the conflicting dynamics of the decision boundary and obtains improved robustness under various perturbation sizes compared to the state-of-the-art defenses.
Researcher Affiliation Academia University of Maryland, College Park New York University {ycxu,ycs,tomg,furongh}@umd.edu goldblum@nyu.edu
Pseudocode No The paper describes methods and processes in narrative text but does not include structured pseudocode or algorithm blocks.
Open Source Code Yes Our code is available at https://github.com/Yuancheng-Xu/Dynamics-Aware-Robust-Training.
Open Datasets Yes Experiments on the CIFAR-10 (Krizhevsky et al., 2009) and Tiny-Image Net (Deng et al., 2009) datasets
Dataset Splits Yes To alleviate robust overfitting (Rice et al., 2020), we perform early stopping on a validation set of size 1024 using projected gradient descent (PGD) attacks with 20 steps.
Hardware Specification Yes All experiments are run on NVIDIA Ge Force RTX 2080 Ti GPU. ... We use one NVDIA RTX A4000.
Software Dependencies No The paper mentions 'Py Torch' and 'Opacus package (Yousefpour et al., 2021)' but does not provide specific version numbers for these software components.
Experiment Setup Yes Models are trained using stochastic gradient descent with momentum 0.9 and weight decay 0.0005 with batch size 256 for 200 epochs on CIFAR-10 and 100 epochs on Tiny-Image Net. We use a cosine learning rate schedule (Loshchilov & Hutter, 2016) without restarts where the initial learning rate is set to 0.1 for all baselines and Dy ART. To alleviate robust overfitting (Rice et al., 2020), we perform early stopping on a validation set of size 1024 using projected gradient descent (PGD) attacks with 20 steps.