Summary of Badsfl: Backdoor Attack Against Scaffold Federated Learning, by Xingshuo Han et al.
BadSFL: Backdoor Attack against Scaffold Federated Learning
by Xingshuo Han, Xuanye Zhang, Xiang Lan, Haozhao Wang, Shengmin Xu, Shen Ren, Jason Zeng, Ming Wu, Michael Heinrich, Tianwei Zhang
First submitted to arxiv on: 25 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes a novel backdoor attack method called BadSFL for federated learning (FL) in non-IID scenarios. The attack leverages Generative Adversarial Network (GAN)-based augmentations to train models that achieve high accuracy on both benign and backdoor samples. It uses a specific feature as the backdoor trigger, ensuring stealthiness, and exploits the scaffold’s control variate to predict the global model’s convergence direction, maintaining persistence. The proposed attack is evaluated on three benchmark datasets, demonstrating its effectiveness, stealthiness, and durability over 60 rounds in the global model or up to 3 times longer than existing baseline attacks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Federated learning (FL) helps train deep learning models without sharing data. But attackers can secretly contaminate these models, making them predict things they shouldn’t. The current strategies for this type of attack don’t work well in real-world scenarios where the training data are not identical and independent. This paper suggests a new way to do this kind of attack, called BadSFL. It uses a special technique called Generative Adversarial Network (GAN) to make the model better at predicting things it should predict, including backdoors. The attackers can control how well the backdoor works by choosing which features to use as triggers. This method was tested on three datasets and showed that it’s effective, sneaky, and lasts a long time. |
Keywords
» Artificial intelligence » Deep learning » Federated learning » Gan » Generative adversarial network
