Adversarial Training and Provable Robustness: A Tale of Two Objectives

Authors: Jiameng Fan, Wenchao Li7367-7376

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We perform both theoretical analysis on the convergence of the proposed technique and experimental comparison with state-of-the-arts. Results on MNIST and CIFAR-10 show that our method can consistently match or outperform prior approaches for provable l robustness.
Researcher Affiliation Academia Jiameng Fan , Wenchao Li Department of Electrical and Computer Engineering, Boston University, Boston {jmfan, wenchao}@bu.edu
Pseudocode Yes Algorithm 1 Weight Updates and Algorithm 2 Joint Training
Open Source Code No The paper does not contain an explicit statement about releasing source code or a link to a code repository for the methodology described.
Open Datasets Yes Results on MNIST and CIFAR-10 show that our method can consistently match or outperform prior approaches for provable l robustness.
Dataset Splits No The paper mentions 'test dataset' and 'test examples' but does not provide specific training/validation/test dataset splits, percentages, or explicit sample counts for reproduction.
Hardware Specification Yes We perform all experiments on a desktop server using at most 4 Ge Force GTX 1080 Ti GPUs.
Software Dependencies No The paper does not specify software dependencies with version numbers.
Experiment Setup Yes Algorithm 2 Joint Training Input Warm-up epochs Tnat and Tadv, ϵtrain ramp-up epochs R, maximum FOSC value cmax... ct=clip(cmax (t R) cmax/T , 0, cmax)... κadv, κIBP, κreg=COMPUTE WEIGHTS(xadv, t, ct)... loss=κadv Ladv(θt)+κIBPLIBP(θt)+κreg LIBP(θt) 2 2... θt+1=θt ηtgfinal(θt) gfinal(θt): stochastic gradient