# Decentralized Federated Learning: Fundamentals, State of the Art, Frameworks, Trends, and Challenges

**Authors:** Enrique Tomás Martínez Beltrán, Mario Quiles Pérez, Pedro Miguel Sánchez Sánchez, Sergio López Bernal, Gérôme Bovet, Manuel Gil Pérez, Gregorio Martínez Pérez, Alberto Huertas Celdrán  
**arXiv:** [2211.08413](https://arxiv.org/abs/2211.08413)  
**Date:** 2022-11-15  
**Summary page:** https://www.summarizepaper.com/en/paper/2211.08413v3/

## Key points

- Federated Learning (FL) is a popular method for training collaborative models without sharing sensitive data
- Centralized FL (CFL) is the most common approach, but it has issues with latency, system failures, and trustworthiness
- Decentralized Federated Learning (DFL) emerged as an alternative to address these issues
- There is a lack of comprehensive study on DFL compared to CFL
- This article aims to bridge the gap by identifying and analyzing key fundamentals of DFL, including federation architectures, network topologies, communication mechanisms, security approaches, and key performance indicators (KPIs)
- The article also explores existing mechanisms to optimize these fundamentals and reviews and compares current DFL frameworks
- Different application scenarios using DFL are analyzed, including healthcare, Industry 4.0, mobile services, and military vehicles
- The article studies the evolution of existing DFL solutions and highlights open challenges in the field

## Summary

In the last decade, Federated Learning (FL) has become increasingly popular as a method for training collaborative models without sharing sensitive data. Centralized FL (CFL) has been the most common approach, where a central entity creates a global model. However, CFL leads to increased latency, vulnerability to system failures, and trustworthiness concerns. Decentralized Federated Learning (DFL) emerged as an alternative to address these issues by promoting decentralized model aggregation and minimizing reliance on centralized architectures. Despite the work done in DFL, there is still a lack of comprehensive study on its main aspects compared to CFL. This article aims to bridge this gap by identifying and analyzing the key fundamentals of DFL, including federation architectures, network topologies, communication mechanisms, security approaches, and key performance indicators (KPIs). It also explores existing mechanisms to optimize these fundamentals. Furthermore, the article reviews and compares the most relevant features of current DFL frameworks. By considering these fundamentals and frameworks, researchers can develop and compare frameworks capable of training DFL-based models with better performance, security, scalability, and trustworthiness across various use cases. The article also analyzes different application scenarios using DFL based on the identified fundamentals and implemented frameworks. Some of the most commonly used scenarios include healthcare for decentralizing clinical records and collaborative diagnosis; Industry 4.0 for improving automated industrial systems efficiency; mobile services for enhancing user experience through personalized models without compromising privacy; and military vehicles for secure collaboration among vehicles in battlefield environments. By examining these application scenarios according to the fundamentals and frameworks they implement, researchers can identify strengths and weaknesses of different approaches and make informed decisions when selecting suitable solutions for real-world scenarios. Finally, the article studies the evolution of existing DFL solutions to provide insights into trends in this field as well as lessons learned from previous implementations. It also highlights open challenges that need further research attention. Overall, this article expands upon the existing literature by comprehensively identifying and describing the fundamentals KPIs optimization mechanisms frameworksand application scenarios of DFL.