Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Survey on Decentralized Federated Learning

Edoardo Gabrielli, Anthony Di Pietro, Dario Fenoglio, Giovanni Pica, Gabriele Tolomei

TL;DR

This survey systematically reviews DFL methods from 2018 through early 2026 and organizes them into two architectural families: traditional distributed FL and blockchain-based FL, and proposes a unified, challenge-driven taxonomy that maps both families to the core bottlenecks they primarily address.

Abstract

Federated learning (FL) enables collaborative training without pooling raw data, but standard FL relies on a central coordinator, which introduces a single point of failure and concentrates trust in the orchestration infrastructure. Decentralized federated learning (DFL) removes the coordinator and replaces client-server orchestration with peer-to-peer coordination, making learning dynamics topology-dependent and reshaping the associated security, privacy, and systems trade-offs. This survey systematically reviews DFL methods from 2018 through early 2026 and organizes them into two architectural families: traditional distributed FL and blockchain-based FL. We then propose a unified, challenge-driven taxonomy that maps both families to the core bottlenecks they primarily address, and we summarize prevailing evaluation practices and their limitations, exposing gaps in the literature. Finally, we distill lessons learned and outline research directions, emphasizing topology-aware threat models, privacy notions that reflect decentralized exposure, incentive mechanisms robust to manipulation, and the need to explicitly define whether the objective is a single global model or personalized solutions in decentralized settings.

A Survey on Decentralized Federated Learning

TL;DR

This survey systematically reviews DFL methods from 2018 through early 2026 and organizes them into two architectural families: traditional distributed FL and blockchain-based FL, and proposes a unified, challenge-driven taxonomy that maps both families to the core bottlenecks they primarily address.

Abstract

Federated learning (FL) enables collaborative training without pooling raw data, but standard FL relies on a central coordinator, which introduces a single point of failure and concentrates trust in the orchestration infrastructure. Decentralized federated learning (DFL) removes the coordinator and replaces client-server orchestration with peer-to-peer coordination, making learning dynamics topology-dependent and reshaping the associated security, privacy, and systems trade-offs. This survey systematically reviews DFL methods from 2018 through early 2026 and organizes them into two architectural families: traditional distributed FL and blockchain-based FL. We then propose a unified, challenge-driven taxonomy that maps both families to the core bottlenecks they primarily address, and we summarize prevailing evaluation practices and their limitations, exposing gaps in the literature. Finally, we distill lessons learned and outline research directions, emphasizing topology-aware threat models, privacy notions that reflect decentralized exposure, incentive mechanisms robust to manipulation, and the need to explicitly define whether the objective is a single global model or personalized solutions in decentralized settings.
Paper Structure (44 sections, 2 equations, 4 figures, 3 tables)

This paper contains 44 sections, 2 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Comparison with previous surveys and contributions
  3. Background and Preliminaries
  4. Federated Learning
  5. Data Partitioning
  6. Machine Learning Models
  7. Scale of Federation
  8. Communication Architecture
  9. Peer-to-Peer Systems (P2P)
  10. Blockchain
  11. Research Method
  12. Research Questions
  13. Search Process
  14. Inclusion/Exclusion Criteria
  15. Data Extraction
  16. ...and 29 more sections

Figures (4)

  • Figure 1: Architectural comparison between standard FL and DFL. (a) FL relies on a central orchestrator for client selection and global aggregation. (b) DFL removes the central server and performs P2P coordination, where nodes iteratively train locally, exchange updates with neighbors, and aggregate received updates.
  • Figure 2: Selection process of articles with the PRISMA flow diagram.
  • Figure 3: Timeline of the works included in our taxonomy, grouped by publication year and categorized as traditional DFL (top) versus blockchain-based FL (bottom). The distribution highlights that blockchain-based proposals are concentrated mainly in 2019--2022 within our corpus, whereas traditional DFL continues to attract sustained research activity in later years (2023--2025), reflecting a shift of emphasis toward lighter-weight P2P coordination and direct mechanisms for robustness, heterogeneity, privacy, and communication efficiency.
  • Figure 4: Mapping between the survey structure and the taxonomy in Table \ref{['tab:taxonomy']}. Each column corresponds to (i) a main challenge dimension in DFL and (ii) the dedicated section where that dimension is discussed in the survey. Within each column, blue boxes denote TD-FL works and red boxes denote BC-FL works, listing representative papers that primarily address that challenge.