Recognising and characterising human activity from wearable inertial signals with machine learning, including deep and self-supervised models. Data-driven classification and modeling of human activity.

Objectives

1. Activity classification and temporal segmentation from wearable IMU signals.

2. Activity recognition from thigh, wrist and upper-body accelerometry

3. Self-supervised and transfer learning to reduce labelling needs

4. Context inference in IMU sensing

5. Robustness to sensor placement, subject and protocol variability