郑州大学学报（理学版）2026，Vol.58Issue(5)：37-44,8.DOI:10.13705/j.issn.1671-6841.2024187
基于区块链和模型共存的联邦蒸馏动态优化方法
Dynamic Optimization Method for Federated Distillation Based on Blockchain and Model Coexistence
摘要
Abstract
A dynamic optimization method was proposed for federated distillation based on blockchain and model coexistence to address the single point of failure risks and heterogeneity issues in the model training process of the federated learning center server.Firstly,a decentralized data sharing and communication platform was built by introducing blockchain technology to overcome the single point of failure risk of the central server.Secondly,a strategy of coexistence between global and personalized models was adopted.And knowledge distillation techniques was introduced in the local training process to achieve more flexible and efficient model training,to solve the problem of heterogeneity.The experimental results showed that the algorithm proposed could ensure high accuracy of personalized models and improve the stability and robustness of global models with large differences in data distribution,which could solve heterogeneity ef-fectively.At the same time,the use of blockchain could eliminate the hidden danger of single point fail-ures in central servers successfully.关键词
联邦学习/区块链/知识蒸馏/异质性/去中心化Key words
federated learning/blockchain/knowledge distillation/heterogeneity/decentralization分类
信息技术与安全科学引用本文复制引用
田钊,谢文顶,牛亚杰,徐旺旺,佘维..基于区块链和模型共存的联邦蒸馏动态优化方法[J].郑州大学学报（理学版）,2026,58(5):37-44,8.基金项目
国家重点研发计划项目(31703-3) （31703-3）
河南省高等学校重点科研项目(24A520045) （24A520045）
郑州大学产学研用科技创新团队培育项目(32214110-23) （32214110-23）
