A large amount of energy and computational cost is required to train AI models in a centralized way, raising concerns about its carbon footprint. Total amount of energy consumed by data centres made up about 1% of global energy use over the past decade – equalling roughly 18 million US homes.
Green AI refers to AI that yields novel results while taking into account computational cost and environmental impact.
Organizations need to conduct rigorous exercises to determine the environmental impact and carbon footprint of, among other things, their model training, infrastructure and compute usage.
Federated Learning decentralizes data processing which reduces the number of communications and requires less energy consumption which decreases carbon footprint.