# LEAF: A Benchmark for Federated Settings

**Authors:** Sebastian Caldas, Sai Meher Karthik Duddu, Peter Wu, Tian Li, Jakub Konečný, H. Brendan McMahan, Virginia Smith, Ameet Talwalkar  
**arXiv:** [1812.01097](https://arxiv.org/abs/1812.01097)  
**Date:** 2018-12-03  
**Summary page:** https://www.summarizepaper.com/en/paper/1812.01097v3/

## Key points

- Authors address challenges in modern federated networks
- Federated networks consist of various devices generating vast amounts of data daily
- Data has potential to enhance user experiences through model learning
- Scale and diversity of data introduce complexities in federated learning, meta-learning, and multi-task learning
- LEAF provides open-source federated datasets and reference implementations
- LEAF is a modular benchmarking solution addressing obstacles related to data heterogeneity and scalability within federated networks

## Summary

In their paper titled "LEAF: A Benchmark for Federated Settings," authors Sebastian Caldas, Sai Meher Karthik Duddu, Peter Wu, Tian Li, Jakub Konečný, H. Brendan McMahan, Virginia Smith, and Ameet Talwalkar address the challenges posed by modern federated networks. These networks consist of various devices such as wearables, mobile phones, and autonomous vehicles that generate vast amounts of data daily. This data holds the potential to enhance user experiences through model learning; however, the scale and diversity of federated data introduce complexities in areas like federated learning, meta-learning, and multi-task learning. Recognizing the need for realistic benchmarks in this evolving landscape of machine learning research, LEAF encompasses a collection of open-source federated datasets, and a series of reference implementations designed to capture the nuances and challenges inherent in practical federated environments. By introducing LEAF as a modular benchmarking solution, This initiative underscores the importance of addressing obstacles related to data heterogeneity and scalability within federated networks to drive innovation and progress in machine learning applications across diverse device ecosystems.