SWTA-Shapley：一种高效的联邦学习贡献评估方法

鲍世豪; 倪郑威

doi:10.11959/j.issn.1000-0801.2025211

您当前的位置：

首页 >

文章列表页 >

SWTA-Shapley：一种高效的联邦学习贡献评估方法

研究与开发 | 更新时间：2026-01-08

- SWTA-Shapley：一种高效的联邦学习贡献评估方法
- SWTA-Shapley: an efficient contribution evaluation method for federated learning
- 电信科学 2025年41卷第12期页码：146-163
- 作者机构：
  
  浙江工商大学信息与电子工程学院，浙江杭州 310018
- 作者简介：
  
  [ "鲍世豪（2000− ），男，浙江工商大学信息与电子工程学院硕士生，主要研究方向为联邦学习。" ]
  [ "倪郑威（1989- ），男，博士，浙江工商大学信息与电子工程学院副研究员，主要研究方向为机器学习、物联网、无线通信等。" ]
- 基金信息：
  
  桐乡市通用人工智能研究院项目(TAGI2-B-2024-0017)
- DOI：10.11959/j.issn.1000-0801.2025211
  中图分类号： TP30;TN91
- 收稿：2025-03-17，
  
  修回：2025-04-12，
  
  录用：2025-05-26，
  
  纸质出版：2025-12-20
- 稿件说明：
移动端阅览
鲍世豪,倪郑威.SWTA-Shapley：一种高效的联邦学习贡献评估方法[J].电信科学,2025,41(12):146-163.

BAO Shihao,NI Zhengwei.SWTA-Shapley: an efficient contribution evaluation method for federated learning[J].Telecommunications Science,2025,41(12):146-163.
鲍世豪,倪郑威.SWTA-Shapley：一种高效的联邦学习贡献评估方法[J].电信科学,2025,41(12):146-163. DOI： 10.11959/j.issn.1000-0801.2025211.

BAO Shihao,NI Zhengwei.SWTA-Shapley: an efficient contribution evaluation method for federated learning[J].Telecommunications Science,2025,41(12):146-163. DOI： 10.11959/j.issn.1000-0801.2025211.

摘要

联邦学习很好地解决了当数据隐私保护导致的“数据孤岛”问题，为了维持联邦学习系统的长期运行，需要通过适当的激励计划来吸引高质量的数据拥有者参与联邦学习训练。如何公平评估参与者对最终联邦学习模型性能的贡献是研究的重点。Shapley值是贡献评估的重要方法，已经被广泛应用。然而，Shapley值的计算复杂度高，这使Shapley值很难应用到现实的场景中。现有许多近似Shapley值算法，如引导截断梯度Shapley（guided trunction gradient Shapley，GTG-Shapley），结合了引导随机抽样、梯度重建子模型、截断3种机制，能够降低计算复杂度，得到接近准确的近似Shapley值。这样的近似算法使Shapley值在现实场景中的应用变成可能，但是GTG-Shapley依旧存在许多不足，所以基于GTG-Shapley提出了一种新的近似Shapley值算法——分层加权截断与自适应规划Shapley（stratified weighted truncation and adaptive programming Shapley，SWTA-Shapley），该算法将GTG-Shapley中的引导随机抽样换成了分层组合的方法，并将截断优化为加权截断，同时为了进一步提高准确率，采用自适应规划的方法充分利用了被截断部分的效用值。实验结果表明，相较于GTG-Shapley，SWTA-Shapley在多种场景中都能取得更高的计算效率，更适用于现实的场景。

Abstract

Federated learning has effectively addressed the “data silo” issue caused by data privacy protection. To maintain the long-term operation of federated learning systems

it is necessary to attract high-quality data owners to participate in federated learning training through appropriate incentive programs. The focus of researches is on how to fairly evaluate the contribution of participants to the performance of the final federated learning model. The Shapley value is an important method for contribution assessment and has been widely applied. However

the computational complexity of Shapley values is particularly high

making it difficult to apply Shapley value in real-world scenarios. Many approximate Shapley value algorithms

such as guided trunction gradient Shapley (GTG-Shapley) that combines guided random sampling

gradient reconstruction of sub-models

and truncation mechanisms

can reduce computational complexity and yield approximate Shapley values that are very close to the accurate values. Such approximation algorithms make the application of Shapley value in real-world scenarios possible

but GTG-Shapley still has many shortcomings. Therefore

based on GTG-Shapley

a new approximate Shapley value algorithm called stratified weighted truncation and adaptive programming Shapley (SWTA-Shapley) was proposed. This algorithm replaced the guided random sampling in GTG-Shapley with a stratified combination method and optimized truncation to weighted truncation. At the same time

to further improve accuracy

adaptive programming method was combined to fully utilize the utility values of the truncated parts. Extensive experimental results show that

compared with GTG-Shapley

SWTA-Shapley can achieve higher computational efficiency in various scenarios and is more applicable to real-world situations.

关键词

Keywords

references

HEMSLEY M . Why the general data protection regulation is likely to disrupt core digital marketing channels in Europe [J ] . Journal of Digital & Social Media Marketing , 2018 , 6 ( 2 ): 137 - 142 .

MCMAHAN B , MOORE E , RAMAGE D , et al . Communication-efficient learning of deep networks from decentralized data [C ] // Proceeding of the Int . Symp . on Artificial Intelligence and Statistics (AISTATS) . 2017 , arXiv: 1602 . 05629 .

YANG Q , LIU Y , CHENG Y , et al . Federated learning [M ] . Heidelberg : Springer , 2019 .

WANG T H , RAUSCH T , ZHANG C , et al . A principled approach to data valuation for federated learning [J ] . Federated Learning: Privacy and Incentive , 2020 : 153 - 167 .

SHI Y Y , SONG H J , XU J . Responsible and Effective federated learning in financial services: a comprehensive survey [C ] // Proceedings of the 2023 62nd IEEE Conference on Decision and Control (CDC) . Singapore : IEEE Press , 2023 : 4229 - 4236 .

LYU X C , HOU X Y , CHEN S , et al . Secure and efficient federated learning with provable performance guarantees via stochastic quantization [J ] . IEEE Transactions on Information Forensics and Security , 2024 , 19 : 4070 - 4085 .

JIA R , DAO DAVID , WANG B X , et al . Towards efficient data valuation based on the Shapley value [C ] // Proceedings of the 22nd International Conference on Artificial Intelligence and Statistics . 2019 , arXiv : 1902 . 10275 .

SHAPLEY L S . A value for n-person games [J ] . Contributions to the Theory of Games , 1953 , 307 – 317 .

GUL F . Bargaining foundations of Shapley value [J ] . Econometrica: Journal of the Econometric Society , 1989 , 57 ( 1 ): 81 - 95 .

AUMANN R J . Economic applications of the shapley value [M ] // Game-Theoretic Methods in General Equilibrium Analysis . Dordrecht : Springer , 1994 : 121 - 133 .

ALGABA E , FRAGNELLI V , SANCHEZ-SORIANO . Handbook of the shapley value [J ] . Chapman and Hall/CRC , 2019 .

MICHALAK T P , RAHWAN T , JENNINGS N R , Szczepański P L , et al . Computational analysis of connectivity games with applications to the investigation of terrorist networks [C ] // Proceedings of the Twenty-Third International Joint Conference on Artificial Intelligence (IJCAI '13) . Palo Alto : AAAI Press , 2013 : 293 - 301 .

LINDELAUF R , HAMERS H , HUSSLAGE B . Cooperative game theoretic centrality analysis of terrorist networks: the cases of Jemaah Islamiyah and al-Qaeda [J ] . European Journal of Operational Research , 2013 , 229 ( 1 ): 230 - 238 .

PETROSJAN L , ZACCOUR G . Time-consistent Shapley value allocation of pollution cost reduction [J ] . Journal of Economic Dynamics and Control , 2003 , 27 ( 3 ): 381 - 398 .

XU Z W , PENG Z X , YANG L , et al . An improved Shapley Value Method For A Green Supply Chain Income Distribution Mechanism [J ] . International Journal of Environmental Research and Public Health , 2018 , 15 ( 9 ): 1976 .

LIAO Z L , ZHU X L . Case study on initial allocation of Shanghai carbon emission trading based on Shapley value [J ] . Journal of Cleaner Production , 2015 , 103 : 338 - 344 .

COHEN S , RUPPIN E , DROR G . Feature selection based on the Shapley value [C ] // Proceedings of the 19th International Joint Conference on Artificial Intelligence . 2005 .

ROZEMBERCZKI B , WATSON L , BAYER P , et al . The Shapley value in machine learning [J ] . arXiv preprint , arXiv: 2202.05594 , 2022 .

CHEN H , IAN C , SCOTT M , et al . Algorithms to estimate Shapley value feature attributions [J ] . Nature Machine Intelligence , 2022 , 5 : 590 - 601 .

CASTRO J , GÓMEZ D , TEJADA J . Polynomial calculation of the Shapley value based on sampling [J ] . Computers and Operations Research , 2008 , 36 ( 5 ): 1726 - 1730 .

GHORBANI A , ZOU Y J . Data Shapley: equitable valuation of data for machine learning [J ] . CoRR , 2019 , abs/1904. 02868 .

SONG T S , TONG Y X , WER S Y . Profit allocation for federated learning [C ] // Proceedings of the 2019 IEEE International Conference on Big Data (Big Data) . Piscataway : IEEE Press , 2019 : 2577 - 2586 .

LIU Z L , CHEN Y Y , YU H , et al . GTG-Shapley: efficient and accurate participant contribution evaluation in federated learning [J ] . ACM Transactions on Intelligent Systems and Technology (TIST) , 2022 , 13 ( 4 ): 1 - 21 .

LECUN Y , BOTTOU L . Gradient-based learning applied to document recognition [J ] . Proceedings of the IEEE , 1998 , 86 ( 11 ): 2278 - 2324 .

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

无人机联邦学习场景下动态选择性同态加密隐私保护研究

基于可解释机器学习模型的5G下行空口关键性能研究

基于模型分割的联邦学习数据隐私保护方法

数字孪生辅助的智能楼宇多模态通信资源管理方法

云计算中基于Shapley值改进遗传算法的虚拟机调度模型