The project aims to advance novel data-efficient machine learning techniques for modelling sensor data obtained by resource-constrained devices, such as microcontrollers, wearables, mobile devices, and IoT. These data are typically high in dimension, noise and uncertainty, and highly varied in its modality, purpose and tasks, ranging from end-consumer wearable sensors and devices capturing human activities and physiological signals to infrastructure sensors and IoT capturing large-scale urban flows and mobility patterns.
This program seeks to enable modelling of human behaviours more efficiently at the edge, with little or no labelled data. It expects to make major breakthroughs in the modelling of human behaviour through a generalised representation learning techniques for different downstream tasks and domains, including transportation and mobility, building construction and management, agriculture and mining, and health and productivity.
The program involves three Australian leading institutions, UNSW, Curtin, and ANU, with strong support from six industry partners: Aurecon, STSA (a subsidiary of SoftBank Japan), SAP, N2N.AI, and HugHealth. The program seeks to train a cohort of twelve students, including four PhD, one Master by Research, and seven Honours students, with the goal to train them on developing and deploying AI on the edge, and embed them in industry-relevant problems.