A Dual Stealthy Backdoor: From Both Spatial and Frequency Perspectives
Authors: Yudong Gao, Honglong Chen, Peng Sun, Junjian Li, Anqing Zhang, Zhibo Wang, Weifeng Liu
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate DUBA extensively on four datasets against popular image classifiers, showing significant superiority over state-of-the-art backdoor attacks in attack success rate and stealthiness. Experimental Settings. Attack Performance Evaluation. Attack Stealthiness. Robustness to Defenses. Ablation Studies. |
| Researcher Affiliation | Academia | 1College of Control Science and Engineering, China University of Petroleum (East China), P.R. China 2College of Computer Science and Electronic Engineering, Hunan University, P.R. China 3School of Cyber Science and Technology, Zhejiang University, P.R. China |
| Pseudocode | No | No, the paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | 1codes: https://github.com/ifen1/Dual-Stealthy-Backdoor |
| Open Datasets | Yes | Cifar10 (Krizhevsky, Hinton et al. 2009), Gtsrb (Stallkamp et al. 2012), Image Net (Deng et al. 2009), and Fer2013 (Goodfellow et al. 2013). |
| Dataset Splits | No | No, the paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce the data partitioning into train, validation, and test sets. |
| Hardware Specification | No | No, the paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | No, the paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment. |
| Experiment Setup | Yes | In our experiments, we randomly select an image with a dog s ear as the initial trigger. During training, α and β are set to 0.4. In the attack phase, α and β are both set to 0.6, λ to 0.7. |