Photoplethysmography (PPG) is widely used in wearables to record heart activity. Analysis of interbeat-intervals (IBI) inferred from PPG recordings provide a powerful means of assessing the activities of the autonomic nervous system (ANS) by observing the interplay between the sympathetic and parasympathetic contributions. Consumer health research has notably used wearable data for sleep, stress, recovery and more recently pain assessment. Leading approaches to evaluate these metrics involve selecting a set of manually engineered features from the IBI time series - e.g resting heart rate and pulse rate variability (PRV) - and feeding them to a small neural network.
Research in clinical diagnosis of cardiac diseases has established that end-to-end deep learning approaches perform best at classifying a wide range of heart arrhythmias from electrocardiography (ECG). We thus expect the use of end-to-end deep learning to achieve significant improvements in sleep, stress, recovery and pain assessment over present approaches. More generally we research how to unify these metrics into a more general measure of well-being.
The aim of this project is to design and build a unified end-to-end deep learning model that is capable of handling raw physiological data to assess and predict our well-being. You will work with a large set of physiological data (IBI and motion data) combined with rich context information and insights into our customers’ well-being inferred from our application. More specifically as part as your research you will study the following topics:
Semi-supervised end-to-end deep learning algorithms that can handle noisy (i.e. physiological and measurement artifacts), sparse (i.e. missing data and unreliable data points) and partially labeled IBI and motion time series collected by wearables
Multi-stage (e.g. global and individual) learning phases to optimize accuracy given the wide inter-patient variability
Comparison of learned feature representations of IBI data with standard PRV metrics
Assessment and prediction of well-being
Master’s degree in Artificial Intelligence, Mathematics, Physics, Computer Science or relevant field
Strong machine learning / deep learning background
Proficient programming skills in Python
You are insightful and you can think outside the box
You can work autonomously
Good communication skills (English)
Duration & Start Date
3 years starting January - September 2019
To apply to this position, please send the following materials to Dr Remi Paulin ([email protected]
resumé or curriculum vitae
cover letter describing your experience, research and professional interests
contact information of three academic / professional references
copy of your transcripts (optional)
Remedee Labs is a startup specialized in active health technology. By pioneering advanced scientific and technological breakthroughs, the company’s vision is to enable people to better use their body’s own resources and become central actors in their own health and well-being, in particular for the management of chronic pain, stress and sleep disorders. Using its expertise in micro and nanotechnologies and medical research, Remedee is turning this vision into a reality. Founded in 2016 in Grenoble, France, the company already boasts a multidisciplinary, international team of more than 20 people.