ByzFL: Research Framework for Robust Federated Learning

2025-05-30

AI Summary

Robust federated learning (FL) lacks a unified, scalable evaluation framework for systematic benchmarking of Byzantine-resilient algorithms. Method: This paper introduces ByzFL—a first-of-its-kind open-source, JSON-driven framework for robust FL research, built on PyTorch and NumPy. It features a modular architecture supporting customizable aggregators, Byzantine attack strategies, heterogeneous data and model distribution simulation, and integrated visualization. Contribution/Results: ByzFL enables systematic, cross-algorithm, cross-attack, and cross-distribution benchmarking, significantly enhancing experimental reproducibility and development efficiency. We validate the effectiveness of mainstream defense methods on standard benchmarks including CIFAR-10 and MNIST. The framework is publicly available as open-source software and has been widely adopted by the research community.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Computer Vision: Adversarial Attacks & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• Responsible Web: Data and user privacy-enhancing technologies for the Web
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

We present ByzFL, an open-source Python library for developing and benchmarking robust federated learning (FL) algorithms. ByzFL provides a unified and extensible framework that includes implementations of state-of-the-art robust aggregators, a suite of configurable attacks, and tools for simulating a variety of FL scenarios, including heterogeneous data distributions, multiple training algorithms, and adversarial threat models. The library enables systematic experimentation via a single JSON-based configuration file and includes built-in utilities for result visualization. Compatible with PyTorch tensors and NumPy arrays, ByzFL is designed to facilitate reproducible research and rapid prototyping of robust FL solutions. ByzFL is available at https://byzfl.epfl.ch/, with source code hosted on GitHub: https://github.com/LPD-EPFL/byzfl.

Problem

• Develop robust federated learning algorithms
• Simulate diverse FL scenarios and attacks
• Enable reproducible research in FL

Innovation

• Unified framework for robust federated learning
• Configurable attacks and robust aggregators
• JSON-based configuration for systematic experimentation