基于模型预分配与自蒸馏的个性化联邦学习方法

doi:10.11772/j.issn.1001-9081.2025010115

《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (1): 10-20.DOI: 10.11772/j.issn.1001-9081.2025010115

基于模型预分配与自蒸馏的个性化联邦学习方法

张珂嘉¹, 方志军¹^,²(), 周南润¹, 史志才³

^1.上海工程技术大学电子电气工程学院，上海 201620
^2.东华大学计算机科学与技术学院，上海 201620
^3.上海中侨职业技术大学信息工程学院，上海 201514

收稿日期:2025-02-07 修回日期:2025-03-14 接受日期:2025-03-19 发布日期:2026-01-10 出版日期:2026-01-10
通讯作者: 方志军
作者简介:张珂嘉（2001—），男，河南新乡人，硕士研究生，主要研究方向：联邦学习
周南润（1976—），男，江西吉安人，教授，博士，主要研究方向：信息与通信工程、网络空间安全
史志才（1964—），男，吉林磐石人，教授，博士，主要研究方向：嵌入式系统、计算机网络、信息安全。
基金资助:
科技创新2030—“新一代人工智能”重大项目(2020AAA0109300);上海市科委地方院校能力建设项目(23010501800)

Personalized federated learning method based on model pre-assignment and self-distillation

Kejia ZHANG¹, Zhijun FANG¹^,²(), Nanrun ZHOU¹, Zhicai SHI³

^1.School of Electronic and Electrical Engineering，Shanghai University of Engineering Science，Shanghai 201620，China
^2.School of Computer Science and Technology，Donghua University，Shanghai 201620，China
^3.School of Information Engineering，Shanghai Zhongqiao Vocational and Technical University，Shanghai 201514，China

Received:2025-02-07 Revised:2025-03-14 Accepted:2025-03-19 Online:2026-01-10 Published:2026-01-10
Contact: Zhijun FANG
About author:ZHANG Kejia， born in 2001， M. S. candidate. His research interests include federated learning.
ZHOU Nanrun， born in 1976， Ph. D.， professor. His research interests include information and communication engineering， cyberspace security.
SHI Zhicai， born in 1964， Ph. D.， professor. His research interests include embedded system， computer network， information security.
Supported by:
Scientific and Technological Innovation 2030-Major Project of “New Generation of Artificial Intelligence”(2020AAA0109300);Shanghai Municipal Science and Technology Commission Local College Capacity Building Project(23010501800)

摘要/Abstract

摘要：

联邦学习（FL）是一种分布式机器学习方法，即利用分布式数据在训练模型的同时保护数据隐私。然而，它在高度异构的数据分布情况时表现不佳。个性化联邦学习（PFL）通过为每个客户端提供个性化模型来解决这一问题。然而，以往的PFL算法主要侧重于客户端本地模型的优化，忽略了服务器端全局模型的优化，导致服务器计算资源没有得到充分利用。针对上述局限性，提出基于模型预分配（PA）与自蒸馏（SD）的PFL方法FedPASD。FedPASD从服务器端和客户端2方面入手：在服务器端，对下一轮客户端模型有针对性地预先分配，这样不仅能提高模型的个性化性能，还能有效利用服务器的计算能力；在客户端，经过分层训练，并通过模型自蒸馏微调使模型更好地适应本地数据分布的特点。在3个数据集CIFAR-10、Fashion-MNIST和CIFAR-100上，将FedPASD与FedCP （Federated Conditional Policy）、FedPAC （Personalization with feature Alignment and classifier Collaboration）和FedALA （Federated learning with Adaptive Local Aggregation）等作为基准的典型算法进行对比实验的结果表明：FedPASD在不同异构设置下的测试准确率都高于基准算法。具体而言，FedPASD在CIFAR-100数据集上，客户端数量为50，参与率为50%的实验设置中，测试准确率较传统FL算法提升了29.05~29.22个百分点，较PFL算法提升了1.11~20.99个百分点；在CIFAR-10数据集上最高可达88.54%测试准确率。

关键词: 联邦学习, 数据异构, 个性化联邦学习, 模型预分配, 自蒸馏

Abstract:

Federated Learning （FL） is a distributed machine learning method that utilizes distributed data for model training while ensuring data privacy. However， it performs poorly in scenarios with highly heterogeneous data distributions. Personalized Federated Learning （PFL） addresses this challenge by providing personalized models for each client. However， the previous PFL algorithms primarily focus on optimizing client local models， while ignoring optimization of server global model. Consequently， server computational resources are not utilized fully. To overcome these limitations，FedPASD， a PFL method based on model pre-assignment and self-distillation， was proposed. FedPASD was operated in both server-side and client-side aspects. On server-side， client models for the next round were pre-assigned targetedly， which not only enhanced model personalization performance， but also utilized server computational resources effectively. On client-side， models were trained hierarchically and fine-tuned using self-distillation to better adapt to local data distribution characteristics. Experimental results on three datasets， CIFAR-10，Fashion-MNIST， and CIFAR-100 of comparing FedPASD with classic algorithms such as FedCP （Federated Conditional Policy），FedPAC （Personalization with feature Alignment and classifier Collaboration）， and FedALA （Federated learning with Adaptive Local Aggregation） as baselines demonstrate that FedPASD achieves higher test accuracy than those of baseline algorithms under various data heterogeneity settings. Specifically，FedPASD achieves a test accuracy improvement of 29.05 to 29.22 percentage points over traditional FL algorithms and outperforms the PFL algorithms by 1.11 to 20.99 percentage points on CIFAR-100 dataset； on CIFAR-10 dataset，FedPASD achieves a maximum accuracy of 88.54%.

Key words: Federated Learning (FL), data heterogeneity, Personalized Federated Learning (PFL), model pre-assignment, self-distillation

中图分类号:

TP181

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, 2026, 46(1): 10-20.

Kejia ZHANG, Zhijun FANG, Nanrun ZHOU, Zhicai SHI. Personalized federated learning method based on model pre-assignment and self-distillation[J]. Journal of Computer Applications, 2026, 46(1): 10-20.

图/表 14

图1 FedPASD整体架构

Fig. 1 Overall architecture of FedPASD

图2 服务器端的模型预分配架构

Fig. 2 Server-side pre-assignment architecture of models

图3 客户端模型的本地训练流程

Fig. 3 Local training process of client model

表1 数据集信息

Tab. 1 Dataset information

数据集	任务	样本数	图片大小	类别数
CIFAR-10	图像分类	60 000	32 $×$ 32	10
Fashion-MNIST	图像分类	70 000	28 $×$ 28	10
CIFAR-100	图像分类	60 000	32 $×$ 32	100

表1 数据集信息

Tab. 1 Dataset information

数据集	任务	样本数	图片大小	类别数
CIFAR-10	图像分类	60 000	32 $×$ 32	10
Fashion-MNIST	图像分类	70 000	28 $×$ 28	10
CIFAR-100	图像分类	60 000	32 $×$ 32	100

图4 CIFAR-10数据集上各客户端数据样本的分布情况

Fig. 4 Distribution of data samples across clients on CIFAR-10 dataset

表2 不同客户端设置下不同算法在3个数据集上的测试准确率比较 ( %)

Tab. 2 Test accuracy comparison of different algorithms on three datasets under different client settings

算法	CIFAR-10		Fashion-MNIST		CIFAR-100
算法	20clients	50clients	20clients	50clients	20clients	50clients
FedAvg^［1］	39.34	38.22	78.80	75.46	20.47	12.40
FedProx^［14］	39.16	40.14	78.98	75.07	20.35	12.56
FedLC^［16］	39.14	40.03	79.00	75.89	20.34	12.39
FedBN^［9］	39.17	38.63	78.79	75.91	20.45	12.41
FedCP^［29］	87.76	83.58	96.32	95.35	46.68	39.44
FedFomo^［31］	87.52	81.60	96.49	94.81	38.09	35.69
FedProto^［33］	86.04	80.39	96.20	93.78	45.52	27.62
FedPAC^［12］	84.03	81.83	95.66	94.50	48.00	39.22
FedALA^［35］	88.14	83.96	96.34	95.46	45.60	37.92
FedGH^［39］	88.00	83.84	96.49	95.56	45.96	40.50
FedAS^［40］	85.84	81.26	96.61	95.64	24.51	20.62
FedPASD	88.54	84.30	96.73	95.55	50.42	41.61

图5 各算法在不同客户端设置下的收敛曲线

Fig. 5 Convergence curves of each algorithm under different client settings

表3 不同算法达到目标准确率所需的通信轮次

Tab. 3 Communication rounds required for different algorithms to achieve target accuracy

算法	CIFAR-10		Fashion-MNIST		CIFAR-100
算法	轮次	提速倍数	轮次	提速倍数	轮次	提速倍数
FedProto^［33］	200	1.00	200	1.00	200	1.00
FedFomo^［31］	54	3.70	158	1.27	—	—
FedPAC^［12］	—	—	—	—	160	1.25
FedCP^［29］	77	2.60	182	1.10	181	1.10
FedALA^［35］	63	3.17	185	1.09	200	1.00
FedGH^［39］	74	2.70	156	1.60	194	1.03
FedAS^［40］	120	1.67	105	1.90	—	—
FedPASD	48	4.17	135	1.48	136	1.47

表4 不同算法在默认实验设置下的AUC指标

Tab. 4 AUC metrics of different algorithms under default experimental setting

算法	CIFAR-10	Fashion-MNIST	CIFAR-100
FedAvg^［1］	0.800 3	0.943 3	0.806 0
FedProx^［14］	0.798 5	0.943 6	0.805 7
FedLC^［16］	0.798 1	0.943 0	0.806 3
FedBN^［9］	0.798 9	0.943 9	0.805 6
FedCP^［29］	0.985 3	0.9980	0.964 7
FedFomo^［31］	0.977 7	0.994 9	0.949 5
FedProto^［33］	0.984 5	0.997 6	0.965 2
FedPAC^［12］	0.944 3	0.990 5	0.961 9
FedALA^［35］	0.892 1	0.975 5	0.952 1
FedGH^［39］	0.968 5	0.971 5	0.908 2
FedAS^［40］	0.980 0	0.993 4	0.938 1
FedPASD	0.9856	0.996 5	0.9724

表5 不同数据异构程度下各算法在3个数据集上的测试准确率比较 ( %)

Tab. 5 Test accuracy comparison of each algorithm on three datasets under different degrees of data heterogeneity

算法	CIFAR-10				Fashion-MNIST				CIFAR-100
算法	$α$ =0.1	$α$ =0.3	$α$ =0.5	$α$ =0.8	$α$ =0.1	$α$ =0.3	$α$ =0.5	$α$ =0.8	$α$ =0.1	$α$ =0.3	$α$ =0.5	$α$ =0.8
FedAvg^［1］	39.34	47.31	52.07	52.92	78.80	83.37	86.00	85.76	20.47	22.27	22.89	23.05
FedProx^［14］	39.16	47.64	52.39	52.59	78.98	83.67	86.06	85.70	20.35	22.06	22.88	22.97
FedLC^［16］	39.14	47.53	52.16	52.71	79.00	83.54	85.80	85.80	20.34	22.08	22.86	22.93
FedBN^［9］	39.17	47.64	52.38	52.56	78.79	83.44	85.86	85.85	20.45	22.07	22.84	22.86
FedCP^［29］	87.76	76.46	70.59	65.76	96.32	91.77	90.02	89.02	46.68	36.15	30.23	25.68
FedFomo^［31］	87.52	76.22	67.04	61.69	96.49	91.44	88.65	87.06	38.09	29.14	23.23	19.80
FedProto^［33］	86.04	73.77	65.17	60.69	96.20	89.77	86.30	84.91	45.52	32.92	27.59	23.69
FedPAC^［12］	84.03	77.04	71.75	67.12	95.66	91.83	90.28	89.40	48.00	37.01	32.23	28.66
FedALA^［35］	88.14	77.45	71.87	67.41	96.34	92.15	90.46	89.46	45.60	37.77	33.80	29.50
FedGH^［39］	88.00	76.52	71.16	66.92	96.49	92.12	90.53	89.44	45.96	36.19	31.85	28.01
FedAS^［40］	85.84	70.01	59.66	52.92	96.61	91.87	89.21	87.71	24.51	17.16	11.92	6.83
FedPASD	88.54	77.96	72.38	67.60	96.73	92.36	90.50	89.55	50.42	39.26	33.86	29.30

表5 不同数据异构程度下各算法在3个数据集上的测试准确率比较 ( %)

Tab. 5 Test accuracy comparison of each algorithm on three datasets under different degrees of data heterogeneity

算法	CIFAR-10				Fashion-MNIST				CIFAR-100
算法	$α$ =0.1	$α$ =0.3	$α$ =0.5	$α$ =0.8	$α$ =0.1	$α$ =0.3	$α$ =0.5	$α$ =0.8	$α$ =0.1	$α$ =0.3	$α$ =0.5	$α$ =0.8
FedAvg^［1］	39.34	47.31	52.07	52.92	78.80	83.37	86.00	85.76	20.47	22.27	22.89	23.05
FedProx^［14］	39.16	47.64	52.39	52.59	78.98	83.67	86.06	85.70	20.35	22.06	22.88	22.97
FedLC^［16］	39.14	47.53	52.16	52.71	79.00	83.54	85.80	85.80	20.34	22.08	22.86	22.93
FedBN^［9］	39.17	47.64	52.38	52.56	78.79	83.44	85.86	85.85	20.45	22.07	22.84	22.86
FedCP^［29］	87.76	76.46	70.59	65.76	96.32	91.77	90.02	89.02	46.68	36.15	30.23	25.68
FedFomo^［31］	87.52	76.22	67.04	61.69	96.49	91.44	88.65	87.06	38.09	29.14	23.23	19.80
FedProto^［33］	86.04	73.77	65.17	60.69	96.20	89.77	86.30	84.91	45.52	32.92	27.59	23.69
FedPAC^［12］	84.03	77.04	71.75	67.12	95.66	91.83	90.28	89.40	48.00	37.01	32.23	28.66
FedALA^［35］	88.14	77.45	71.87	67.41	96.34	92.15	90.46	89.46	45.60	37.77	33.80	29.50
FedGH^［39］	88.00	76.52	71.16	66.92	96.49	92.12	90.53	89.44	45.96	36.19	31.85	28.01
FedAS^［40］	85.84	70.01	59.66	52.92	96.61	91.87	89.21	87.71	24.51	17.16	11.92	6.83
FedPASD	88.54	77.96	72.38	67.60	96.73	92.36	90.50	89.55	50.42	39.26	33.86	29.30

表6 不同客户端参与率下不同算法在CIFAR-10和CIFAR-100数据集上的测试准确率对比 ( %)

Tab. 6 Test accuracy comparison of different algorithms on CIFAR-10 and CIFAR-100 datasets under different client participation rates

算法	CIFAR-10			CIFAR-100
算法	$r$ = 30%	$r$ = 50%	$r$ = 80%	$r$ = 30%	$r$ = 50%	$r$ = 80%
FedAvg^［1］	38.03	38.22	40.34	11.90	12.40	12.71
FedProx^［14］	38.01	40.14	40.21	12.11	12.56	12.87
FedLC^［16］	39.17	40.03	40.01	12.15	12.39	12.75
FedBN^［9］	37.42	38.63	39.74	12.15	12.41	12.67
FedCP^［29］	83.26	83.58	83.90	38.36	39.44	40.27
FedFomo^［31］	80.25	81.60	82.22	32.76	35.69	37.14
FedProto^［33］	78.91	80.39	81.13	18.83	27.62	37.20
FedPAC^［12］	82.94	81.83	79.22	38.31	39.22	40.01
FedALA^［35］	83.15	83.96	83.93	37.97	37.92	37.10
FedGH^［39］	83.74	83.84	83.96	39.82	40.50	40.69
FedAS^［40］	81.67	81.26	80.70	20.14	20.62	23.20
FedPASD	83.72	84.30	84.64	40.05	41.61	43.51

表6 不同客户端参与率下不同算法在CIFAR-10和CIFAR-100数据集上的测试准确率对比 ( %)

Tab. 6 Test accuracy comparison of different algorithms on CIFAR-10 and CIFAR-100 datasets under different client participation rates

算法	CIFAR-10			CIFAR-100
算法	$r$ = 30%	$r$ = 50%	$r$ = 80%	$r$ = 30%	$r$ = 50%	$r$ = 80%
FedAvg^［1］	38.03	38.22	40.34	11.90	12.40	12.71
FedProx^［14］	38.01	40.14	40.21	12.11	12.56	12.87
FedLC^［16］	39.17	40.03	40.01	12.15	12.39	12.75
FedBN^［9］	37.42	38.63	39.74	12.15	12.41	12.67
FedCP^［29］	83.26	83.58	83.90	38.36	39.44	40.27
FedFomo^［31］	80.25	81.60	82.22	32.76	35.69	37.14
FedProto^［33］	78.91	80.39	81.13	18.83	27.62	37.20
FedPAC^［12］	82.94	81.83	79.22	38.31	39.22	40.01
FedALA^［35］	83.15	83.96	83.93	37.97	37.92	37.10
FedGH^［39］	83.74	83.84	83.96	39.82	40.50	40.69
FedAS^［40］	81.67	81.26	80.70	20.14	20.62	23.20
FedPASD	83.72	84.30	84.64	40.05	41.61	43.51

表7 超参数μ的不同取值对算法测试准确率的影响

Tab. 7 Impact of different hyperparameter μ value on algorithm test accuracy

$μ$	测试准确率/%
$μ$	CIFAR-10	Fashion-MNIST	CIFAR-100
0.0	88.19	96.60	50.23
0.2	88.29	96.61	50.27
0.5	88.54	96.73	50.42
0.8	88.62	96.64	50.39

表7 超参数μ的不同取值对算法测试准确率的影响

Tab. 7 Impact of different hyperparameter μ value on algorithm test accuracy

$μ$	测试准确率/%
$μ$	CIFAR-10	Fashion-MNIST	CIFAR-100
0.0	88.19	96.60	50.23
0.2	88.29	96.61	50.27
0.5	88.54	96.73	50.42
0.8	88.62	96.64	50.39

表8 超参数τ的不同取值对算法测试准确率的影响

Tab. 8 Impact of different hyperparameter τ value on algorithm test accuracy

$τ$	测试准确率/%
$τ$	CIFAR-10	Fashion-MNIST	CIFAR-100
2	88.54	96.73	50.42
5	88.32	96.68	50.67
8	88.40	96.65	50.49
10	88.22	96.70	50.38

表8 超参数τ的不同取值对算法测试准确率的影响

Tab. 8 Impact of different hyperparameter τ value on algorithm test accuracy

$τ$	测试准确率/%
$τ$	CIFAR-10	Fashion-MNIST	CIFAR-100
2	88.54	96.73	50.42
5	88.32	96.68	50.67
8	88.40	96.65	50.49
10	88.22	96.70	50.38

表9 不同模块对算法测试准确率的影响

Tab. 9 Impact of different modules on algorithm test accuracy

模块	测试准确率/%
模块	CIFAR-10	Fashion-MNIST	CIFAR-100
None	85.77	95.66	45.52
SD	88.24	96.64	50.10
PA	86.28	96.02	48.49
Both	88.54	96.73	50.42

参考文献 41

[1]	McMAHAN 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.
[2]	YAN Z， WICAKSANA J， WANG Z， et al. Variation-aware federated learning with multi-source decentralized medical image data ［J］. IEEE Journal of Biomedical and Health Informatics， 2021， 25（7）： 2615-2628.
[3]	CETINKAYA A E， AKIN M， SAGIROGLU S. Improving performance of federated learning based medical image analysis in non-IID settings using image augmentation ［C］// Proceedings of the 2021 International Conference on Information Security and Cryptology. Piscataway： IEEE， 2021： 69-74.
[4]	ZHENG W， YAN L， GOU C， et al. Federated meta-learning for fraudulent credit card detection ［C］// Proceedings of the 29th International Joint Conferences on Artificial Intelligence. California： ijcai.org， 2020： 4654-4660.
[5]	LIANG X， LIU Y， CHEN T， et al. Federated transfer reinforcement learning for autonomous driving ［M］// RAZAVI-FAR R， WANG B， TAYLOR M E， et al. Federated and transfer learning， ALO 27. Cham： Springer， 2023： 357-371.
[6]	REISIZADEH A， MOKHTARI A， HASSANI H， et al. FedPAQ： a communication-efficient federated learning method with periodic averaging and quantization ［C］// Proceedings of the 23rd International Conference on Artificial Intelligence and Statistics. New York： JMLR.org， 2020： 2021-2031.
[7]	李少波，杨磊，李传江，等.联邦学习概述：技术，应用及未来［J］.计算机集成制造系统， 2022， 28（7）： 2119-2138.
	LI S B， YANG L， LI C J， et al. Overview of federated learning： technology， applications and future ［J］. Computer Integrated Manufacturing Systems， 2022， 28（7）： 2119-2138.
[8]	WANG J， YANG X， CUI S， et al. Towards personalized federated learning via heterogeneous model reassembly ［C］// Proceedings of the 37th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2023： 29515-29531.
[9]	LI X， JIANG M， ZHANG X， et al. FedBN： federated learning on non-iid features via local batch normalization ［EB/OL］. ［2024-06-26］. .
[10]	JIN H， BAI D， YAO D， et al. Personalized edge intelligence via federated self-knowledge distillation ［J］. IEEE Transactions on Parallel and Distributed Systems， 2023， 34（2）： 567-580.
[11]	KAIROUZ P， McMAHAN H B， AVENT B， et al. Advances and open problems in federated learning ［J］. Foundations and Trends in Machine Learning， 2021， 14（1/2）： 1-210.
[12]	XU J， TONG X， HUANG S L. Personalized federated learning with feature alignment and classifier collaboration ［EB/OL］. ［2024-11-02］. .
[13]	ZHU H， XU J， LIU S， et al. Federated learning on non-IID data： a survey ［J］. Neurocomputing， 2021， 465： 371-390.
[14]	LI T， SAHU A K， ZAHEER M， et al. Federated optimization in heterogeneous networks ［EB/OL］. ［2024-11-02］. .
[15]	KARIMIREDDY S P， KALE S， MOHRI M， et al. SCAFFOLD： stochastic controlled averaging for federated learning ［C］// Proceedings of the 37th International Conference on Machine Learning. New York： JMLR.org， 2020： 5132-5143.
[16]	ZHANG J， LI Z， LI B， et al. Federated learning with label distribution skew via logits calibration ［C］// Proceedings of the 39th International Conference on Machine Learning. New York： JMLR.org， 2022： 26311-26329.
[17]	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.
[18]	LIU Z， WU F， WANG Y， et al. FedCL： federated contrastive learning for multi-center medical image classification ［J］. Pattern Recognition， 2023， 143： No.109739.
[19]	TAN A Z， YU H， CUI L， et al. Towards personalized federated learning ［J］. IEEE Transactions on Neural Networks and Learning Systems， 2023， 34（12）： 9587-9603.
[20]	SABAH F， CHEN Y， YANG Z， et al. Model optimization techniques in personalized federated learning： a survey ［J］. Expert Systems with Applications， 2024， 243： No.122874.
[21]	FALLAH A， MOKHTRI A， OZDAGLAR A. Personalized federated learning with theoretical guarantees： a model-agnostic meta-learning approach ［C］// Proceedings of the 34th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2020： 3557-3568.
[22]	FINN C， ABBEEL P， LEVINE S. Model-agnostic meta-learning for fast adaptation of deep networks ［C］// Proceedings of the 34th International Conference on Machine Learning. New York： JMLR.org， 2017： 1126-1135.
[23]	LIN T， KONG L， STICH S U， et al. Ensemble distillation for robust model fusion in federated learning ［C］// Proceedings of the 34th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2020： 2351-2363.
[24]	LI T， HU S， BEIRAMI A， et al. Ditto： fair and robust federated learning through personalization ［C］// Proceedings of the 38th International Conference on Machine Learning. New York： JMLR.org， 2021： 6357-6368.
[25]	DINH C T， TRAN N H， NGUYEN T D. Personalized federated learning with Moreau envelopes ［C］// Proceedings of the 34th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2020： 21394-21405.
[26]	COLLINS L， HASSANI H， MOKHTARI A， et al. Exploiting shared representations for personalized federated learning ［C］// Proceedings of the 38th International Conference on Machine Learning. New York： JMLR.org， 2021： 2089-2099.
[27]	ARIVAZHAGAN M G， AGGARWAL V， SINGH A K， et al. Federated learning with personalization layers ［EB/OL］. ［2024-10-03］. .
[28]	OH J， KIM S， YUN S Y. FedBABU： towards enhanced representation for federated image classification ［EB/OL］. ［2024-11-02］. .
[29]	ZHANG J， HUA Y， WANG H， et al. FedCP： separating feature information for personalized federated learning via conditional policy ［C］// Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York： ACM， 2023： 3249-3261.
[30]	HUANG Y， CHU L， ZHOU Z， et al. Personalized cross-silo federated learning on non-IID data ［C］// Proceedings of the 35th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2021： 7865-7873.
[31]	ZHANG M， SAPRA K， FIDLER S， et al. Personalized federated learning with first order model optimization ［EB/OL］. ［2024-11-02］. .
[32]	LIU X， LI H， XU G， et al. Adaptive privacy-preserving federated learning ［J］. Peer-to-Peer Networking and Applications， 2020， 13（6）： 2356-2366.
[33]	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.
[34]	LI H， CAI Z， WANG J， et al. FedTP： federated learning by transformer personalization ［J］. IEEE Transactions on Neural Networks and Learning Systems， 2024， 35（10）： 13426-13440.
[35]	ZHANG J， HUA Y， WANG H， et al. FedALA： adaptive local aggregation for personalized federated learning ［C］// Proceedings of the 37th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2023： 11237-11244.
[36]	LUO J， WU S. Adapt to adaptation： learning personalization for cross-silo federated learning ［C］// Proceedings of the 31st International Joint Conferences on Artificial Intelligence. California： ijcai.org， 2022： 2166-2173.
[37]	KRIZHEVSKY A. Learning multiple layers of features from tiny images ［EB/OL］. ［2024-11-02］. .
[38]	XIAO H， RASUL K， VOLLGRAF R. Fashion-MNIST： a novel image dataset for benchmarking machine learning algorithms ［EB/OL］. ［2024-11-02］. .
[39]	YI L， WANG G， LIU X， et al. FedGH： heterogeneous federated learning with generalized global header ［C］// Proceedings of the 31st ACM International Conference on Multimedia. New York： ACM， 2023： 8686-8696.
[40]	YANG X， HUANG W， YE M. FedAS： bridging inconsistency in personalized federated learning ［C］// Proceedings of the 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2024： 11986-11995.
[41]	PASZKE A， GROSS S， MASSA F， et al. PyTorch： an imperative style， high-performance deep learning library ［C］// Proceedings of the 33rd International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2019： 8026-8037.

基于模型预分配与自蒸馏的个性化联邦学习方法

Personalized federated learning method based on model pre-assignment and self-distillation

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 14

参考文献 41

相关文章 15

编辑推荐

Metrics

[1]	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》唯一官方网站, 2026, 46(1): 21-32.
[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]	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.
[4]	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.
[5]	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.
[6]	张宏扬, 张淑芬, 谷铮. 面向个性化与公平性的联邦学习算法[J]. 《计算机应用》唯一官方网站, 2025, 45(7): 2123-2131.
[7]	张一鸣, 曹腾飞. 基于本地漂移和多样性算力的联邦学习优化算法[J]. 《计算机应用》唯一官方网站, 2025, 45(5): 1447-1454.
[8]	范亚州, 李卓. 能耗约束下分层联邦学习模型质量优化的节点协作机制[J]. 《计算机应用》唯一官方网站, 2025, 45(5): 1589-1594.
[9]	陈庆礼, 郭渊博, 方晨. 面向数据异构的聚类联邦学习算法[J]. 《计算机应用》唯一官方网站, 2025, 45(4): 1086-1094.
[10]	项钰斐, 倪郑威. 基于演化博弈的分层联邦学习边缘联合动态分析[J]. 《计算机应用》唯一官方网站, 2025, 45(4): 1077-1085.
[11]	曾辉, 熊诗雨, 狄永正, 史红周. 基于剪枝的大模型联邦参数高效微调技术[J]. 《计算机应用》唯一官方网站, 2025, 45(3): 715-724.
[12]	林海力, 李京. 基于工作证明的联邦学习懒惰客户端识别方法[J]. 《计算机应用》唯一官方网站, 2025, 45(3): 856-863.
[13]	徐超, 张淑芬, 陈海田, 彭璐璐, 张帅华. 基于自适应差分隐私与客户选择优化的联邦学习方法[J]. 《计算机应用》唯一官方网站, 2025, 45(2): 482-489.
[14]	王心妍, 杜嘉程, 钟李红, 徐旺旺, 刘伯宇, 佘维. 融合电力数据的纵向联邦学习企业排污预测模型[J]. 《计算机应用》唯一官方网站, 2025, 45(2): 518-525.
[15]	陈海田, 陈学斌, 马锐奎, 张帅华. 面向遥感数据的基于本地差分隐私的联邦学习隐私保护方案[J]. 《计算机应用》唯一官方网站, 2025, 45(2): 506-517.