Flint: A Platform for Federated Learning Integration

By Ewen Wang et al

Published on June 10, 2023

Read the original document by opening this link in a new tab.

Table of Contents

1. Introduction
2. Background
3. System Design
4. Real World Measurements
5. Proxy Data Generator
6. Data Locality

Summary

This paper presents a device-cloud collaborative FL platform called FLINT that integrates with an existing machine learning platform. It addresses the challenges of cross-device federated learning, focusing on privacy, security, and performance. The platform provides tools to measure real-world constraints, assess infrastructure capabilities, evaluate model training performance, and estimate system resource requirements. The paper also introduces a decision workflow to evaluate the trade-offs of cross-device FL. Real-world measurements and on-device benchmarks are shared to provide insights into system constraints and performance. The platform enables modelers to benchmark existing models under realistic heterogeneous conditions and generate per-device proxy datasets for evaluation. Additionally, FLINT offers a feature catalog to manage device-based features, data locality, and feature transformations. The paper aims to guide the integration of federated learning into production systems and facilitate informed decision-making for businesses.