# Flower: A Friendly Federated Learning Research Framework

**Authors:** Daniel J. Beutel, Taner Topal, Akhil Mathur, Xinchi Qiu, Javier Fernandez-Marques, Yan Gao, Lorenzo Sani, Kwing Hei Li, Titouan Parcollet, Pedro Porto Buarque de Gusmão, Nicholas D. Lane  
**arXiv:** [2007.14390](https://arxiv.org/abs/2007.14390)  
**Date:** 2020-07-28  
**Summary page:** https://www.summarizepaper.com/en/paper/2007.14390v5/

## Key points

- "Flower: A Friendly Federated Learning Research Framework" addresses challenges in implementing Federated Learning (FL) on edge devices
- FL allows collaborative learning of a shared prediction model while keeping training data on the device itself
- Existing research frameworks lack support for scalable workloads on diverse edge devices, but Flower offers new facilities for large-scale FL experiments and consideration of heterogeneous device scenarios
- Flower has successfully performed FL experiments with up to 15 million clients using only a pair of high-end GPUs
- Researchers can seamlessly transition their experiments from Flower to real devices, making it a critical tool for studying and developing Federated Learning techniques within the research community
- Flower's open-source nature and mobile-friendly interface enhance its accessibility and usability for researchers in machine learning and computer vision

## Summary

"Flower: A Friendly Federated Learning Research Framework" by Daniel J. Beutel, Taner Topal, Akhil Mathur, Xinchi Qiu, Javier Fernandez-Marques, Yan Gao, Lorenzo Sani, Kwing Hei Li, Titouan Parcollet, Pedro Porto Buarque de Gusmão and Nicholas D. Lane presents a comprehensive solution to the challenges faced in implementing Federated Learning (FL) on edge devices. FL allows for collaborative learning of a shared prediction model while keeping training data on the device itself. This eliminates the need to store data in the cloud but poses obstacles in scaling and dealing with heterogeneous systems. Existing research frameworks for simulating FL algorithms lack support for scalable workloads on diverse edge devices. However, Flower stands out by offering new facilities that enable large-scale FL experiments and consideration of richly heterogeneous device scenarios. The framework has successfully performed FL experiments with up to 15 million clients using only a pair of high-end GPUs. Researchers can seamlessly transition their experiments from Flower to real devices to explore different aspects of the design space. This makes Flower a critical tool for studying and developing Federated Learning techniques within the research community. Its open-source nature and mobile-friendly interface further enhance its accessibility and usability for researchers in machine learning and computer vision.