PreComp-VFL: Enhancing Embedding Representation in Vertical Federated Learning for Medical Data
- Publisher:
- Institute of Electrical and Electronics Engineers (IEEE)
- Publication Type:
- Conference Proceeding
- Citation:
- 2025 9th Cyber Security in Networking Conference (CSNet), 2025, 00, pp. 1-5
- Issue Date:
- 2025-10-22
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Vertical federated learning (VFL) in healthcare enables institutions holding different features of the same patients to collaboratively train machine learning models without exposing raw data. However, most existing VFL approaches focus on secure aggregation and privacy-preserving computation, while overlooking client-side preprocessing. This limitation can result in less informative embedding representations and increased communication overhead. To address this, we propose PreComp-VFL, a client-level preprocessing method that integrates unsupervised feature selection with dimensionality reduction techniques. PreComp-VFL allows clients to transmit compressed, informative, and privacy-preserving embeddings without requiring label access. We used four real-world medical datasets in a number of experiments demonstrating our proposed scheme and achieved improved model accuracy and F1 Score compared to standard VFL. We also show that our proposed scheme achieves significant reductions in communication costs relative to server-side feature selection scheme.
Please use this identifier to cite or link to this item: