Summary of Meta Stackelberg Game: Robust Federated Learning Against Adaptive and Mixed Poisoning Attacks, by Tao Li et al.
Meta Stackelberg Game: Robust Federated Learning against Adaptive and Mixed Poisoning Attacks
by Tao Li, Henger Li, Yunian Pan, Tianyi Xu, Zizhan Zheng, Quanyan Zhu
First submitted to arxiv on: 22 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Cryptography and Security (cs.CR); Computer Science and Game Theory (cs.GT)
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 tackles the security concerns in Federated Learning (FL) by developing a robust and adaptive defense mechanism. The authors formulate FL under mixed attacks as a Bayesian Stackelberg Markov game, allowing them to propose a meta-Stackelberg defense. This defense is composed of pre-training with reinforcement learning-based attacks and online adaptation using meta-reinforcement learning. An efficient meta-learning approach is developed to solve the game, ensuring convergence to the first-order ε-meta-equilibrium point in O(ε^-2) gradient iterations with O(ε^-4) samples per iteration. Experimental results demonstrate the framework’s effectiveness against strong model poisoning and backdoor attacks of uncertain and unknown types. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Federated learning (FL) is a way for many devices to work together without sharing their data. But, this process can be attacked by bad actors. The researchers in this paper came up with a new way to defend FL against these threats. They used game theory to create a defense system that adapts to different types of attacks. This system is made up of two parts: one that trains the model and another that updates it as needed. The authors tested their approach and found that it works well against strong attacks. |
Keywords
* Artificial intelligence * Federated learning * Meta learning * Reinforcement learning
