Summary of Building Gradient Bridges: Label Leakage From Restricted Gradient Sharing in Federated Learning, by Rui Zhang et al.
Building Gradient Bridges: Label Leakage from Restricted Gradient Sharing in Federated Learning
by Rui Zhang, Ka-Ho Chow, Ping Li
First submitted to arxiv on: 17 Dec 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: 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 A novel federated learning attack called Gradient Bridge (GDBR) is introduced, which recovers the label distribution of training data from shared gradients. The attack leverages the relationship between layer-wise gradients and tracks the flow of gradients to analytically derive batch training labels. Existing lightweight defenses that restrict gradient sharing are shown to be inadequate in mitigating this privacy leakage. GDBR can accurately recover more than 80% of labels in various federated learning settings, highlighting the need for effective defense schemes. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Federated learning is a way to train AI models using data from different devices without sharing that data. This helps keep personal information private. But, new attacks are emerging that could put this privacy at risk. One attack, called Gradient Bridge (GDBR), can figure out what labels were used to train a model just by looking at the gradients shared during training. This is a problem because existing defenses don’t work well against GDBR. The paper shows that GDBR can accurately recover over 80% of labels in different scenarios, highlighting the need for better defense strategies. |
Keywords
* Artificial intelligence * Federated learning
