![[2602.12708] Mixture of Predefined Experts: Maximizing Data Usage on Vertical Federated Learning](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.12708] Mixture of Predefined Experts: Maximizing Data Usage on Vertical Federated Learning
Summary
The paper introduces Split-MoPE, a novel framework for Vertical Federated Learning that maximizes data usage by integrating predefined experts, enhancing performance in privacy-sensitive domains.
Why It Matters
As Vertical Federated Learning becomes increasingly vital in sectors like finance and healthcare, addressing the challenges of data misalignment is crucial. Split-MoPE offers a robust solution that enhances data utilization and model performance while maintaining privacy, making it relevant for researchers and practitioners in machine learning and data science.
Key Takeaways
- Split-MoPE integrates Split Learning with a Mixture of Predefined Experts to optimize data usage.
- The framework is designed to handle sample misalignment, a common issue in real-world federated learning scenarios.
- Split-MoPE achieves superior performance with reduced communication needs compared to traditional methods.
- It provides interpretability by quantifying each participant's contribution to predictions.
- Extensive evaluations show its effectiveness on both vision and tabular datasets.
Computer Science > Machine Learning arXiv:2602.12708 (cs) [Submitted on 13 Feb 2026] Title:Mixture of Predefined Experts: Maximizing Data Usage on Vertical Federated Learning Authors:Jon Irureta, Gorka Azkune, Jon Imaz, Aizea Lojo, Javier Fernandez-Marques View a PDF of the paper titled Mixture of Predefined Experts: Maximizing Data Usage on Vertical Federated Learning, by Jon Irureta and 4 other authors View PDF HTML (experimental) Abstract:Vertical Federated Learning (VFL) has emerged as a critical paradigm for collaborative model training in privacy-sensitive domains such as finance and healthcare. However, most existing VFL frameworks rely on the idealized assumption of full sample alignment across participants, a premise that rarely holds in real-world scenarios. To bridge this gap, this work introduces Split-MoPE, a novel framework that integrates Split Learning with a specialized Mixture of Predefined Experts (MoPE) architecture. Unlike standard Mixture of Experts (MoE), where routing is learned dynamically, MoPE uses predefined experts to process specific data alignments, effectively maximizing data usage during both training and inference without requiring full sample overlap. By leveraging pretrained encoders for target data domains, Split-MoPE achieves state-of-the-art performance in a single communication round, significantly reducing the communication footprint compared to multi-round end-to-end training. Furthermore, unlike existing proposals that address sam...
