Samuel S. Wu

Guanhong Miao, A. Adam Ding, Samuel S. Wu

The rapid development in Internet of Medical Things (IoMT) boosts the opportunity for real-time health monitoring using various data types such as electroencephalography (EEG) and electrocardiography (ECG). Security issues have significantly impeded the e-healthcare system implementation. Three important challenges for privacy preserving system need to be addressed: accurate diagnosis, privacy protection without compromising accuracy, and computation efficiency. It is essential to guarantee prediction accuracy since disease diagnosis is strongly related to health and life. By implementing matrix encryption method, we propose a real-time disease diagnosis scheme using support vector machine (SVM). A biomedical signal provided by the client is diagnosed such that the server does not get any information about the signal as well as the final result of the diagnosis while the proposed scheme also achieves confidentiality of the SVM classifier and the server's medical data. The proposed scheme has no accuracy degradation. Experiments on real-world data illustrate the high efficiency of the proposed scheme. It takes less than 1 second to derive the disease diagnosis result using a device with 4Gb RAMs, suggesting the feasibility to implement real-time privacy preserving health monitoring.

A. Adam Ding, Samuel S. Wu, Guanhong Miao et al.

Differential privacy schemes have been widely adopted in recent years to address issues of data privacy protection. We propose a new Gaussian scheme combining with another data protection technique, called random orthogonal matrix masking, to achieve $(\varepsilon, δ)$-differential privacy (DP) more efficiently. We prove that the additional matrix masking significantly reduces the rate of noise variance required in the Gaussian scheme to achieve $(\varepsilon, δ)-$DP in big data setting. Specifically, when $\varepsilon \to 0$, $δ\to 0$, and the sample size $n$ exceeds the number $p$ of attributes by $(n-p)=O(ln(1/δ))$, the required additive noise variance to achieve $(\varepsilon, δ)$-DP is reduced from $O(ln(1/δ)/\varepsilon^2)$ to $O(1/\varepsilon)$. With much less noise added, the resulting differential privacy protected pseudo data sets allow much more accurate inferences, thus can significantly improve the scope of application for differential privacy.