Scholarlink

<p>In federated recommendation systems, different clients can have different data resources and computational capabilities, necessitating the design of personalized models for different clients for better recommendation accuracy. The existing heterogeneous federated recommendation framework, although effective, faces challenges in designing fully personalized local models for each client to enhance recommendation accuracy. To address the above challenge, this paper proposes HeteFedAT, a heterogeneous federated recommendation framework based on adversarial training. First, clients are allowed to design their own personalized models locally. Then, each client downloads the global model from the server and performs adversarial training between the global and local personalized models. The personalized model is updated locally on the client to capture personalized preferences, while the global model is updated uniformly on the server to learn global information. Based on this, HeteFedAT achieves accurate modeling of client personalized preferences and effective sharing of global information. Finally, extensive experiments are conducted to compare HeteFedAT with seven existing federated recommendation framework in five public datasets, namely Flixster, Douban, Filmtrust, ML-100K and ML-1&#xa0;M, to validate the effectiveness of the proposed HeteFedAT in this paper.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

The heterogeneous federated recommendation framework based on adversarial training

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月31日

查看全文
  • Suqi Zhang,
  • Yajuan Li,
  • Wang Zhao

摘要

In federated recommendation systems, different clients can have different data resources and computational capabilities, necessitating the design of personalized models for different clients for better recommendation accuracy. The existing heterogeneous federated recommendation framework, although effective, faces challenges in designing fully personalized local models for each client to enhance recommendation accuracy. To address the above challenge, this paper proposes HeteFedAT, a heterogeneous federated recommendation framework based on adversarial training. First, clients are allowed to design their own personalized models locally. Then, each client downloads the global model from the server and performs adversarial training between the global and local personalized models. The personalized model is updated locally on the client to capture personalized preferences, while the global model is updated uniformly on the server to learn global information. Based on this, HeteFedAT achieves accurate modeling of client personalized preferences and effective sharing of global information. Finally, extensive experiments are conducted to compare HeteFedAT with seven existing federated recommendation framework in five public datasets, namely Flixster, Douban, Filmtrust, ML-100K and ML-1 M, to validate the effectiveness of the proposed HeteFedAT in this paper.