pith. sign in

arxiv: 2605.30336 · v1 · pith:NH5ASQYZnew · submitted 2026-05-28 · 💻 cs.LG

Fairness-Aware Federated Learning with Trajectory Shapley Value

Daniel Kuznetsov , Ziqi Wang This is my paper

Pith reviewed 2026-06-29 08:59 UTC · model grok-4.3

classification 💻 cs.LG
keywords federated learningtrajectory shapley valueadaptive aggregationclient contributionfairnessdynamic weightingoptimization trajectory
0
0 comments X

The pith

Trajectory Shapley Value measures each client's influence on the global model's optimization path to set dynamic aggregation weights in federated learning.

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

The paper proposes the Trajectory Shapley Value as a metric to assess how each client affects the sequence of global model updates during federated training. It replaces fixed weights with per-round dynamic weights computed from a validation-based utility that remains consistent across time steps. This design lets the server respond to clients that contribute unevenly or act adversarially without requiring prior knowledge of their data distributions. A reader would care because conventional aggregation can bias the shared model toward whichever clients happen to participate more or send larger updates at any moment. If the approach holds, training becomes both faster to converge and less prone to instability from heterogeneous participation.

Core claim

The central claim is that client influence in federated learning can be quantified by the change each client's update produces along the entire optimization trajectory, rather than by a single-round or fixed contribution score. The Trajectory Shapley Value uses a temporally consistent validation utility to compute these influences. FedTSV then converts the resulting values into dynamic client weights for the server aggregation step, allowing real-time adjustment to heterogeneous and adversarial participation patterns.

What carries the argument

Trajectory Shapley Value (TSV), a contribution metric that evaluates how each client influences the optimization trajectory of the global model using a validation-based, temporally consistent utility.

If this is right

  • FedTSV accelerates convergence on standard benchmark datasets compared with fixed-weight schemes.
  • The method improves robustness when some clients participate adversarially or drop out unpredictably.
  • Contribution assessments become more equitable across clients with differing data volumes or qualities.
  • The framework supplies a principled basis for fairness-aware aggregation rules in distributed training.

Where Pith is reading between the lines

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

  • TSV weights could be cached across rounds to reduce computation when client sets change slowly.
  • The same trajectory-based scoring might apply to measuring influence in decentralized optimization beyond the federated server-client setting.
  • Combining TSV with explicit privacy mechanisms could quantify the fairness cost of noise injection.

Load-bearing premise

A validation-based, temporally consistent utility function can accurately and unbiasedly quantify each client's influence on the global optimization trajectory.

What would settle it

In a controlled experiment with one client known to send updates that slow convergence, the TSV-derived weights fail to reduce that client's aggregation share relative to uniform averaging while still reaching the same final accuracy.

Figures

Figures reproduced from arXiv: 2605.30336 by Daniel Kuznetsov, Ziqi Wang.

Figure 1
Figure 1. Figure 1: Performance comparison on CIFAR-10 and MNIST. [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Client-contribution comparison on CIFAR-10. Green: [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
read the original abstract

Federated learning is an emerging distributed paradigm that addresses the challenges posed by heterogeneous, privacy-sensitive data. It enables multiple clients to train a model collaboratively by aggregating their local updates at a server. However, conventional aggregation schemes typically use fixed weights that fail to reflect unequal and time-varying client contributions, leading to biased and unstable learning. To improve fairness and stability, we propose the Trajectory Shapley Value (TSV), a contribution metric that evaluates how each client influences the optimization trajectory of the global model using a validation-based, temporally consistent utility. Building on TSV, we design FedTSV, an adaptive aggregation method that converts per-round evaluations into dynamic client weights, allowing the server to respond to heterogeneous and adversarial participation in real time. Experiments on benchmark datasets show that FedTSV accelerates convergence, improves robustness, and yields more equitable contribution assessments, thereby providing a principled foundation for fairness-aware federated optimization.

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

1 major / 0 minor

Summary. The paper proposes Trajectory Shapley Value (TSV), a contribution metric that evaluates each client's influence on the global model's optimization trajectory via a validation-based, temporally consistent utility function. It introduces FedTSV, an adaptive aggregation scheme that converts per-round TSV scores into dynamic client weights to handle heterogeneous and adversarial participation. Experiments on benchmark datasets are reported to demonstrate accelerated convergence, improved robustness to adversaries, and more equitable contribution assessments compared to conventional fixed-weight aggregation.

Significance. If the TSV utility is shown to be an unbiased estimator of marginal trajectory influence and FedTSV yields stable fairness improvements without degrading accuracy, the work would supply a concrete, per-round mechanism for fairness-aware aggregation in non-IID federated learning. The absence of derivations, proofs, or detailed error analysis in the provided abstract limits assessment of whether these gains are parameter-free or merely reparameterized.

major comments (1)
  1. [Abstract] Abstract: the central claim that the validation-based utility 'accurately and unbiasedly quantifies each client's influence' is load-bearing for both the TSV definition and the fairness guarantees of FedTSV, yet the abstract supplies no argument that a fixed server validation distribution remains representative of the optimization trajectory's effect on heterogeneous client distributions. Under non-IID data this choice risks systematic bias in the resulting dynamic weights.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for highlighting a key assumption in our abstract. We agree the claim about unbiased quantification merits qualification and will revise the manuscript to address representativeness of the validation distribution under non-IID conditions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the validation-based utility 'accurately and unbiasedly quantifies each client's influence' is load-bearing for both the TSV definition and the fairness guarantees of FedTSV, yet the abstract supplies no argument that a fixed server validation distribution remains representative of the optimization trajectory's effect on heterogeneous client distributions. Under non-IID data this choice risks systematic bias in the resulting dynamic weights.

    Authors: We acknowledge the point: the abstract's phrasing is stronger than the supporting argument provided there. The full manuscript motivates the validation set as a proxy for trajectory progress (Section 3.2), but does not derive unbiasedness or prove representativeness under arbitrary non-IID partitions. In the revision we will (i) replace the absolute claim in the abstract with 'approximates client influence via a temporally consistent validation utility,' (ii) add a dedicated paragraph in Section 3.3 discussing the fixed validation assumption and its potential bias, and (iii) include an additional experiment varying the validation distribution to quantify sensitivity. These changes will make the limitations explicit while preserving the empirical evidence that FedTSV improves fairness metrics on the reported benchmarks. revision: yes

Circularity Check

0 steps flagged

No circularity; TSV utility defined externally via validation set

full rationale

The derivation introduces TSV as a Shapley-style metric computed from a validation-based, temporally consistent utility function that is independent of the resulting client weights. FedTSV then applies these pre-computed values as dynamic aggregation coefficients. No equation reduces the claimed influence scores or convergence improvements to a fitted parameter or self-referential definition; the utility remains an external input. No self-citations are load-bearing in the provided text, and the central construction does not collapse to renaming or tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Abstract-only review; the central claim rests on the domain assumption that validation-based utilities can serve as temporally consistent contribution measures. TSV is introduced as a new entity without independent evidence supplied in the abstract.

axioms (1)
  • domain assumption A validation-based utility function can provide a temporally consistent measure of client contribution to the optimization trajectory.
    This premise directly supports the definition of TSV in the abstract.
invented entities (1)
  • Trajectory Shapley Value (TSV) no independent evidence
    purpose: Contribution metric that evaluates client influence on the global model's optimization trajectory
    Newly defined in the paper; no external validation or prior reference supplied in the abstract.

pith-pipeline@v0.9.1-grok · 5679 in / 1308 out tokens · 24645 ms · 2026-06-29T08:59:39.726393+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

26 extracted references · 5 canonical work pages · 1 internal anchor

  1. [1]

    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

    2020

  2. [2]

    Advances and open problems in federated learning,

    P. Kairouz, H. B. McMahanet al., “Advances and open problems in federated learning,”Foundations and Trends® in Machine Learning, vol. 14, no. 1–2, pp. 1–210, 2021

    2021

  3. [3]

    Hierarchical split federated learning: Convergence analysis and sys- tem optimization,

    Z. Lin, W. Wei, Z. Chen, C.-T. Lam, X. Chen, Y . Gao, and J. Luo, “Hierarchical split federated learning: Convergence analysis and sys- tem optimization,”IEEE Transactions on Mobile Computing, vol. 24, no. 10, pp. 9352–9367, 2025

    2025

  4. [4]

    Federated multiagent deep reinforcement learning approach via physics-informed reward for multimicrogrid energy management,

    Y . Li, S. He, Y . Li, Y . Shi, and Z. Zeng, “Federated multiagent deep reinforcement learning approach via physics-informed reward for multimicrogrid energy management,”IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 5, pp. 5902–5914, 2024

    2024

  5. [5]

    Challenges and directions in computational fluid–structure interaction,

    Y . Bazilevs, K. Takizawa, and T. E. Tezduyar, “Challenges and directions in computational fluid–structure interaction,”Mathematical Models and Methods in Applied Sciences, vol. 23, no. 02, pp. 215–221, 2013

    2013

  6. [6]

    A review of modelling techniques for floating offshore wind turbines,

    A. Otter, J. Murphy, V . Pakrashi, A. Robertson, and C. Desmond, “A review of modelling techniques for floating offshore wind turbines,” Wind Energy, vol. 25, no. 5, pp. 831–857, 2022

    2022

  7. [7]

    A comprehensive review of wind power prediction based on machine learning: Models, applications, and challenges,

    Z. Liu, H. Guo, Y . Zhang, and Z. Zuo, “A comprehensive review of wind power prediction based on machine learning: Models, applications, and challenges,”Energies, vol. 18, no. 2, 2025. [Online]. Available: https://www.mdpi.com/1996-1073/18/2/350

    2025

  8. [8]

    Communication-efficient learning of deep networks from decentral- ized data,

    B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentral- ized data,” inProceedings of the 20th International Conference on Artificial Intelligence and Statistics. PMLR, Apr. 2017, pp. 1273– 1282

    2017

  9. [9]

    SCAFFOLD: Stochastic controlled averaging for federated learning,

    S. P. Karimireddy, S. Kale, M. Mohri, S. Reddi, S. Stich, and A. T. Suresh, “SCAFFOLD: Stochastic controlled averaging for federated learning,” inProceedings of the 37th International Conference on Machine Learning, ser. Proceedings of Machine Learning Research, H. D. III and A. Singh, Eds., vol. 119. PMLR, 13–18 Jul 2020, pp. 5132–5143

    2020

  10. [10]

    FedADMM-InSa: An inexact and self-adaptive ADMM for federated learning,

    Y . Song, Z. Wang, and E. Zuazua, “FedADMM-InSa: An inexact and self-adaptive ADMM for federated learning,”Neural Networks, vol. 181, p. 106772, Jan. 2025. [Online]. Available: https: //www.sciencedirect.com/science/article/pii/S0893608024006968

    2025

  11. [11]

    Beyond ADMM: A unified client-variance-reduced adaptive federated learning framework,

    S. Wang, Y . Xu, Z. Wang, T.-H. Chang, T. Q. S. Quek, and D. Sun, “Beyond ADMM: A unified client-variance-reduced adaptive federated learning framework,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 8, 2023, pp. 10 175–10 183

    2023

  12. [12]

    Notes on the N-Person Game — II: The Value of an N- Person Game,

    L. S. Shapley, “Notes on the N-Person Game — II: The Value of an N- Person Game,” RAND Corporation, Tech. Rep. RM-670, Aug. 1951. [Online]. Available: https://www.rand.org/pubs/research memoranda/ RM0670.html

    1951

  13. [13]

    Measure contribution of participants in federated learning,

    G. Wang, C. X. Dang, and Z. Zhou, “Measure contribution of participants in federated learning,” in2019 IEEE International Conference on Big Data (Big Data). IEEE, 2019, pp. 2597–2604. [Online]. Available: https://ieeexplore.ieee.org/document/9006179

    work page arXiv 2019

  14. [14]

    On the volatility of shapley-based contribution metrics in federated learning,

    A. Geimer, B. Fiz, and R. State, “On the volatility of shapley-based contribution metrics in federated learning,” in2025 International Joint Conference on Neural Networks (IJCNN). IEEE, 2025, pp. 1–8

    2025

  15. [15]

    Gradient driven rewards to guarantee fairness in collaborative machine learn- ing,

    X. Xu, L. Lyu, X. Ma, C. Miao, C. S. Foo, and B. K. H. Low, “Gradient driven rewards to guarantee fairness in collaborative machine learn- ing,” inAdvances in Neural Information Processing Systems, vol. 34. Curran Associates, Inc., 2021, pp. 16 104–16 117

    2021

  16. [16]

    A multi-objective optimization framework for decentralized learning with coordination constraints,

    R. Morales and U. Biccari, “A multi-objective optimization framework for decentralized learning with coordination constraints,” Jul. 2025, arXiv:2507.13983

    work page arXiv 2025

  17. [17]

    GTG-Shapley: Efficient and accurate participant contribution evaluation in federated learning,

    Z. Liu, Y . Chen, H. Yu, Y . Liu, and L. Cui, “GTG-Shapley: Efficient and accurate participant contribution evaluation in federated learning,” ACM Transactions on Intelligent Systems and Technology, vol. 13, no. 4, pp. 1–21, 2022

    2022

  18. [18]

    ShapleyFL: Robust federated learning based on shapley value,

    Q. Sun, X. Li, J. Zhang, L. Xiong, W. Liu, J. Liu, Z. Qin, and K. Ren, “ShapleyFL: Robust federated learning based on shapley value,” in Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, ser. KDD ’23. New York, NY , USA: Association for Computing Machinery, 2023, p. 2096–2108. [Online]. Available: https://doi.org/10.114...

    work page doi:10.1145/3580305.3599500 2023

  19. [19]

    Peters,Game Theory

    H. Peters,Game Theory. Berlin, Heidelberg: Springer, 2008. [Online]. Available: http://link.springer.com/10.1007/978-3-540-69291-1

    work page doi:10.1007/978-3-540-69291-1 2008

  20. [20]

    Efficient sampling approaches to shapley value approximation,

    J. Zhang, Q. Sun, J. Liu, L. Xiong, J. Pei, and K. Ren, “Efficient sampling approaches to shapley value approximation,”Proc. ACM Manag. Data, vol. 1, no. 1, May 2023

    2023

  21. [21]

    Towards efficient data valuation based on the shapley value,

    R. Jia, D. Dao, B. Wang, F. A. Hubis, N. Hynes, N. M. G ¨urel, B. Li, C. Zhang, D. Song, and C. J. Spanos, “Towards efficient data valuation based on the shapley value,” inProceedings of the Twenty- Second International Conference on Artificial Intelligence and Statis- tics, ser. Proceedings of Machine Learning Research, K. Chaudhuri and M. Sugiyama, Eds....

    2019

  22. [22]

    How to backdoor federated learning,

    E. Bagdasaryan, A. Veit, Y . Hua, D. Estrin, and V . Shmatikov, “How to backdoor federated learning,” inProceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics, ser. Proceedings of Machine Learning Research, S. Chiappa and R. Calandra, Eds., vol. 108. PMLR, 26–28 Aug 2020, pp. 2938– 2948

    2020

  23. [23]

    Gradient-based learning applied to document recognition,

    Y . LeCun, L. Bottou, Y . Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,”Proceedings of the IEEE, vol. 86, no. 11, pp. 2278–2324, 1998

    1998

  24. [24]

    Learning multiple layers of features from tiny images,

    A. Krizhevsky and G. Hinton, “Learning multiple layers of features from tiny images,” University of Toronto, Tech. Rep. TR- 2009, 2009. [Online]. Available: https://www.cs.toronto.edu/ ∼kriz/ learning-features-2009-TR.pdf

    2009

  25. [25]

    Measuring the effects of non-identical data distribution for federated visual classification,

    T.-M. H. Hsu, H. Qi, and M. Brown, “Measuring the effects of non-identical data distribution for federated visual classification,”

  26. [26]

    Measuring the Effects of Non-Identical Data Distribution for Federated Visual Classification

    [Online]. Available: https://arxiv.org/abs/1909.06335 6

    work page internal anchor Pith review Pith/arXiv arXiv 1909