Internet of Medical Things (IoMT) facilitate revolutionary development in healthcare services, recognized as smart healthcare. By collecting big healthcare data and utilizing artificial intelligence algorithms, P4-medicine can be realized in intelligent diagnosis, risk analysis, and health management. As more data is collected, privacy and security become imperatives in healthcare research, and split learning is ideal for big data predictions, but relative research is at an early stage without systematically designing and considering data characteristics. In this paper, a Heterogeneous Transfer learning for spliT Prediction System (HTTPS) is proposed. HTTPS converts the dataset into both sparse and dense feature matrices, subsequently directing them into the sparse and dense embedding networks. For privacy considerations, embedding networks are designed as split learning to embed local features, and can further transfer experience from heterogeneous data. Experimental findings demonstrate that HTTPS outperforms the benchmark systems and has strong transferability. Furthermore, the designed mechanism motivates users to share some personal information to obtain precise predictions, and still provides a general model for privacy-conscious users.

中文翻译：

---

HTTPS：根据医疗数据评估 spliT 预测系统的异构迁移学习


医疗物联网（IoMT）促进了医疗服务的革命性发展，被称为智慧医疗。通过采集健康医疗大数据，利用人工智能算法，实现P4医学的智能诊断、风险分析、健康管理。随着收集的数据越来越多，隐私和安全成为医疗保健研究的当务之急，而分割学习是大数据预测的理想选择，但相关研究还处于早期阶段，没有系统地设计和考虑数据特征。在本文中，提出了一种用于 spliT 预测系统（HTTPS）的异构迁移学习。 HTTPS 将数据集转换为稀疏和稠密特征矩阵，随后将它们引导到稀疏和稠密嵌入网络中。出于隐私考虑，嵌入网络被设计为分裂学习来嵌入局部特征，并且可以进一步传递来自异构数据的经验。实验结果表明，HTTPS 的性能优于基准系统，并且具有很强的可传输性。此外，所设计的机制激励用户共享一些个人信息以获得精确的预测，并且仍然为注重隐私的用户提供了通用模型。