Dartmouth Events

ZOOM LINK
Meeting ID: 977 7792 8244
Passcode: 813105

Wearable medical devices are expected to provide automatic monitoring and processing of physiological signals, and more importantly, to be capable of identifying abnormal signals and contacting medical systems when necessary. Such devices are the key components in the future "unmanned medical nursing systems." Machine learning is a promising solution in digital health systems and has recently been applied to the continuous monitoring of physiological signals with on-sensor processing. Machine learning algorithms that can accommodate real-time processing without too much data storage and data movement are preferred in wearable sensor applications.

This talk presents novel smart wearable wireless biosensor frameworks to address sensing, processing, and communication problems systematically by exercising three complementary research objectives:

1. Analog-to-feature converter for electrocardiography (ECG) sensing based on dynamic predictive level-crossing sampling sensors; 
2. Linear-kernel rotation support vector machine for patient-specific data with on-sensor inference for arrhythmia detection; and
3. Power-efficient encoder and decoder for wireless asynchronous transmission of oversampling data.

The key idea is to exploit the advantages of combining oversampling sensors and power-efficient machine learning algorithms for a hardware-friendly framework in wireless biomedical sensors.