《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (1): 21-32.DOI: 10.11772/j.issn.1001-9081.2024121817
菅银龙1,2,3, 陈学斌1,2,3(
), 景忠瑞1,2,3, 钟琪1,2,3, 张镇博1,2,3
收稿日期:2024-12-27
修回日期:2025-03-04
接受日期:2025-03-10
发布日期:2026-01-10
出版日期:2026-01-10
通讯作者:
陈学斌
作者简介:菅银龙(2001—),男,河南商丘人,硕士研究生,CCF会员,主要研究方向:数据安全、隐私保护基金资助:
Yinlong JIAN1,2,3, Xuebin CHEN1,2,3(
), Zhongrui JING1,2,3, Qi ZHONG1,2,3, Zhenbo ZHANG1,2,3
Received:2024-12-27
Revised:2025-03-04
Accepted:2025-03-10
Online:2026-01-10
Published:2026-01-10
Contact:
Xuebin CHEN
About author:JIAN Yinlong, born in 2001, M. S. candidate. His research interests include data security, privacy protection.Supported by:摘要:
针对非独立同分布(Non-IID)场景下,联邦学习系统面临收敛缓慢和模型准确率降低等挑战,提出联邦学习中基于条件生成对抗网络的数据增强方案(FDA-GAN)。首先,设计一种类别选择的条件生成器为每个类别添加独立的网络模块,并将标签作为条件信息,以更精确地提取各类别的特定特征;其次,提出一种覆盖类别的客户端选择策略来基于客户端的综合奖励,选择包含尽可能多类别的客户端集合参与训练,确保生成对抗网络(GAN)能学习到完整的类别分布;最后,利用生成样本扩充客户端的本地数据集,以优化本地数据的特征构成,减小客户端之间的偏差。实验结果表明,FDA-GAN在狄利克雷数据划分下,相较于CAP-GAN (Collaborated gAme Parallel learning based on GAN)的MNIST Score (MNIST inception Score)和Mode Score指标上分别提升了2.67和1.08, 在FID (Fréchet Inception Distance)和MMD (Maximum Mean Discrepancy)指标上分别降低了55.12和2.56;在不同的Non-IID场景下, FedAvg (Federated Averaging)和FedProx (Federated Proximal)算法在结合FDA-GAN后,在50轮通信轮次内达到收敛,并且准确率提升了至少30.36个百分点。可见, FDA-GAN可以提高生成样本的质量与多样性,而且与基线算法结合后可以大幅提高联邦模型的准确率和收敛速度。
中图分类号:
菅银龙, 陈学斌, 景忠瑞, 钟琪, 张镇博. 联邦学习中基于条件生成对抗网络的数据增强方案[J]. 计算机应用, 2026, 46(1): 21-32.
Yinlong JIAN, Xuebin CHEN, Zhongrui JING, Qi ZHONG, Zhenbo ZHANG. Data augmentation scheme based on conditional generative adversarial network in federated learning[J]. Journal of Computer Applications, 2026, 46(1): 21-32.
| 训练阶段 | 参数 | 值 |
|---|---|---|
| GAN模型训练 | GAN训练轮数 | 100 |
| 训练批次大小 | 64 | |
| 生成器输入噪声维度 | 100 | |
| 优化算法 | Adam | |
| 优化器的学习率 | 0.000 2 | |
| 优化器的超参数 | 0.5,0.999 | |
| 客户端数 | 20 | |
| 客户端参与比例 | 0.5 | |
| 联邦模型训练 | 全局训练轮数 | 200 |
| 本地训练轮数 | 1 | |
| 客户端数 | 20 | |
| 客户端参与比例 | 0.5 | |
| 训练批次大小 | 10 | |
| 优化算法 | SGD | |
| 优化器的学习率 | 0.01 |
表1 实验参数设置
Tab. 1 Experimental parameter setting
| 训练阶段 | 参数 | 值 |
|---|---|---|
| GAN模型训练 | GAN训练轮数 | 100 |
| 训练批次大小 | 64 | |
| 生成器输入噪声维度 | 100 | |
| 优化算法 | Adam | |
| 优化器的学习率 | 0.000 2 | |
| 优化器的超参数 | 0.5,0.999 | |
| 客户端数 | 20 | |
| 客户端参与比例 | 0.5 | |
| 联邦模型训练 | 全局训练轮数 | 200 |
| 本地训练轮数 | 1 | |
| 客户端数 | 20 | |
| 客户端参与比例 | 0.5 | |
| 训练批次大小 | 10 | |
| 优化算法 | SGD | |
| 优化器的学习率 | 0.01 |
| 方案 | IID | DIR | Fully Non-IID | |||
|---|---|---|---|---|---|---|
| FL-GAN | 123.25 | 5.15 | 174.75 | 3.95 | 211.80 | 3.15 |
| MD-GAN | 109.39 | 6.20 | 119.51 | 5.07 | 149.34 | 4.12 |
| CAP-GAN | ||||||
| FDA-GAN | 75.31 | 8.99 | 41.98 | 9.40 | 34.61 | 9.72 |
表2 MNIST数据集上的FID与MNIST Score指标对比结果
Tab. 2 Comparison results of FID and MNIST Score metrics on MNIST dataset
| 方案 | IID | DIR | Fully Non-IID | |||
|---|---|---|---|---|---|---|
| FL-GAN | 123.25 | 5.15 | 174.75 | 3.95 | 211.80 | 3.15 |
| MD-GAN | 109.39 | 6.20 | 119.51 | 5.07 | 149.34 | 4.12 |
| CAP-GAN | ||||||
| FDA-GAN | 75.31 | 8.99 | 41.98 | 9.40 | 34.61 | 9.72 |
| 方案 | IID | DIR | Fully Non-IID | |||
|---|---|---|---|---|---|---|
| FL-GAN | 5.10 | 5.95 | 4.97 | 24.68 | 4.14 | 34.17 |
| MD-GAN | 7.31 | 5.24 | 5.73 | 13.80 | 5.77 | 16.48 |
| CAP-GAN | 8.61 | |||||
| FDA-GAN | 8.37 | 0.29 | 8.26 | 0.96 | 0.42 | |
表3 FashionMNIST数据集上的Mode Score与MMD指标对比结果
Tab. 3 Comparison results of Mode Score and MMD metrics on FashionMNIST dataset
| 方案 | IID | DIR | Fully Non-IID | |||
|---|---|---|---|---|---|---|
| FL-GAN | 5.10 | 5.95 | 4.97 | 24.68 | 4.14 | 34.17 |
| MD-GAN | 7.31 | 5.24 | 5.73 | 13.80 | 5.77 | 16.48 |
| CAP-GAN | 8.61 | |||||
| FDA-GAN | 8.37 | 0.29 | 8.26 | 0.96 | 0.42 | |
| 算法 | MNIST | FashionMNIST | ||
|---|---|---|---|---|
| DIR场景 | Fully Non-IID场景 | DIR场景 | Fully Non-IID场景 | |
| FDA-GAN+FedAvg | 96.23 | |||
| FDA-GAN+FedProx | 94.09 | 83.96 | 78.92 | |
| FedAvg | 65.54 | 50.90 | 50.71 | 31.86 |
| FedProx | 65.40 | 51.49 | 50.57 | 34.50 |
表4 不同算法的模型准确率对比 ( %)
Tab. 4 Comparison of model accuracy among different algorithms
| 算法 | MNIST | FashionMNIST | ||
|---|---|---|---|---|
| DIR场景 | Fully Non-IID场景 | DIR场景 | Fully Non-IID场景 | |
| FDA-GAN+FedAvg | 96.23 | |||
| FDA-GAN+FedProx | 94.09 | 83.96 | 78.92 | |
| FedAvg | 65.54 | 50.90 | 50.71 | 31.86 |
| FedProx | 65.40 | 51.49 | 50.57 | 34.50 |
| 方案 | ||
|---|---|---|
| FL-GAN | 171.45 | 4.21 |
| MD-GAN | 116.72 | 5.42 |
| CAP-GAN | ||
| FDA-GAN | 38.74 | 9.60 |
表5 固定客户端参与数为10的指标对比
Tab. 5 Comparison of metrics when the number of participating clients is fixed to 10
| 方案 | ||
|---|---|---|
| FL-GAN | 171.45 | 4.21 |
| MD-GAN | 116.72 | 5.42 |
| CAP-GAN | ||
| FDA-GAN | 38.74 | 9.60 |
| 方案 | 参与比例为0.10 | 参与比例为0.25 | 参与比例为0.50 | 参与比例为0.75 | ||||
|---|---|---|---|---|---|---|---|---|
| FL-GAN | 193.97 | 3.14 | 194.75 | 3.55 | 174.75 | 3.95 | 177.32 | 4.02 |
| MD-GAN | 147.67 | 3.68 | 140.17 | 4.27 | 119.51 | 5.07 | 116.95 | 5.50 |
| CAP-GAN | ||||||||
| FDA-GAN | 43.01 | 9.35 | 42.25 | 9.44 | 41.98 | 9.40 | 38.63 | 9.66 |
表6 不同客户端参与比例下的FID与MNIST Score指标对比(DIR)
Tab. 6 Comparison of FID and MNIST Score metrics under different client participation ratios (DIR)
| 方案 | 参与比例为0.10 | 参与比例为0.25 | 参与比例为0.50 | 参与比例为0.75 | ||||
|---|---|---|---|---|---|---|---|---|
| FL-GAN | 193.97 | 3.14 | 194.75 | 3.55 | 174.75 | 3.95 | 177.32 | 4.02 |
| MD-GAN | 147.67 | 3.68 | 140.17 | 4.27 | 119.51 | 5.07 | 116.95 | 5.50 |
| CAP-GAN | ||||||||
| FDA-GAN | 43.01 | 9.35 | 42.25 | 9.44 | 41.98 | 9.40 | 38.63 | 9.66 |
| 方案 | 客户端数为10 | 客户端数为30 | 客户端数为40 | 客户端数为50 | ||||
|---|---|---|---|---|---|---|---|---|
| FL-GAN | 3.81 | 47.60 | 4.61 | 31.19 | 4.87 | 27.39 | 4.90 | 24.16 |
| MD-GAN | 4.76 | 20.60 | 5.84 | 17.33 | 6.33 | 14.90 | 6.43 | 13.72 |
| CAP-GAN | 8.66 | 8.71 | ||||||
| FDA-GAN | 8.47 | 0.51 | 0.41 | 0.41 | 8.77 | 0.40 | ||
表7 不同客户端数的Mode Score与MMD指标对比(Fully Non-IID)
Tab. 7 Comparison of Mode Score and MMD metrics with different numbers of clients (Fully Non-IID)
| 方案 | 客户端数为10 | 客户端数为30 | 客户端数为40 | 客户端数为50 | ||||
|---|---|---|---|---|---|---|---|---|
| FL-GAN | 3.81 | 47.60 | 4.61 | 31.19 | 4.87 | 27.39 | 4.90 | 24.16 |
| MD-GAN | 4.76 | 20.60 | 5.84 | 17.33 | 6.33 | 14.90 | 6.43 | 13.72 |
| CAP-GAN | 8.66 | 8.71 | ||||||
| FDA-GAN | 8.47 | 0.51 | 0.41 | 0.41 | 8.77 | 0.40 | ||
| [1] | SATTLER F, MÜLLER K R, SAMEK W. Clustered federated learning: model-agnostic distributed multitask optimization under privacy constraints [J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 32(8): 3710-3722. |
| [2] | LV Y, DING H, WU H, et al. FedRDS: federated learning on non-IID data via regularization and data sharing [J]. Applied Sciences, 2023, 13(23): No.12962. |
| [3] | LI Q, HE B, SONG D. Model-contrastive federated learning [C]// Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 10708-10717. |
| [4] | 宋华伟,李升起,万方杰,等.非独立同分布场景下的联邦学习优化方法[J].计算机工程, 2024, 50(3): 166-172. |
| SONG H W, LI S Q, WAN F J, et al. Federated learning optimization method in non-IID scenarios [J]. Computer Engineering, 2024, 50(3): 166-172. | |
| [5] | ZHANG J, LI A, TANG M, et al. Fed-CBS: a heterogeneity-aware client sampling mechanism for federated learning via class-imbalance reduction [C]// Proceedings of the 2023 International Conference on Machine Learning. New York: JMLR.org, 2023: 41354-41381. |
| [6] | JEONG E, OH S, KIM H, et al. Communication-efficient on-device machine learning: federated distillation and augmentation under non-IID private data [EB/OL]. [2024-01-19]. . |
| [7] | 汤凌韬,王迪,刘盛云.面向非独立同分布数据的联邦学习数据增强方案[J].通信学报, 2023, 44(1): 164-176. |
| TANG L T, WANG D, LIU S Y. Data augmentation scheme for federated learning with non-IID data [J]. Journal on Communications, 2023, 44(1): 164-176. | |
| [8] | ZHAO Y, LI M, LAI L, et al. Federated learning with non-IID data [EB/OL]. [2024-04-08]. . |
| [9] | HEUSEL M, RAMSAUER H, UNTERTHINER T, et al. GANs trained by a two time-scale update rule converge to a local Nash equilibrium [C]// Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2017: 6629-6640. |
| [10] | SATTLER F, WIEDEMANN S, MÜLLER K R, et al. Robust and communication-efficient federated learning from non-i.i.d data [J]. IEEE Transactions on Neural Networks and Learning Systems, 2020, 31(9): 3400-3413. |
| [11] | LI T, SAHU A K, TALWALKAR A, et al. Federated learning: challenges, methods, and future directions [J]. IEEE Signal Processing Magazine, 2020, 37(3): 50-60. |
| [12] | McMAHAN H B, MOORE E, RAMAGE D, et al. Communication-efficient learning of deep networks from decentralized data [C]// Proceedings of the 20th International Conference on Artificial Intelligence and Statistics. New York: JMLR.org, 2017: 1273-1282. |
| [13] | LI T, SAHU A K, ZAHEER M, et al. Federated optimization in heterogeneous networks [EB/OL]. [2024-10-16]. . |
| [14] | ARIVAZHAGAN M G, AGGARWAL V, SINGH A K, et al. Federated learning with personalization layers [EB/OL]. [2024-10-16]. . |
| [15] | MU X, SHEN Y, CHENG K, et al. FedProc: prototypical contrastive federated learning on non-IID data [J]. Future Generation Computer Systems, 2023, 143: 93-104. |
| [16] | TAN Y, LONG G, LIU L, et al. FedProto: federated prototype learning across heterogeneous clients [C]// Proceedings of the 36th AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2022: 8432-8440. |
| [17] | ZHU H, XU J, LIU S, et al. Federated learning on non-IID data: a survey [J]. Neurocomputing, 2021, 465: 371-390. |
| [18] | ZHANG W, WANG X, ZHOU P, et al. Client selection for federated learning with non-IID data in mobile edge computing [J]. IEEE Access, 2021, 9: 24462-24474. |
| [19] | 张祖篡,陈学斌,高瑞,等.基于标签分类的联邦学习客户端选择方法[J].计算机应用, 2024, 44(12): 3759-3765. |
| ZHANG Z C, CHEN X B, GAO R, et al. Federated learning client selection method based on label classification [J]. Journal of Computer Applications, 2024, 44(12): 3759-3765. | |
| [20] | ZHANG Y, XU C, YANG H H, et al. DPP-based client selection for federated learning with non-IID data [C]// Proceedings of the 2023 IEEE International Conference on Acoustics, Speech and Signal Processing. Piscataway: IEEE, 2023: 1-5. |
| [21] | YOON T, SHIN S, HWANG S J, et al. FedMix: approximation of mixup under mean augmented federated learning [EB/OL]. [2024-10-16]. . |
| [22] | GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al. Generative adversarial nets [C]// Proceedings of the 28th International Conference on Neural Information Processing Systems — Volume 2. Cambridge: MIT Press, 2014: 2672-2680. |
| [23] | MIRZA M, OSINDERO S. Conditional generative adversarial nets [EB/OL]. [2024-07-13]. . |
| [24] | LEE M B, KIM Y H, PARK K R. Conditional generative adversarial network-based data augmentation for enhancement of iris recognition accuracy [J]. IEEE Access, 2019, 7: 122134-122152. |
| [25] | LI Z, SHAO J, MAO Y, et al. Federated learning with GAN-based data synthesis for non-iid clients [C]// Proceedings of the 2022 International Workshop on Trustworthy Federated Learning, LNCS 13448. Cham: Springer, 2023: 17-32. |
| [26] | GOLHAR D, JOSHI N, JOSHI M V. Generative adversarial network: use of Non-IID samples [EB/OL]. [2024-07-22]. . |
| [27] | HARDY C, LE MERRER E, SERICOLA B. MD-GAN: multi-discriminator generative adversarial networks for distributed datasets [C]// Proceedings of the 2019 IEEE International Parallel and Distributed Processing Symposium. Piscataway: IEEE, 2019: 866-877. |
| [28] | ZHANG J, ZHAO L, YU K, et al. A novel federated learning scheme for generative adversarial networks [J]. IEEE Transactions on Mobile Computing, 2024, 23(5): 3633-3649. |
| [29] | CHEN W, WANG Y, YUAN Y. Combinatorial multi-armed bandit: general framework and applications [C]// Proceedings of the 30th International Conference on Machine Learning. New York: JMLR.org, 2013: 151-159. |
| [30] | AUER P, CESA-BIANCHI N, FISCHER P, et al. Finite-time analysis of the multiarmed bandit problem [J]. Machine Learning, 2002, 47(2/3): 235-256. |
| [31] | DENG L. The MNIST database of handwritten digit images for machine learning research [best of the web] [J]. IEEE Signal Processing Magazine, 2012, 29(6): 141-142. |
| [32] | XIAO H, RASUL K, VOLLGRAF R. Fashion-MNIST: a novel image dataset for benchmarking machine learning algorithms [EB/OL]. [2024-05-10]. . |
| [33] | MA Z, LIU Y, MIAO Y, et al. FlGAN: GAN-based unbiased federated learning under non-IID settings [J]. IEEE Transactions on Knowledge and Data Engineering, 2023, 36(4): 1566-1581. |
| [34] | SALIMANS T, GOODFELLOW I, ZAREMBA W, et al. Improved techniques for training GANs [C]// Proceedings of the 30th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2016: 2234-2242. |
| [35] | CHE T, LI Y, JACOB A P, et al. Mode regularized generative adversarial networks [EB/OL]. [2024-11-21]. . |
| [36] | GRETTON A, BORGWARDT K M, RASCH M J, et al. A kernel two-sample test [J]. Journal of Machine Learning Research, 2012, 13: 723-773. |
| [1] | 樊娜, 罗闯, 张泽晖, 张梦瑶, 穆鼎. 基于改进生成对抗网络的车辆轨迹语义隐私保护机制[J]. 《计算机应用》唯一官方网站, 2026, 46(1): 169-180. |
| [2] | 俞浩, 范菁, 孙伊航, 金亚东, 郗恩康, 董华. 边缘异构下的联邦分割学习优化方法[J]. 《计算机应用》唯一官方网站, 2026, 46(1): 33-42. |
| [3] | 王丽芳, 任文婧, 郭晓东, 张荣国, 胡立华. 用于低剂量CT图像降噪的多路特征生成对抗网络[J]. 《计算机应用》唯一官方网站, 2026, 46(1): 270-279. |
| [4] | 李世伟, 周昱峰, 孙鹏飞, 刘伟松, 孟竹喧, 廉浩杰. 基于煤尘对激光雷达电磁波散射和吸收效应的点云数据增强方法[J]. 《计算机应用》唯一官方网站, 2026, 46(1): 331-340. |
| [5] | 昝志辉, 王雅静, 李珂, 杨智翔, 杨光宇. 基于SAA-CNN-BiLSTM网络的多特征融合语音情感识别方法[J]. 《计算机应用》唯一官方网站, 2026, 46(1): 69-76. |
| [6] | 张珂嘉, 方志军, 周南润, 史志才. 基于模型预分配与自蒸馏的个性化联邦学习方法[J]. 《计算机应用》唯一官方网站, 2026, 46(1): 10-20. |
| [7] | 李玟, 李开荣, 杨凯. 基于数据增强的子图感知对比学习[J]. 《计算机应用》唯一官方网站, 2026, 46(1): 1-9. |
| [8] | 王闯, 俞璐, 陈健威, 潘成, 杜文博. 开集域适应综述[J]. 《计算机应用》唯一官方网站, 2025, 45(9): 2727-2736. |
| [9] | 邓伊琳, 余发江. 基于LSTM和可分离自注意力机制的伪随机数生成器[J]. 《计算机应用》唯一官方网站, 2025, 45(9): 2893-2901. |
| [10] | 梁一鸣, 范菁, 柴汶泽. 基于双向交叉注意力的多尺度特征融合情感分类[J]. 《计算机应用》唯一官方网站, 2025, 45(9): 2773-2782. |
| [11] | 李莉, 宋涵, 刘培鹤, 陈汉林. 基于数据增强和残差网络的敏感信息命名实体识别[J]. 《计算机应用》唯一官方网站, 2025, 45(9): 2790-2797. |
| [12] | 俞浩, 范菁, 孙伊航, 董华, 郗恩康. 联邦学习统计异质性综述[J]. 《计算机应用》唯一官方网站, 2025, 45(9): 2737-2746. |
| [13] | 周金, 李玉芝, 张徐, 高硕, 张立, 盛家川. 复杂电磁环境下的调制识别网络[J]. 《计算机应用》唯一官方网站, 2025, 45(8): 2672-2682. |
| [14] | 苏锦涛, 葛丽娜, 肖礼广, 邹经, 王哲. 联邦学习中针对后门攻击的检测与防御方案[J]. 《计算机应用》唯一官方网站, 2025, 45(8): 2399-2408. |
| [15] | 葛丽娜, 王明禹, 田蕾. 联邦学习的高效性研究综述[J]. 《计算机应用》唯一官方网站, 2025, 45(8): 2387-2398. |
| 阅读次数 | ||||||
|
全文 |
|
|||||
|
摘要 |
|
|||||