Abstract
[Google Meet]
https://meet.google.com/tpc-jsyg-xzy
The expansion of wearable technologies has transformed health monitoring from isolated snapshots into continuous streams of high-dimensional, multimodal physiological data. Yet these signals are highly nonstationary with variable qualities, making it difficult to extract reliable and physiologically meaningful information. To address such challenges, many may turn to deep learning. While these methods have achieved remarkable success in industrial applications, its yet-black-box nature poses serious risks in medicine. Even when such models accurately predict certain indices, they often generate predictions devoid of understanding. This inherent opacity does not advance science, particularly the kind needed in healthcare, but serves instead as a warning. Mathematics, grounded in scientific principles, offers a path forward. By embedding AI in rigorous theory, we can reveal interpretable physiological structures and the governing rules of our bodies. This talk will demonstrate how "mathematical learning" transforms raw physiological signals into trustworthy insights, showing that interpretability and scientific understanding with mathematics, not merely data and computation, lie at the heart of tomorrow's medical AI.
https://meet.google.com/tpc-jsyg-xzy
The expansion of wearable technologies has transformed health monitoring from isolated snapshots into continuous streams of high-dimensional, multimodal physiological data. Yet these signals are highly nonstationary with variable qualities, making it difficult to extract reliable and physiologically meaningful information. To address such challenges, many may turn to deep learning. While these methods have achieved remarkable success in industrial applications, its yet-black-box nature poses serious risks in medicine. Even when such models accurately predict certain indices, they often generate predictions devoid of understanding. This inherent opacity does not advance science, particularly the kind needed in healthcare, but serves instead as a warning. Mathematics, grounded in scientific principles, offers a path forward. By embedding AI in rigorous theory, we can reveal interpretable physiological structures and the governing rules of our bodies. This talk will demonstrate how "mathematical learning" transforms raw physiological signals into trustworthy insights, showing that interpretability and scientific understanding with mathematics, not merely data and computation, lie at the heart of tomorrow's medical AI.
Bio
Dr. Wu received his PhD in Mathematics from Princeton University in 2011, under the supervision of Professor Ingrid Daubechies (and unofficially Professor Amit Singer). Following his postdoctoral research position from 2011 to 2014, Dr. Wu joined the University of Toronto's Department of Mathematics as a tenure-track Assistant Professor until June 2017. From there, he transitioned to Duke University, initially as a tenured Associate Professor in the Department of Mathematics and Department of Statistical Science (from June 2017 to July 2022) and later as a tenured Professor (from August 2022 to August 2023). In August 2023, he joined New York University as a tenured Professor at the Courant Institute of Mathematical Sciences. Before coming to mathematics, Dr. Wu received his medical degree from National Yang Ming University (now National Yang Ming Chiao Tung University), Taiwan, in 2003. Following his medical studies, he practiced as a resident doctor at Taipei Veterans General Hospital, Taiwan. Dr. Wu was awarded the Sloan Research Fellowship in 2015 and the PIMS Early Career Award by the Canadian Applied and Industrial Society (CAIMS) in 2017. His research interests encompass a broad spectrum, spanning mathematics, statistics, biomedical engineering, and medicine, with a particular emphasis on biomedical signal processing, theoretical advancements, and their clinical applications.