Summary of Byzantine-resilient Federated Learning Employing Normalized Gradients on Non-iid Datasets, by Shiyuan Zuo et al.
Byzantine-resilient Federated Learning Employing Normalized Gradients on Non-IID Datasets
by Shiyuan Zuo, Xingrun Yan, Rongfei Fan, Li Shen, Puning Zhao, Jie Xu, Han Hu
First submitted to arxiv on: 18 Aug 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Distributed, Parallel, and Cluster Computing (cs.DC)
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 Federated Normalized Gradients Algorithm (Fed-NGA) is a novel approach for practical federated learning systems, which tackles the challenges of Byzantine attacks and data heterogeneity. This algorithm achieves a trade-off between adaptability to different loss function types and robustness to heterogeneous datasets without compromising on optimality gap. Fed-NGA normalizes uploaded local gradients to unit vectors, resulting in a time complexity of O(pM), where p represents model parameter dimension and M is the number of participating clients. The algorithm can adapt to both non-convex loss functions and non-IID datasets simultaneously with zero optimality gap at a rate of O(1/T^(1/2 – δ)), where T is the iteration number and δ is in (0, 1/2). For strongly convex loss functions, the rate can be improved to linear. Experimental results demonstrate the superiority of Fed-NGA over baseline methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Federated learning lets devices share information without sharing their data. This helps keep personal info private, but it also makes training harder because devices have different data and might try to trick the system. The new algorithm, called Federated Normalized Gradients Algorithm (Fed-NGA), makes training faster and more accurate while keeping sensitive data safe. It works by normalizing the information each device sends in a way that makes it easier for all devices to work together. |
Keywords
» Artificial intelligence » Federated learning » Loss function
