Implementation and Analysis of Federated Learning Framework for Eco-Friendly Smart Services
With the rapid growth of artificial intelligence, the energy consumption and carbon emissions of distributed learning systems have become critical concerns. This thesis presents a federated learning (FL) framework integrating Docker, the Flower library, and OMNeT++/INET to simulate a network of 15 c...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Fecha de publicación: | 2025 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:upcommons.upc.edu:2117/450256 |
| Acceso en línea: | https://hdl.handle.net/2117/450256 |
| Access Level: | acceso abierto |
| Palabra clave: | Federated learning (Machine learning) Machine learning Federated Learning Aprenentatge automàtic Àrees temàtiques de la UPC::Informàtica::Intel·ligència artificial::Aprenentatge automàtic |
| Sumario: | With the rapid growth of artificial intelligence, the energy consumption and carbon emissions of distributed learning systems have become critical concerns. This thesis presents a federated learning (FL) framework integrating Docker, the Flower library, and OMNeT++/INET to simulate a network of 15 clients and a central server within containerized environments. We model diverse network conditions—varying topologies, bandwidth limits, latency, and client churn—and employ both CodeCarbon and custom measurement scripts to quantify communication energy costs, resource utilization, and digital carbon footprint (DCF). Experimental results demonstrate that our FL setup maintains high model accuracy under unstable client availability while significantly reducing carbon emissions. Comparative analysis reveals that our cross-validated estimation method aligns more closely with real-world network characteristics than CodeCarbon alone. These findings offer practical guidance for designing energy-efficient, low-carbon intelligent services. |
|---|