Towards Ultra-low Power Wearable Health Sensing with Sparse Sampling and Asymmetric Communication

Abstract

Wearable sensors offer tremendous opportunities for accelerating biomedical discovery, and improving population-scale health and wellness. There is a growing appetite for health analytics -- we are no longer content with wearables that count steps and calories, we want to measure physiology, behavior, activities, cognition, affect, and other parameters with the expectation that such data will lead to deep insights that can improve quality of life. But a chasm separates expectations and reality. How do we extract such insights from sensor platforms with tiny energy budgets? How do we communicate high-rate sensor data to the cloud for enabling deep analytics while operating within these energy budgets? How do we deal with noise, confounders, and artifacts that make insights hard to extract from signals collected in real-world settings? In this talk, I will discuss a few strategies to tackle these problems. I will discuss how we can design an low-power computational eyeglass that continually tracks eye and visual context by leveraging sparsity, how we can transfer data at Megabits/second from wearables while operating at tens of micro-watts of power, and how we can leverage these techniques in the context of mobile health.