2026年 02期

An Efficient Vertical Federated Learning Model Enhanced by Adaptive Gradient Sparsification

摘要(Abstract):

为了解决纵向联邦学习模型训练过程中模型参数传输通信开销较大的问题,提出一种基于自适应梯度稀疏化增强的高效纵向联邦学习模型;在各参与方上传加密梯度之前,进行梯度稀疏化,进而优化模型参数传输效率;建立定量稀疏化阈值的端到端动态自适应映射算法,实现超参数阈值的动态自适应求解;基于各参与方私有数据,构建映射模型的输入特征指标集,实现数据驱动的梯度阈值求解过程,提高阈值的求解精度。实验仿真结果表明,相较于基准对比模型,提出的基于自适应梯度稀疏化增强的高效纵向联邦学习模型训练速度平均提升24.4%,且电力数据异常检测准确率平均提升9%,在保障检测准确率的同时,有效提高了纵向联邦学习的建模效率。

关键词(KeyWords):纵向联邦学习;梯度高效传输算法;梯度稀疏化;神经网络

基金项目(Foundation): 国家重点研发计划项目(2022YFB2702805); 国网山东省电力公司科技项目(520626230045)

作者(Author):刘冬兰,赵夫慧,王睿,张昊,刘新,常英贤

DOI:10.13349/j.cnki.jdxbn.20251125.001

参考文献(References):

[1] LIU F X, ZHENG Z M, SHI Y X, et al. A survey on federated learning: a perspective from multi-party computation[J]. Frontiers of Computer Science, 2024, 18: 181336.

[2] LIU Y, KANG Y, ZOU T Y, et al. Vertical federated learning: concepts, advances, and challenges[J]. IEEE Transactions on Knowledge and Data Engineering, 2024, 36(7): 3615.

[3] 肖雄, 唐卓, 肖斌, 等. 联邦学习的隐私保护与安全防御研究综述[J]. 计算机学报, 2023, 46(5): 1019.

[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.

[5] WANG G Y, GU B, ZHANG Q S, et al. A unified solution for privacy and communication efficiency in vertical federated learning[J]. Advances in Neural Information Processing Systems, 2023, 36: 13480.

[6] LIU D W. Accelerating intra-party communication in vertical federated learning with RDMA[C] // DistributedML’20: Proceedings of the 1st Workshop on Distributed Machine Learning. New York: ACM, 2020: 14.

[7] KHAN A, ten THIJ M, WILBIK A. Communication-efficient vertical federated learning[J]. Algorithms, 2022, 15(8): 273.

[8] CASTIGLIA T J, DAS A, WANG S Q, et al. Compressed-VFL: communication-efficient learning with vertically partitioned data[C] // CHAUDHURI K, JEGELKA S, SONG L, et al. Proceedings of the 39th International Conference on Machine Learning. [S. l. ]: ML Research Press, 2022: 2738.

[9] 张宇航, 嵩天. 嵌入和梯度双向压缩的高效纵向联邦学习[J]. 计算机系统应用, 2024, 33(10): 190.

[10] SUN J W, XU Z Y, YANG D, et al. Communication-efficient vertical federated learning with limited overlapping samples[C] // Proceedings of the IEEE/ CVF International Conference on Computer Vision, October 2-6, 2023, Paris, France. New York: IEEE, 2023: 5203.

[11] FU R, WU Y C, XU Q Q, et al. FEAST: a communicationefficient federated feature selection framework for relational data[J]. Proceedings of the ACM on Management of Data, 2023, 1(1): Article 107.

[12] XIE C L, CHEN P Y, LI Q B, et al. Improving privacy-preserving vertical federated learning by efficient communication with ADMM[C] //2024 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML), April 9-11, 2024, Toronto, ON, Canada. New York: IEEE, 2024: 443.

[13] FU F C, MIAO X P, JIANG J W, et al. Towards communication-efficient vertical federated learning training via cache-enabled local updates[C] // ÖZCAN F, FREIRE J, LIN X M. Proceedings of the VLDB Endowment, Volume 15, Issue 10. Trondheim: VLDB Endowment, 2022: 2111.

[14] 唐晓, 陈芳, 许强, 等. 改进鲸鱼算法优化的多维度深度极限学习机短期负荷预测[J]. 山东电力技术, 2023, 50(1): 2.

[15] 吕秋霞, 孙亮, 车延华, 等. 基于深度置信网络的配电网负荷预测[J]. 山东电力技术, 2023, 50(8): 21.

[16] 庄立生. 融合气象特征的BP 神经网络电力系统短期负荷预测[J]. 山东电力技术, 2023, 50(11): 53.

[17] 李谟兴, 何永秀, 柳洋, 等. 基于大数据与机器学习的配电网电缆线路工程造价预测[J]. 山东电力技术, 2023, 50(1): 41.

[18] 白国政. 基于贝叶斯网络的电力变压器局部放电故障检测[J]. 计算机测量与控制, 2023, 31(9): 92.

[19] 舒一飞, 刘兴杰, 康洁莹, 等. 基于EMD 的异常用电检测方法[J]. 应用科技, 2022, 49(2): 93.

[20] 马一杰, 陈君, 刘松. 基于长短时记忆网络的电力负荷异常检测[J]. 云南大学学报(自然科学版), 2020, 42(增刊2): 56.

[21] 唐豫川, 苏彦莽, 何少华, 等. 基于DCE-LAE 的电力负荷异常检测方法[J]. 计算机工程与设计, 2023, 44(6): 1695.

[22] 蔡嘉辉, 王琨, 董康, 等. 基于DenseNet 和随机森林的电力用户窃电检测[J]. 计算机应用, 2021, 41(增刊1): 77.