Subject

This master’s thesis proposes the exploration and implementation of generative AI (GenAI) models on low-power devices, ranging from single embedded systems and mobile devices to more capable platforms such as embedded GPUs. The motivation stems from the growing need to reduce the energy footprint of GenAI applications while enhancing data privacy by enabling offline, local execution. The research will address the technical challenges associated with deploying complex models like large language models (LLMs) and transformers on constrained hardware. Key areas of investigation will include understanding the architectures of GenAI systems, evaluating and developing frameworks for efficient offline deployment. This work aims to contribute to sustainable and privacy-preserving AI by bridging the gap between cutting-edge GenAI and practical, energy-efficient applications in real-world low-power environments.

Kind of work

The work to be done involves:

1.- Literature study
2.- Evaluation of the most interesting solutions and frameworks
3.- Identification of the main limitation and explore novel approaches
4.- Implement and test for a realistic use case (TBD) the proposed approaches for edge deployment of GenAI
5.- Writing the thesis and preparing the public defense

Number of Students

1 - 2

Expected Student Profile

• Interested/experience in AI and embedded/edge technologies.
• Experience in programming languages such as C/C++ and python.