Scholarlink

Protecting patient privacy while enabling model training and data analysis is crucial. Federated Learning (FL) is a leading privacy-preserving approach that allows model training while keeping data within healthcare institutions and sharing only data aggregates. However, there is limited work on process discovery algorithms for federated execution in healthcare. Although a federated version of the Alpha Algorithm (AA) has been proposed, its inherent limitations restrict its practical use. In this paper, we propose a federated adaptation of the enhanced Alpha+ Algorithm (AA+). We formally demonstrate the equivalence between the results of the distributed and centralized algorithms, and provide an open-source software implementation. In preliminary test results we show the capabilities of the proposed federated algorithm.

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

Towards Distributed Process Discovery in Healthcare: Testing and Proving the Feasibility of the Federated Alpha+ Algorithm

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Leonardo Nucciarelli,
  • Roberto Gatta,
  • Andrada Mihaela Tudor,
  • Erica Tavazzi,
  • Giovanni Arcuri,
  • Mauro Vallati,
  • Gema Ibanez-Sanchez,
  • Zoe Valero-Ramon,
  • Carlos Fernández Llatas,
  • Andrea Damiani

摘要

Protecting patient privacy while enabling model training and data analysis is crucial. Federated Learning (FL) is a leading privacy-preserving approach that allows model training while keeping data within healthcare institutions and sharing only data aggregates. However, there is limited work on process discovery algorithms for federated execution in healthcare. Although a federated version of the Alpha Algorithm (AA) has been proposed, its inherent limitations restrict its practical use. In this paper, we propose a federated adaptation of the enhanced Alpha+ Algorithm (AA+). We formally demonstrate the equivalence between the results of the distributed and centralized algorithms, and provide an open-source software implementation. In preliminary test results we show the capabilities of the proposed federated algorithm.