Scholarlink

<p>Survival models find vast applications in biomedical studies. However, survival data used to train these models are usually distributed, censored and facing a growing concern for data privacy. In addition to these challenges, the inherently skewed nature of survival time distributions has received comparatively limited attention. To address these issues, this study introduces a novel federated learning framework that leverages a multi-task variational autoencoder (MVAE) for data augmentation, specifically designed to mitigate both censoring and data skewness problems in survival analysis. Experimental results from extensive simulated and real world survival datasets have demonstrated the effectiveness of the proposed methodology with possible deployments at the server or the clients.</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

Federated survival analysis via data augmentation using multi-task variational autoencoder

  • 期刊论文

  • 发表日期：

    2026年2月11日

查看全文
  • WenJun Wang,
  • Yi Zou,
  • Zhuan Zhang,
  • Hong Wang

摘要

Survival models find vast applications in biomedical studies. However, survival data used to train these models are usually distributed, censored and facing a growing concern for data privacy. In addition to these challenges, the inherently skewed nature of survival time distributions has received comparatively limited attention. To address these issues, this study introduces a novel federated learning framework that leverages a multi-task variational autoencoder (MVAE) for data augmentation, specifically designed to mitigate both censoring and data skewness problems in survival analysis. Experimental results from extensive simulated and real world survival datasets have demonstrated the effectiveness of the proposed methodology with possible deployments at the server or the clients.