Effective Backdoor Defense by Exploiting Sensitivity of Poisoned Samples

Authors: Weixin Chen, Baoyuan Wu, Haoqian Wang

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive results on three benchmark datasets demonstrate the superior defense performance against eight types of backdoor attacks, to state-of-the-art backdoor defenses. Codes are available at: https://github.com/SCLBD/Effective_backdoor_defense.
Researcher Affiliation Academia Weixin Chen1, Baoyuan Wu2 , Haoqian Wang1 1Tsinghua Shenzhen International Graduate School, Tsinghua University 2School of Data Science, Shenzhen Research Institute of Big Data, The Chinese University of Hong Kong, Shenzhen
Pseudocode Yes More details are in Algorithm 1 in Appendix A.1
Open Source Code Yes Codes are available at: https://github.com/SCLBD/Effective_backdoor_defense.
Open Datasets Yes We evaluate all attacks on 3 benchmark datasets, CIFAR-10 [3], CIFAR-100 [3] and an Image Net subset [33, 9], with Res Net-18 [34] as the base model.
Dataset Splits No The paper references using 'Dtrain' for training and evaluates 'Test ACC' and 'Test ASR', but does not explicitly state the dataset split percentages (e.g., train/validation/test percentages or counts) within the main text.
Hardware Specification No The paper does not provide specific details about the hardware used for experiments, such as GPU or CPU models.
Software Dependencies No The paper does not provide specific software dependency versions (e.g., library names with version numbers) required to replicate the experiments.
Experiment Setup Yes Poisoning rate is set to 10% in all attacks. For our proposed methods, we use αc = 20%, αp = 5% and τ =rotate+affine in all experiments. Other details can be seen in Appendix C.5.