Inferring Human Behavior and Health from Imperfect Wearable Signals by Hao Zhou

Abstract: While modern healthcare remains largely episodic and expensive, human health unfolds continuously in daily life. Low-cost wearable devices offer a unique opportunity to bridge this gap; however, current systems are often hindered by noisy sensing, partial observability, and limited generalization beyond controlled environments. This raises a fundamental challenge: can ubiquitous, low-cost wearables reliably infer human states and deliver clinically meaningful endpoints?

In this talk, I answer this question by co-designing everyday wearable sensing and structure-aware learning grounded in anatomy, biomechanics, and physiology. Rather than treating wearable data as generic time series, I embed domain structure into model design. This transforms imperfect, partial observations into robust and interpretable inference for fine-grained motion analysis, muscular activity modeling, and cardiovascular health assessment, moving toward scalable real-world deployment.

Looking forward, I outline a vision for intelligent and trustworthy wearable health computing built on three pillars: (1) novel, low-cost sensing paradigms, (2) multimodal reasoning for behavior and health, and (3) privacy-preserving on-device AI. By leveraging the synergy between computer science and public health, this vision shifts the paradigm from episodic care to accessible, proactive, and inclusive monitoring, evolving wearables into universal health companions that enable earlier clinical intervention and broaden equitable access to care worldwide.

Bio: Hao Zhou is a Ph.D. candidate in Computer Science and Engineering at The Pennsylvania State University, advised by Professor Mahanth Gowda. His research focuses on AI-powered mobile and wearable systems and time-series representation learning, with an emphasis on understanding human behavior and health from low-cost, everyday sensors. He has published extensively at leading venues including ACM MobiCom, IMWUT/UbiComp, MobiSys, IoTDI, ICLR, and ICASSP, and his work has received multiple honors, including the Best Paper Award on IoT Edge AI at IoTDI, the MobiSys Rising Star, and the Penn State Alumni Association Dissertation Award. His work advances both accessible wearable computing, including sign language recognition in collaboration with Gallaudet University, and scalable smart health systems through large-scale wearable foundation models developed in collaboration with Samsung Research America.