This project builds novel methods for deep neural network inference to achieve a target accuracy under the resource constraints of embedded devices. The developed technical solutions enhance the device functionalities and extend the applicability of solutions based on artificial intelligence to diverse scenarios. The devised methods allow to run machine learning tasks on resource constrained devices, therefore, they reduce the reliance on cloud computing providers; they also ensure data ownership and privacy. The conceived solutions facilitate the integration of embedded systems with value-added services, thus fostering the creation of new digital ecosystems in key sectors including the Industrial Internet. Finally, the outcomes of the project enable a greater level of automation in the industry, resulting in improved utilization of natural resources and less waste generation; efficient computation also reduces the per-user carbon footprint, thereby supporting a resource-wise economy.

2023

2027