The ODEON consortium is proud to announce a new scientific publication authored by partners from the National Technical University of Athens (NTUA), Intracom Telecom, and Barbara IoT, recently published in Energy Reports (Elsevier), titled “Isomorphic structured pruning of temporal CNNs for scalable NILM on edge devices”
A breakthrough in scalable AI for energy monitoring
This research introduces Isomorphic Structured Pruning (ISP), a novel method to optimize Deep Learning (DL) models for Non-Intrusive Load Monitoring (NILM) in resource-constrained edge environments.
By grouping and ranking equivalent substructures within neural networks, the ISP approach achieves remarkable efficiency improvements:
- Up to 97% reduction in model size
- 42× computational efficiency gains
- Over 85% lower CPU usage and energy consumption
All these gains are achieved with negligible loss in accuracy, making ISP a powerful solution for real-world deployment of NILM algorithms on edge devices.
Contributing to ODEON’s vision
These results illustrate how cutting-edge AI research can make energy disaggregation both scalable and efficient, paving the way for smarter, greener, and more distributed energy systems — a core ambition of the ODEON project.
Read the full paper: ScienceDirect