Summary of Detrigger: a Gradient-centric Approach to Backdoor Attack Mitigation in Federated Learning, by Kichang Lee et al.
DeTrigger: A Gradient-Centric Approach to Backdoor Attack Mitigation in Federated Learning
by Kichang Lee, Yujin Shin, Jonghyuk Yun, Songkuk Kim, Jun Han, JeongGil Ko
First submitted to arxiv on: 19 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR)
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 The proposed DeTrigger framework is a scalable and efficient approach to detect and mitigate backdoor attacks in Federated Learning (FL) systems. By leveraging insights from adversarial attack methodologies, DeTrigger employs gradient analysis with temperature scaling to identify and isolate backdoor triggers, allowing for precise model weight pruning of backdoor activations without compromising benign model knowledge. The framework demonstrates significant improvement over traditional methods, achieving up to 251x faster detection and mitigating backdoor attacks by up to 98.9% on four widely used datasets. DeTrigger’s robustness and scalability make it an effective solution for protecting FL environments against sophisticated backdoor threats. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Federated Learning (FL) is a way to train AI models together with many devices without sharing their data. But this makes it hard to keep the models safe from bad guys who might try to trick them into making wrong predictions. This paper proposes a new way called DeTrigger that can detect and stop these attacks. It uses some clever math tricks to figure out which parts of the model are being controlled by the attackers, and then it removes those parts without hurting the rest of the model. The team tested DeTrigger on four different sets of data and found that it worked really well, keeping the bad guys from causing trouble 98.9% of the time. |
Keywords
» Artificial intelligence » Federated learning » Pruning » Temperature
