Decentralized Federated Learning: Fundamentals, State of the Art, Frameworks, Trends, and Challenges
Open accessThe DOI matches its Crossref record; 230 of the 230 references it cites were found in a registry or an index.
Abstract
In recent years, Federated Learning (FL) has gained relevance in training collaborative models without sharing sensitive data. Since its birth, Centralized FL (CFL) has been the most common approach in the literature, where a central entity creates a global model. However, a centralized approach leads to increased latency due to bottlenecks, heightened vulnerability to system failures, and trustworthiness concerns affecting the entity responsible for the global model creation. Decentralized Federated Learning (DFL) emerged to address these concerns by promoting decentralized model aggregation and minimizing reliance on centralized architectures. However, despite the work done in DFL, the literature has not (i) studied the main aspects differentiating DFL and CFL; (ii) analyzed DFL frameworks to create and evaluate new solutions; and (iii) reviewed application scenarios using DFL. Thus, this article identifies and analyzes the main fundamentals of DFL in terms of federation architectures, topologies, communication mechanisms, security approaches, and key performance indicators. Additionally, the paper at hand explores existing mechanisms to optimize critical DFL fundamentals. Then, the most relevant features of the current DFL frameworks are reviewed and compared. After that, it analyzes the most used DFL application scenarios, identifying solutions based on the fundamentals and frameworks previously defined. Finally, the evolution of existing DFL solutions is studied to provide a list of trends, lessons learned, and open challenges.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Data breaches within third-party services could lead to the loss of proprietary vehicle models or private data, compromising user privacy and vehicle security on a large scale [7, 12, 22].”
From Privacy-preserving Machine Learning in Internet of Vehicle Applications: Fundamentals, Recent Advances, and Future Direction (2025) · cites this paper as [12]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“In DFL, achieving consensus among participating nodes is crucial to ensure that model updates are coherent and reflect the contributions of all peers [12].”
From Privacy-preserving Machine Learning in Internet of Vehicle Applications: Fundamentals, Recent Advances, and Future Direction (2025) · cites this paper as [12]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Data security is another critical challenge, as the transmission and storage of sensitive data create vulnerabilities that may lead to breaches, exposing both individual and aggregate information [7, 12, 22].”
From Privacy-preserving Machine Learning in Internet of Vehicle Applications: Fundamentals, Recent Advances, and Future Direction (2025) · cites this paper as [12]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“The choice of topology affects the overall privacy, robustness, flexibility, fault tolerance, and communication costs of the network [12].”
From Privacy-preserving Machine Learning in Internet of Vehicle Applications: Fundamentals, Recent Advances, and Future Direction (2025) · cites this paper as [12]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“While CFL is effective in various scenarios, it has limitations such as a single point of failure, bottlenecks at the server, and various privacy risks associated with centralized data aggregation [12].”
From Privacy-preserving Machine Learning in Internet of Vehicle Applications: Fundamentals, Recent Advances, and Future Direction (2025) · cites this paper as [12]
Experimental — accuracy not yet measured
4 more claims about this paper
Create a free account to see every claim other papers make about this one.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Items 201–231 of 231
On the Decentralization of Blockchain-enabled Asynchronous Federated Learning
Wilhelmi, F., Guerra, E., & Dini, P. (2023) · IEEE International Conference on Network Softwarization (NetSoft) · doi:10.1109/netsoft57336.2023.10175411 · Cited by 9 · cited 1×
“Wilhelmi et al. [193] studied the impact of ledger inconsistencies on DFL performance.”
BackgroundA Hybrid Incentive Mechanism for Decentralized Federated Learning
Qi, M., Wang, Z., Chen, S., & Xiang, Y. (2022) · Distributed Ledger Technologies Research and Practice · doi:10.1145/3538226 · Cited by 8 · cited 2×
“DLT technologies, such as Blockchain, provide an ideal solution for this requirement by ensuring the immutability of exchanged models [68].”
BackgroundFederated Learning for Maritime Environments:Use Cases, Experimental Results, and Open IssuesPreprint
Giannopoulos, A., Gkonis, P. K., Bithas, P. S., Νομικός, Ν., Ntroulias, G., & Trakadas, P. T. (2023) · doi:10.36227/techrxiv.22133549.v1 · Cited by 8
ZeroAccess botnet investigation and analysis
Rawat, R. S., Diwakar, M., & Verma, P. (2021) · International Journal of Information Technology · doi:10.1007/s41870-021-00693-z · Cited by 7 · cited 2×
“In this sense, DFL can be subject to various attacks, including malware deployment and adversarial attacks, which can affect federation models and decentralized data, similar to the CFL approach [76].”
Scatterbrained: A flexible and expandable pattern for decentralized machine learningPreprint
Wilt, M., Matelsky, J., & Gearhart, A. S. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2112.07718 · Cited by 7
DeFL: Decentralized Weight Aggregation for Cross-silo Federated LearningPreprint
Han, J., Han, Y., Huang, G., & Ma, Y. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2208.00848 · Cited by 7
Decentralized Zero-Trust Framework for Digital Twin-based 6GPreprint
Ridhawi, I. A., Otoum, S., & Aloqaily, M. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2302.03107 · Cited by 7 · cited 1×
“In this sense, the use of advanced technologies such as 5G [138] or 6G [139] can help reduce communication limitations, improving network bandwidth and enable faster and more efficient data exchange.”
BackgroundAdvancements in Federated Learning: Models, Methods, and PrivacyPreprint
Chen, H., Wang, H., Long, Q., et al. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2302.11466 · Cited by 7 · cited 1×
“Additionally, Chen et al. [35] analyzed the advances in FL, including aspects such as communications, privacy, and security requirements in DFL.”
BackgroundKhelghatdoust, M., & Mahdavi, M. (2022) · Lecture notes in networks and systems · doi:10.1007/978-3-030-99587-4_17 · Cited by 6 · cited 2×
“Finally, some application scenarios of gossip communications in DFL are providing accurate recommendations (e.g., location logs, movie ratings) [72] or in Distributed Online Social Networks (DOSN) where users can directly communicate only with their immediate neighbors in a social graph [73].”
BackgroundFederatedScope: A Flexible Federated Learning Platform for HeterogeneityPreprint
Xie, Y., Wang, Z., Gao, D., et al. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2204.05011 · Cited by 6
CRAWDAD dataset uportorwthaachen/vanetjamming2012 (v. 2014-05-12)
Puñal, O., Pereira, C., Aguiar, A., & Gross, J. (2014) · CRAWDAD wireless network data archive · doi:10.15783/c7ts35 · Cited by 5 · cited 2×
“[208] 2020 Federated learning-based security architecture for jamming attack detection for FANET Cross-device Centralized Decentralized Simulation Three-layered NN FedAvg Asynchronous Random CRAWDAD [209] NS3 FANET [210]”
MethodInfluentialEfficient Fully Distributed Federated Learning with Adaptive Local LinksPreprint
Georgatos, E., Mavrokefalidis, C., & Berberidis, K. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2203.12281 · Cited by 5
Decentralized Federated Learning: A Defense Against Gradient Inversion Attack
Guang-xi, L., Xiong, Z., Li, R., & Li, W. (2023) · Lecture notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering · doi:10.1007/978-3-031-27041-3_4 · Cited by 5 · cited 1×
“However, exchanging gradients or meta-level information during distributed training can reveal information about the training data and model updates, making exchanging model weights preferable for privacy preservation [81].”
BackgroundFederatedTrust: A Solution for Trustworthy Federated LearningPreprint
Sánchez, P. M. S., Celdrán, A. H., Xie, N., et al. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2302.09844 · Cited by 4 · cited 1×
“(cid:129) (cid:129) Security and privacy: The preservation of security and privacy is paramount in DFL, even while allowing effective model training [95].”
BackgroundSelf-citationP4L: Privacy Preserving Peer-to-Peer Learning for Infrastructureless SetupsPreprint
Arapakis, I., Papadopoulos, P. D., Katevas, K., & Perino, D. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2302.13438 · Cited by 4 · cited 1×
“Finally, it is worth mentioning that other studies have explored the use of innovative techniques in DFL, such as P2P with knowledge distillation, for various applications such as social network scenarios or IoT device identification [201], [204], [206].”
BackgroundAl-Abiad, M. S., Hassan, M. Z., & Hossain, M. J. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2303.00032 · Cited by 4 · cited 2×
“[212] 2023 Resource-efficient framework for mmWave aerial-terrestrial integrated networks using UAVs Cross-device Decentralized Simulation N/S Custom algorithm Asynchronous Scheduling MNIST [172] CIFAR-10 [152]”
MethodSánchez, P. M. S., Celdrán, A. H., Beltrán, E. T. M., et al. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2210.11061 · Cited by 3
Self-citationDecentral and Incentivized Federated Learning Frameworks: A Systematic Literature Review
Witt, L., Heyer, M., Toyoda, K., et al. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2205.07855 · Cited by 3
Tejasree, A. D., Neeharika, B., & Mary, S. P. (2021) · Lecture notes in electrical engineering · doi:10.1007/978-981-15-7504-4_49 · Cited by 2 · cited 1×
“[2] 2022 Collaboratively training between healthcare device for creating a robust model without raw data exchange Cross-device Decentralized Simulation 3D CNN Custom FedAvg (RingAVG) Synchronous Asynchronous Random COVID-19 CT [165] Eye Disease [166] Skin Cancer [167] [22] 2022 Ring topology for DGM to ensure security…”
MethodTowards More Suitable Personalization in Federated Learning via Decentralized Partial Model TrainingPreprint
Shi, Y., Liu, Y., Sun, Y., Lin, Z., Shen, L., Wang, X., & Tao, D. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2305.15157 · Cited by 2 · cited 1×
“Similarly, the distinct data distribution across nodes in DFL could be addressed using federated sampling techniques, which help maintain data diversity while ensuring statistical accuracy [119].”
BackgroundMitigating Communications Threats in Decentralized Federated Learning through Moving Target DefensePreprint
Beltrán, E. T. M., Sánchez, P. M. S., Bernal, S. L., et al. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2307.11730 · Cited by 2 · cited 2×
“Specifically, random neighbor selection and dynamic IP/port switching offer enhanced resilience against communication-based threats [83].”
BackgroundSelf-citationGómez, Á. L. P., Beltrán, E. T. M., Sánchez, P. M. S., & Celdrán, A. H. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2308.03554 · Cited by 2 · cited 1×
“Moreover, anomaly detection has significantly improved by incorporating time series analysis techniques [84].”
MethodSelf-citationFedER: Federated Learning through Experience Replay and Privacy-Preserving Data SynthesisPreprint
Pennisi, M., Salanitri, F. P., Bellitto, G., et al. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2206.10048 · Cited by 2 · cited 2×
“[157] 2022 Exploit concepts from experience repetition and adversarial generation research Cross-device Centralized Decentralized Simulation GAN FedAvg FedProx FedBN Synchronous Asynchronous Random X-ray data [158], [159] HAM10000 [160]”
BackgroundMethodOn the Decentralization of Blockchain-enabled Asynchronous Federated LearningPreprint
Wilhelmi, F., Guerra, E., & Dini, P. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2205.10201 · Cited by 1 · cited 1×
“Wilhelmi et al. [193] studied the impact of ledger inconsistencies on DFL performance.”
BackgroundPersonalized Decentralized Federated Learning with Knowledge DistillationPreprint
Jeong, E., & Kountouris, M. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2302.12156 · Cited by 1
Decentralized Online Federated G-Network Learning for Lightweight Intrusion DetectionPreprint
Nakıp, M., Gül, B. C., & Gelenbe, E. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2306.13029 · Cited by 1 · cited 1×
“Finally, the telecommunications sector benefits from DFL in optimizing network performance and anomaly detection, allowing network providers to share insights without compromising proprietary information [159].”
BackgroundDecentralized Event-Triggered Federated Learning with Heterogeneous Communication ThresholdsPreprint
Zehtabi, S., Hosseinalipour, S., & Brinton, C. G. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2204.03726 · cited 2×
“However, they cause higher communication costs and lower generalization due to staleness [60].”
BackgroundFLoBC: A Decentralized Blockchain-Based Federated Learning FrameworkPreprint
Ghanem, M. C., Dawoud, F., Gamal, H., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2112.11873
CFL: Cluster Federated Learning in Large-scale Peer-to-Peer NetworksPreprint
Chen, Q., Wang, Z., Zhou, Y., et al. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2204.03843
Lumos: Heterogeneity-aware Federated Graph Learning over Decentralized DevicesPreprint
Pan, Q., Zhu, Y., & Chu, L. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2303.00492 · cited 1×
“To improve the previous scenarios, Pan et al. [199] utilized a tree constructor to improve representation capability given the limited structural information and a Monte Carlo Markov Chain-based algorithm to mitigate workload imbalance caused by degree heterogeneity.”
Communication-Efficient Design for Quantized Decentralized Federated LearningPreprint
Chen, L., Liu, W., Chen, Y., et al. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2303.08423 · cited 2×
“A recent contribution includes a method that applies the Lloyd-Max algorithm to DFL to minimize quantization distortion, with the resulting LMDFL algorithm able to adjust quantization levels adaptively [102].”
Method
Research works printed but not linked: 13
- 2022, DISCO—Distributed collaborative machine learning
- Li, 2022, Mining latent relationships among clients: Peer-to-peer federated learning with adaptive neighbor matching, arXiv:2203.12285 · arXiv
- 2022, Tensorflow federated
- Go, 2009, Twitter sentiment classification using distant supervision
- 2016, Geo dataset browser
- Dos, 2021, Predicting human eye diseases
- Savazzi, 2020, Federated learning: mmWave MIMO radar dataset for testing
- 10.1145/3340531.3411866 · DOI
- 2016, Avito context ad clicks
- 2021, IoT device identification
- Koch, 2021, On digital ethics for artificial intelligence in hybrid military operations
- Reiser, 2022, Creiser/drone-detection
- Fisher, 2019, CVONLINE: Image databases
Cited by
Decentralized Federated Learning: A Survey and Perspective
Yuan, L., Wang, Z., Sun, L., Yu, P. S., & Brinton, C. G. (2024) · IEEE Internet of Things Journal · doi:10.1109/jiot.2024.3407584 · Cited by 290
Chellapandi, V. P., Yuan, L., Brinton, C. G., Żak, S. H., & Wang, Z. (2024) · IEEE Transactions on Intelligent Vehicles · doi:10.1109/tiv.2023.3332675 · Cited by 236
Abbas, S. R., Abbas, Z., Zahir, A., & Lee, S. W. (2024) · Healthcare · doi:10.3390/healthcare12242587 · Cited by 228
Distributed Foundation Models for Multi-Modal Learning in 6G Wireless Networks
Du, J., Lin, T., Jiang, C., Yang, Q., Bader, C. F., & Han, Z. (2024) · IEEE Wireless Communications · doi:10.1109/mwc.009.2300501 · Cited by 187
“SL enhances privacy by transmitting only non-sensitive features, reduces bandwidth, and boosts efficiency by allocating lighter tasks to local devices and distributed paradigm, federated learning (FL) is well-suited for assisting the data parallelism [11].”
BackgroundA systematic review of federated learning: Challenges, aggregation methods, and development tools
Guendouzi, S. B., Ouchani, S., Assaad, H. E., & Zaher, M. E. (2023) · Journal of Network and Computer Applications · doi:10.1016/j.jnca.2023.103714 · Cited by 199
Metadata
- Authors
- Martínez Beltrán, E. T., Pérez, M. Q., Sánchez, P. M. S., Bernal, S. L., Bovet, G., Pérez, M. G., Pérez, G. M., & Celdrán, A. H.
- Affiliations
- Department of Information and Communications Engineering, University of Murcia, Murcia, Spain
- Universidad de Murcia (ES)
- Bundesamt für Rüstung armasuisse (CH)
- University of Zurich (CH)
- ORCID
- Venue
- IEEE Communications Surveys & Tutorials
- Published
- 2023
- Volume
- 25
- Issue
- 4
- Pages
- 2983-3013
- ISSN
- 1553-877X (ISSN-L)
- 2373-745X
- OpenAlex
- W4386766438
- Cited by
- 625
- References
- 220
- Citations by year
- 2026: 175 · 2025: 282 · 2024: 145 · 2023: 23
- FWCI
- 58.92
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- IEEE Communications Surveys & Tutorials (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- Zurich Open Repository and Archive (University of Zurich) (submittedVersion)
- arXiv (Cornell University)
- Funders
- European Union (PID2021-122466OB-I00, TED2021-129300B-I00)
- the Swiss Federal Office for Defense Procurement (CyberForce, DEFENDIS)
- Universit?t Z?rich
- Fundaci?n S?neca (21629/FPI/21)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
- Physical Sciences › Computer Science › Information Systems › Blockchain Technology Applications and Security
- Physical Sciences › Computer Science › Computer Networks and Communications › IoT and Edge/Fog Computing
- Keywords
- decentralized federated learning, communication mechanisms, performance indicators, federated learning frameworks, application scenarios, open problems
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.