Summary of How to Defend Against Large-scale Model Poisoning Attacks in Federated Learning: a Vertical Solution, by Jinbo Wang et al.
How to Defend Against Large-scale Model Poisoning Attacks in Federated Learning: A Vertical Solution
by Jinbo Wang, Ruijin Wang, Fengli Zhang
First submitted to arxiv on: 16 Nov 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 The proposed defense mechanism, VERT, tackles the issue of large-scale model poisoning attacks in federated learning (FL) by transforming the problem from a horizontal solution to a vertical solution. This approach leverages the predictable nature of the model’s convergence process and utilizes historical gradients information to predict user gradients. A low-dimensional vector projector is designed to reduce computational complexity while maintaining defense efficacy. Experimental results demonstrate VERT’s efficiency, scalability, and excellent performance in defending against large-scale (>=80%) model poisoning attacks under various FL scenarios. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Federated learning helps many devices learn together without sharing their data. But what if some of these devices try to trick the system? This can happen with “model poisoning” attacks. The current defenses don’t work well when there are a lot of attackers (more than 50%). Researchers have come up with a new way to stop this by using the pattern of how models change over time. They call it VERT, which stands for Vertical Ensemble-based Robust Training. It predicts what gradients (model updates) each device will send and compares them to the actual ones. This helps pick the right combination of gradients to keep the model safe. To make it faster, they use a special tool that reduces the size of the data while keeping its important features. Tests show that VERT works well even with many attackers. |
Keywords
* Artificial intelligence * Federated learning
