Scholarlink

<p>In modern manufacturing value chains, achieving optimal product quality and sustainability necessitates collaboration across interconnected stakeholders. Conventional Latent Variable Model Inversion (LVMI) methods, though widely applied in process optimization, face limitations in cross-organizational settings due to data privacy concerns, as they typically require sharing sensitive data. To address this, the present work introduces a novel, two-phase approach for privacy-preserving value chain optimization called Secure LVMI. The approach first utilizes Privacy-Preserving Partial Least Squares (P3LS) to collaboratively build an integrated process model without sharing raw data, followed by the secure estimation of statistical control limits using Secure Multi-Party Computation (MPC). Subsequently, leveraging the established model and limits, a combination of MPC and a Cooperative Coevolution Genetic Algorithm (CCGA) performs secure model inversion to find optimal process settings. The effectiveness of Secure LVMI is demonstrated through experiments on simulated datasets. By providing solutions with quality comparable to centralized methods while maintaining privacy, this framework offers potential for broad application in industries where privacy concerns restrict traditional data sharing and optimization techniques.</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

Vertical federated latent variable model inversion for collaborative value chain optimization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月7日

查看全文
  • Du Nguyen Duy,
  • Ramin Nikzad-Langerodi,
  • Josef Scharinger,
  • Michael Affenzeller

摘要

In modern manufacturing value chains, achieving optimal product quality and sustainability necessitates collaboration across interconnected stakeholders. Conventional Latent Variable Model Inversion (LVMI) methods, though widely applied in process optimization, face limitations in cross-organizational settings due to data privacy concerns, as they typically require sharing sensitive data. To address this, the present work introduces a novel, two-phase approach for privacy-preserving value chain optimization called Secure LVMI. The approach first utilizes Privacy-Preserving Partial Least Squares (P3LS) to collaboratively build an integrated process model without sharing raw data, followed by the secure estimation of statistical control limits using Secure Multi-Party Computation (MPC). Subsequently, leveraging the established model and limits, a combination of MPC and a Cooperative Coevolution Genetic Algorithm (CCGA) performs secure model inversion to find optimal process settings. The effectiveness of Secure LVMI is demonstrated through experiments on simulated datasets. By providing solutions with quality comparable to centralized methods while maintaining privacy, this framework offers potential for broad application in industries where privacy concerns restrict traditional data sharing and optimization techniques.