The emergence of specialized edge devices for machine learning tasks has enabled the efficient processing and classification of data collected by resource-constrained devices in the Internet of Things. However, with the increasing demand for applications like critical monitoring in smart cities, it has become necessary to develop strategies to make these systems sustainable in terms of energy consumption.
This paper introduces an energy-aware approach to design and deploy self-adaptive AI-based applications that can balance application objectives, such as accuracy in object detection and frames processing rate, with energy consumption. The problem of determining the set of configurations required for self-adaptation is addressed using a meta-heuristic search procedure that requires only a small number of empirical samples. The final set of configurations is selected using weighted gray relational analysis and then mapped to the operation modes of the self-adaptive application.
The proposed approach is validated using an AI-based application for pedestrian detection. The results demonstrate that the self-adaptive application can outperform non-adaptive baseline configurations by saving up to 81% of energy while only sacrificing between 2% and 6% in accuracy.