Implementation of measurement of power efficiency of computing platforms

Context

Running Machine Learning (ML) models is often computation- and resource-intense, as well as it is difficult to deploy on embedded devices with limited energy budget. To run ML models on these devices, numerous optimization and deployment techniques exist. However, the majority of these techniques does not consider the power consumption metrics of the target devices. Accurate acquisition of these metrics into existing ML frameworks allows to optimize the models for low-power target devices.

Goals 

In this thesis the methodology of the power efficiency measurement of integrated AI systems will be explored. The hardware setup for performing measurements of power consumption, execution time and model accuracy will be implemented. To evaluate the design, several hardware platforms will be compared running different AI models.

Requirements

• Programming experience in Python and C/C++;
• Basic knowledge of neural networks;
• Basic knowledge of microcontrollers;
• Motivation and interest in solving technical problems independently.