Summary of Byzantine-robust Federated Learning: Impact Of Client Subsampling and Local Updates, by Youssef Allouah et al.
Byzantine-Robust Federated Learning: Impact of Client Subsampling and Local Updates
by Youssef Allouah, Sadegh Farhadkhani, Rachid GuerraouI, Nirupam Gupta, Rafael Pinot, Geovani Rizk, Sasha Voitovych
First submitted to arxiv on: 20 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 paper addresses the issue of adversarial clients in federated learning (FL), which can manipulate the training process arbitrarily. To robustify FL, a simple averaging operation is replaced with a robust averaging rule. However, previous research has overlooked the impact of client subsampling and local steps on the performance of FL algorithms like FedRo. This study analyzes the effects of these factors on FedRo’s convergence, presenting conditions for nearly-optimal convergence when client subsampling is limited and showing that the rate of improvement in learning accuracy diminishes with increased subsampling. Additionally, the paper demonstrates that carefully choosing step-sizes can reduce the learning error due to Byzantine clients as local steps increase. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Federated learning is a way for many devices to work together to train a single model without sharing their individual data. But what if some of these devices are trying to trick the system? This paper looks at how to make federated learning more robust against these “bad” devices, called Byzantine clients. The usual way to do this is by using a special kind of averaging rule. However, people have forgotten to consider two important things: which devices get picked for training and how many times each device does its own calculations before sending the results back. This paper studies what happens when these factors are taken into account and shows that if too many devices are chosen or do their calculations too many times, it can actually make things worse. But with careful choices, federated learning can become more reliable and accurate. |
Keywords
* Artificial intelligence * Federated learning
