Improving Ensemble Robustness by Collaboratively Promoting and Demoting Adversarial Robustness

Authors: Anh Tuan Bui, Trung Le, He Zhao, Paul Montague, Olivier deVel, Tamas Abraham, Dinh Phung6831-6839

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive and comprehensive experiments to demonstrate that our proposed method outperforms the state-of-the-art ensemble baselines, at the same time can detect a wide range of adversarial examples with a nearly perfect accuracy. Our code is available at: https://github.com/tuananhbui89/Crossing-Collaborative Ensemble.
Researcher Affiliation Collaboration Anh Tuan Bui1, Trung Le1, He Zhao1 Paul Montague2, Olivier de Vel2, Tamas Abraham2, Dinh Phung1 1Monash University, Australia 2Defence Science and Technology Group, Australia
Pseudocode No The paper describes the proposed method using prose and mathematical equations but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks or figures.
Open Source Code Yes Our code is available at: https://github.com/tuananhbui89/Crossing-Collaborative Ensemble.
Open Datasets Yes We use CIFAR10 and CIFAR100 as the benchmark datasets in our experiment.
Dataset Splits No The paper states 'Both datasets have 50,000 training images and 10,000 test images,' which defines the training and test sets, but does not explicitly provide details for a separate validation split or how it was derived.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory specifications, or cloud resources) used for running the experiments.
Software Dependencies No The paper mentions software like 'cleverhans', 'foolbox', and 'tensorflow' but does not provide specific version numbers for these or any other key software components used in the experiments.
Experiment Setup Yes Specifically, the configuration for the CIFAR10 dataset is k = 10, ϵ = 8/255, η = 2/255 and that for the CIFAR100 dataset is k = 10, ϵ = 0.01, η = 0.001. For the CIFAR10 dataset with Res Net architecture, we use the same setting in (Pang et al. 2019) which is k = 10, ϵ U(0.01, 0.05), η = ϵ/10.