pith. sign in

arxiv: 2605.16089 · v1 · pith:HGR6GQ3Vnew · submitted 2026-05-15 · 💻 cs.LG · cs.AI

Centralized vs Decentralized Federated Learning: A trade-off performance analysis

Chaimaa Medjadji , Guilain Leduc , Sylvain Kubler , Yves Le Traon This is my paper

Pith reviewed 2026-05-20 20:10 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords federated learningcentralized federated learningdecentralized federated learningsemi-decentralized federated learningperformance trade-offsFedstellar simulatorMNIST datasetMLP classifier
0
0 comments X

The pith

Experiments using Fedstellar simulator on MNIST with MLP reveal performance trade-offs across centralized, decentralized, and semi-decentralized federated learning.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tries to establish the respective strengths, limitations, and trade-offs among centralized, decentralized, and semi-decentralized federated learning by running direct experimental comparisons. It fills a gap where few studies have tested all three architectures side by side on concrete performance indicators such as accuracy, communication load, and convergence. A sympathetic reader would care because many IoT and edge applications must pick an architecture that balances privacy, scalability, and efficiency without central data storage. The results give practitioners a basis for choosing the right setup according to their specific constraints.

Core claim

This paper overcomes the lack of experimental comparisons by conducting analyses using the Fedstellar simulator, MNIST dataset, and MLP classifier to understand the respective strengths and limitations of Centralized (CFL), Decentralized (DFL), and Semi-decentralized (SDFL) federated learning, as well as trade-offs between different performance indicators.

What carries the argument

Side-by-side experimental runs in the Fedstellar simulator that measure accuracy, communication cost, and convergence speed for the three architectures on the same MNIST classification task with an MLP model.

If this is right

  • Centralized federated learning tends to reach higher final accuracy when a reliable central coordinator is available.
  • Decentralized federated learning reduces single-point failures and communication bottlenecks at the cost of slower or less stable convergence.
  • Semi-decentralized federated learning supplies an intermediate option that trades some accuracy for lower communication overhead and better fault tolerance.
  • Application designers can select the architecture by matching their priorities for privacy, scalability, and latency to the observed trade-offs.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the simulator's network and data-partition models match real edge environments, the reported trade-offs could directly guide early architecture decisions in IoT projects.
  • Testing the same setups under non-independent and identically distributed data partitions would likely expose additional convergence and accuracy penalties not captured in the current MNIST runs.
  • The comparison method itself could be reused to benchmark federated learning against other privacy-preserving distributed training approaches.

Load-bearing premise

The performance differences seen in the Fedstellar simulator with MNIST and MLP reflect the trade-offs that would appear in real-world federated learning deployments with varied data, models, and network conditions.

What would settle it

Repeating the same comparison on a different dataset such as CIFAR-10 or with a convolutional neural network and finding that the relative ranking of the three architectures reverses on key metrics.

Figures

Figures reproduced from arXiv: 2605.16089 by Chaimaa Medjadji, Guilain Leduc, Sylvain Kubler, Yves Le Traon.

Figure 1
Figure 1. Figure 1: Federated Learning Architectures for its simplicity and efficiency, is called federated averaging (FedAvg). FedAvg computes the parameter-wise arithmetic mean across the local models. Currently, most of the FL applications rely on a CFL architecture, as illustrated in [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Methodology underlying the method of evaluating and identifing the trade-offs between different KPIs. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: KPIs introduced by Beltran et al. [5] to evaluate (D)FL architectures/implementations. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Performance Evaluation Metrics with |N| = 3 : (a) Accuracy (b) Loss (c) Precision (d) Recall (e) Bytes exchanged in the FL network (f) CPU usage percentage. The DFL is the most efficient one with an accuracy of 97,84%, a loss of 0.072925 using the less resources usage. The SDFL can reach a very close performance to the DFL one with an accuracy of 96.31% and a loss of 0.133429, but it requires more time (se… view at source ↗
Figure 5
Figure 5. Figure 5: Performance Evaluation Metrics with |N| = 4 : (a) Accuracy (b) Loss (c) Bytes exchanged in the federation network (d) CPU usage percentage. The DFL and the SDFL are the most performant architectures with an accuracy of 97.61% and 97.43% respectively. However, the SDFL is faster in term of convergence in comparison with the DFL with a loss of 0.08832. The SDFL is being close to the DFL in term of resource u… view at source ↗
Figure 6
Figure 6. Figure 6: Performance Evaluation Metrics with |N| = 6 : (a) Accuracy (b) Loss (c) Bytes exchanged in the federation network (d) CPU usage percentage. The DFL and the SDFL are the most efficient architectures with an accuracy of 97.5% and 97.22% respectively. However, the SDFL is faster in term of convergence in comparison with the , which is (the DFL) the less communication costing architecture, with a loss of 0.090… view at source ↗
Figure 7
Figure 7. Figure 7: Performance Evaluation Metrics with |N| = 8 : (a) Accuracy (b) Loss (c) Bytes exchanged in the federation network (d) CPU usage percentage. The DFL and the SDFL still the most performant architectures with an accuracy more than 97% . However, the SDFL is faster in term of convergence in comparison with the DFL one. The SDFL is being really close to the DFL in term of resource usage. The CPU usage is being … view at source ↗
Figure 8
Figure 8. Figure 8: Accuracy average in all scenarios: (a) CFL (b) DFL (c) SDFL. In CFL, increasing the number of participants in the network leads to a decrease in [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
read the original abstract

Federated Learning (FL) has emerged as a promising paradigm for collaborative model training across distributed edge devices while preserving data privacy especially with the huge increase amount of data due to the adoption of technologies which contributes to the growing number of IoT devices. Storing this amount of data centrally is challenging due to issues like limited communication, privacy, and regulations. FL can be Centralized (CFL), Decentralized (DFL), and Semi-decentralized (SDFL). Choosing the right FL architecture depends on the application's needs. However, very few research studies have experimentally compared these three types of architectures to not only understand the respective strengths and limitations, but also trade-offs between different performance indicators. This paper overcome this lack of analysis, conducting experimental analyses using the Fedstellar simulator, MNIST dataset, and MLP classifier.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper claims to address a gap in the literature by experimentally comparing Centralized Federated Learning (CFL), Decentralized Federated Learning (DFL), and Semi-decentralized Federated Learning (SDFL) architectures. It conducts this analysis using the Fedstellar simulator on the MNIST dataset with an MLP classifier to identify respective strengths, limitations, and trade-offs across performance indicators such as accuracy and efficiency.

Significance. If the results hold under broader conditions, the work would provide useful empirical guidance for selecting among CFL, DFL, and SDFL based on application-specific trade-offs. The reliance on a public simulator and standard dataset is a positive step toward reproducibility.

major comments (2)
  1. [§4 (Experimental Setup)] §4 (Experimental Setup): The central claim that the experiments reveal general performance trade-offs among the three architectures rests on the assumption that MNIST + MLP + Fedstellar produces representative differences. However, MNIST is homogeneous and low-complexity while MLP is shallow; this minimizes data heterogeneity, model size, and communication bottlenecks that dominate real FL deployments, so observed rankings or curves may be artifacts of the narrow regime rather than intrinsic architectural properties.
  2. [Results section] Results section (e.g., accuracy and communication tables): Without reported statistical significance, error bars, or multiple random seeds, it is unclear whether the measured differences between CFL/DFL/SDFL exceed experimental noise, weakening the trade-off conclusions.
minor comments (2)
  1. [Abstract] Abstract: 'huge increase amount of data' is grammatically incomplete and should read 'huge increase in the amount of data'.
  2. [Abstract] Abstract: 'This paper overcome this lack of analysis' should be 'This paper overcomes this lack of analysis'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below and indicate the changes we will make in the revised version.

read point-by-point responses

  1. Referee: [§4 (Experimental Setup)] §4 (Experimental Setup): The central claim that the experiments reveal general performance trade-offs among the three architectures rests on the assumption that MNIST + MLP + Fedstellar produces representative differences. However, MNIST is homogeneous and low-complexity while MLP is shallow; this minimizes data heterogeneity, model size, and communication bottlenecks that dominate real FL deployments, so observed rankings or curves may be artifacts of the narrow regime rather than intrinsic architectural properties.

    Authors: We selected MNIST and the MLP architecture to establish a controlled, reproducible baseline using the public Fedstellar simulator, as the referee correctly notes is a positive step for reproducibility. This choice isolates the effects of the three FL architectures (CFL, DFL, SDFL) from confounding factors such as data heterogeneity or large model sizes, allowing clearer attribution of observed trade-offs in accuracy and efficiency. While we acknowledge that real deployments often involve greater heterogeneity and complexity, the current results provide initial empirical guidance in a standardized setting. In the revised manuscript we will expand Section 4 and add a dedicated limitations paragraph to explicitly discuss the scope of the present experiments and outline extensions to more complex datasets and models. revision: partial

  2. Referee: Results section (e.g., accuracy and communication tables): Without reported statistical significance, error bars, or multiple random seeds, it is unclear whether the measured differences between CFL/DFL/SDFL exceed experimental noise, weakening the trade-off conclusions.

    Authors: We agree that reporting variability and statistical significance would strengthen the reliability of the trade-off conclusions. In the revised manuscript we will rerun all experiments with multiple random seeds, report mean performance together with standard deviation (error bars) in tables and figures, and include appropriate statistical tests to confirm that observed differences between architectures are significant. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical comparison using external simulator and public dataset

full rationale

The paper conducts an experimental analysis comparing CFL, DFL, and SDFL architectures via the Fedstellar simulator on the MNIST dataset with an MLP classifier. No equations, derivations, fitted parameters renamed as predictions, or load-bearing self-citations appear in the text. The central claim of filling a research gap rests on running standard simulations and reporting observed performance trade-offs, which is self-contained against external benchmarks and does not reduce to any self-definition or input-by-construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper rests on standard federated learning domain assumptions about privacy preservation through model-update sharing only; no free parameters, invented entities, or ad-hoc axioms are indicated in the abstract.

axioms (1)
  • domain assumption Federated learning preserves data privacy by sharing only model updates rather than raw data.
    Core premise stated in the abstract when describing FL motivation and challenges with central storage.

pith-pipeline@v0.9.0 · 5672 in / 1318 out tokens · 67390 ms · 2026-05-20T20:10:34.308336+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

31 extracted references · 31 canonical work pages · 3 internal anchors

  1. [1]

    ”Federated learning: Challenges, methods, and future directions.” IEEE signal processing magazine 37.3 (2020): 50-60

    Li, Tian, et al. ”Federated learning: Challenges, methods, and future directions.” IEEE signal processing magazine 37.3 (2020): 50-60

    work page 2020

  2. [2]

    Zhang, Jiale, et al. ”Poisoning attack in federated learning using generative adversarial nets.” 2019 18th IEEE international conference on trust, security and privacy in computing and communications/13th IEEE international conference on big data science and engineering (TrustCom/BigDataSE). IEEE, 2019

    work page 2019

  3. [3]

    ”Advances and open problems in federated learn- ing.” Foundations and trends® in machine learning 14.1–2 (2021): 1- 210

    Kairouz, Peter, et al. ”Advances and open problems in federated learn- ing.” Foundations and trends® in machine learning 14.1–2 (2021): 1- 210

    work page 2021

  4. [4]

    ”Fully decentralized federated learning.” Third workshop on bayesian deep learning (NeurIPS)

    Lalitha, Anusha, et al. ”Fully decentralized federated learning.” Third workshop on bayesian deep learning (NeurIPS). V ol. 2. 2018

    work page 2018

  5. [5]

    ”Decentralized federated learn- ing: Fundamentals, state of the art, frameworks, trends, and challenges.” IEEE Communications Surveys & Tutorials (2023)

    Beltr ´an, Enrique Tom ´as Mart´ınez, et al. ”Decentralized federated learn- ing: Fundamentals, state of the art, frameworks, trends, and challenges.” IEEE Communications Surveys & Tutorials (2023)

    work page 2023

  6. [6]

    ”Decentralized federated learning: A survey and perspective.” arXiv preprint arXiv:2306.01603 (2023)

    Yuan, Liangqi, et al. ”Decentralized federated learning: A survey and perspective.” arXiv preprint arXiv:2306.01603 (2023)

    work page arXiv 2023

  7. [7]

    ”Evaluation of machine learning algorithms for anomaly detection.” 2020 international conference on cyber security and protection of digital services (cyber security)

    Elmrabit, Nebrase, et al. ”Evaluation of machine learning algorithms for anomaly detection.” 2020 international conference on cyber security and protection of digital services (cyber security). IEEE, 2020

    work page 2020

  8. [8]

    ”Approches d’apprentissage automatique pour la d ´etection du Spam Web: exploration de diverses caract ´eristiques.” (2018)

    Ait Mahammed, Fatima. ”Approches d’apprentissage automatique pour la d ´etection du Spam Web: exploration de diverses caract ´eristiques.” (2018)

    work page 2018

  9. [9]

    & Kl ´en, R

    Rainio, O., Teuho, J. & Kl ´en, R. Evaluation metrics and statistical tests for machine learning. Sci Rep 14, 6086 (2024)

    work page 2024

  10. [10]

    Taming the cross entropy loss

    Martinez, M., & Stiefelhagen, R. Taming the cross entropy loss. In Pattern Recognition: 40th German Conference, GCPR 2018, Stuttgart, Germany, October 9-12, 2018, Proceedings 628–637, V ol. 40. Springer (2019)

    work page 2018

  11. [11]

    ”Fedstellar: A platform for decentralized federated learning.” Expert Systems with Applications 242 (2024): 122861

    Beltr ´an, Enrique Tom ´as Mart ´ınez, et al. ”Fedstellar: A platform for decentralized federated learning.” Expert Systems with Applications 242 (2024): 122861

    work page 2024

  12. [12]

    ”Multilayer perceptron and neural networks.” WSEAS Transactions on Circuits and Systems 8.7 (2009): 579-588

    Popescu, Marius-Constantin, et al. ”Multilayer perceptron and neural networks.” WSEAS Transactions on Circuits and Systems 8.7 (2009): 579-588

    work page 2009

  13. [13]

    Federated Learning: Strategies for Improving Communication Efficiency

    Konecn ´y, Jakub, et al. ”Federated learning: Strategies for improving communication efficiency.” arXiv preprint arXiv:1610.05492 8 (2016)

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  14. [14]

    Towards Federated Learning at Scale: System Design

    Bonawitz, K., Eichner, H., Grieskamp, W., et al.: Towards federated learning at scale: System design. arXiv preprint arXiv:1902.01046 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 1902

  15. [15]

    In: Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security

    Bonawitz, K., Ivanov, V ., Kreuter, B., Marcedone, A., McMahan, H.B., Patel, S., Ramage, D., Segal, A., Seth, K.: Practical secure aggregation for privacy-preserving machine learning. In: Proceedings of the 2017 ACM SIGSAC Conference on Computer and Communications Security. pp. 1175–1191. ACM (2017)

    work page 2017

  16. [16]

    Federated learning: Challenges, methods, and future directions,

    T. Li, A. K. Sahu, A. Talwalkar, and V . Smith, “Federated learning: Challenges, methods, and future directions,” IEEE Signal Processing Magazine, vol. 37, no. 3, pp. 50–60, 2020

    work page 2020

  17. [17]

    Federated learning in mobile edge networks: A comprehensive survey,

    W. Y . B. Lim et al., “Federated learning in mobile edge networks: A comprehensive survey,” IEEE Communications Surveys & Tutorials, vol. 22, no. 3, pp. 2031–2063, 2020

    work page 2031

  18. [18]

    Federated learning for Internet of Things: A comprehensive survey,

    D. C. Nguyen, M. Ding, P. N. Pathirana, A. Seneviratne, J. Li, and H. V . Poor, “Federated learning for Internet of Things: A comprehensive survey,” IEEE Communications Surveys & Tutorials, vol. 23, no. 3, pp. 1622–1658, 2021

    work page 2021

  19. [19]

    Federated learning for Internet of Things: Recent advances, taxonomy, and open challenges,

    L. U. Khan, W. Saad, Z. Han, E. Hossain, and C. S. Hong, “Federated learning for Internet of Things: Recent advances, taxonomy, and open challenges,” IEEE Communications Surveys & Tutorials, vol. 23, no. 3, pp. 1759–1799, 2021

    work page 2021

  20. [20]

    A survey on security and privacy of federated learning,

    V . Mothukuri, R. M. Parizi, S. Pouriyeh, Y . Huang, A. Dehghantanha, and G. Srivastava, “A survey on security and privacy of federated learning,” Future Generation Computer Systems, vol. 115, pp. 619– 640, 2021

    work page 2021

  21. [21]

    Fusion of federated learning and industrial Internet of Things: A survey,

    P. Boobalan et al., “Fusion of federated learning and industrial Internet of Things: A survey,” Computer Networks, vol. 212, p. 109048, 2022

    work page 2022

  22. [22]

    ”Federated learning for healthcare domain-pipeline, applications and challenges.” ACM Transactions on Computing for Healthcare 3.4 (2022): 1-36

    Joshi, Madhura, Ankit Pal, and Malaikannan Sankarasubbu. ”Federated learning for healthcare domain-pipeline, applications and challenges.” ACM Transactions on Computing for Healthcare 3.4 (2022): 1-36

    work page 2022

  23. [23]

    Decentral and incentivized federated learning frameworks: A systematic literature review,

    L. Witt, M. Heyer, K. Toyoda, W. Samek, and D. Li, “Decentral and incentivized federated learning frameworks: A systematic literature review,” arXiv preprint arXiv:2205.07855, 2022

    work page arXiv 2022

  24. [24]

    ”Decentralized privacy using blockchain-enabled federated learning in fog computing.” IEEE Internet of Things Journal 7.6 (2020): 5171-5183

    Qu, Youyang, et al. ”Decentralized privacy using blockchain-enabled federated learning in fog computing.” IEEE Internet of Things Journal 7.6 (2020): 5171-5183

    work page 2020

  25. [25]

    A systematic literature review on blockchain enabled federated learning framework for Internet of Vehicles,

    M. Billah, S. T. Mehedi, A. Anwar, Z. Rahman, and R. Islam, “A systematic literature review on blockchain enabled federated learning framework for Internet of Vehicles,” arXiv preprint arXiv:2203.05192, 2022

    work page arXiv 2022

  26. [26]

    Survey on federated-learning approaches in distributed environment,

    R. Gupta and T. Alam, “Survey on federated-learning approaches in distributed environment,” Wireless Personal Communications, Mar 2022

    work page 2022

  27. [27]

    Blockchain-based federated learning in UA Vs beyond 5G networks: A solution taxonomy and future directions,

    D. Saraswat et al., “Blockchain-based federated learning in UA Vs beyond 5G networks: A solution taxonomy and future directions,” IEEE Access, vol. 10, pp. 33 154–33 182, 2022

    work page 2022

  28. [28]

    ”Blockchain-enabled federated learning: A survey.” ACM Computing Surveys 55.4 (2022): 1-35

    Qu, Youyang, et al. ”Blockchain-enabled federated learning: A survey.” ACM Computing Surveys 55.4 (2022): 1-35

    work page 2022

  29. [29]

    ”Semi-decentralized federated learning with collaborative relaying.” 2022 IEEE International Symposium on Infor- mation Theory (ISIT)

    Yemini, Michal, et al. ”Semi-decentralized federated learning with collaborative relaying.” 2022 IEEE International Symposium on Infor- mation Theory (ISIT). IEEE, 2022

    work page 2022

  30. [30]

    BrainTorrent: A Peer-to-Peer Environment for Decentralized Federated Learning

    Roy, Abhijit Guha, et al. ”Braintorrent: A peer-to-peer environment for decentralized federated learning.” arXiv preprint arXiv:1905.06731 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 1905

  31. [31]

    ”Trustworthy artificial intelligence: a review.” ACM computing surveys (CSUR) 55.2 (2022): 1-38

    Kaur, Davinder, et al. ”Trustworthy artificial intelligence: a review.” ACM computing surveys (CSUR) 55.2 (2022): 1-38

    work page 2022