Sustainable Distributed ML Paradigms for Green Edge Intelligence: Analysis and Framework

The proliferation of Internet of Things devices and emphasis on sustainable computing have highlighted the issue of high energy consumption and carbon footprints in traditional centralized machine learning, posing challenges for green edge intelligence. This research presents a systematic comparative analysis of twelve distributed ML paradigms designed for energyefficient edge environments, focusing on sustainability. The analysis evaluates metrics such as latency, privacy, energy use, carbon footprint, and scalability, resulting in a framework for paradigm selection. Findings reveal trade-offs, with federated learning using green scheduling achieving optimal balance through $\mathbf{9 4. 2 \%}$ semantic preservation and $37 \%$ carbon reduction, while neuromorphic computing suits ultra-low energy needs with submilliwatt consumption. Insights guide sustainable deployments in healthcare, autonomous systems, and smart cities, advancing global sustainability in electronics and communications.