Project Description

The project investigates the possibilities for more efficient training of deep learning models at the edge. It will explore how low-power devices can contribute to the training of models without the use of high-powered GPUs. Approaches such as federated learning will be explored, within the constraints of these edge computing devices. The project can include general modeling and simulation, or actual implementation of models and training on a variety of edge devices including microcontrollers, systems-on-chip, and FPGAs. Students should have some experience of working on such architectures.

Project Deliverables

Experimental testbed to evaluate alternative approaches and results for different datasets.