Skip to content

Machine Learning @ the Extreme Edge

Principal investigator: dr. ir. J. R. Verbiest

Today's challenges are real-time and energy-efficient information extraction and processing at the edge using Artificial Intelligence. However, there is a recent trend to implement machine learning on end-point devices. These devices are located on the extreme edge, the border between the analog (physical world) and the digital world. They consist of one or more sensors and a resource-constrained embedded device, a device with a limited amount of memory, computing power, and power consumption.

The challenge is the development of accurate, energy-efficient machine learning models for deployment on these end-point devices.


workflow


Machine Learning at the Extreme Edge (ML@E2dge) looks at how a developer can apply machine learning in the development of accurate, energy-efficient, and intelligent (wireless) (battery-powered) end-point devices and systems. A machine learning model is developed, optimized, and deployed. We limit ourselves to a supervised machine learning regression problem. The project uses open-source software frameworks such as scikit-learn, TensorFlow, and Keras combined with Edge Impulse Studio. For experiment tracking, we use the MLOps tool Weights and Biases.

Project Supervisory Group

  • Zorginstelling Heder, Antwerp, Belgium.
  • Centre for Health and Technology, University of Antwerp, Belgium
  • Capetech BVBA, Leuven, Belgium
  • Faculty of Rehabilitation Sciences, Rehabilitation Research Group (Reval), Hasselt University, Belgium
  • Department of Rehabilitation Sciences and Physiotherapy Research group MOVANT, University of Antwerp, Belgium
  • Edge Impulse Inc. San Jose, USA.
  • Comate Engineering & Design BVBA, Leuven, Belgium
  • VR Base, Mechelen, Belgium.
  • Hersenletsel Liga.
  • Centre of Expertise The Cycle of Care, Karel de Grote University of Applied Sciences and Arts, Antwerp, Belgium.


The supervisory group operates as a soundboard toward the implementation possibilities of the project results.


This research is supported by the Karel de Grote University of Applied Sciences and Arts through funding by the Flemish government specifically allocated to practice-based research at universities of applied sciences.