Sustainable AI in the Cloud: Exploring Machine Learning Energy Use in the Cloud

Abstract

In light of the increasing urgency regarding climate change due to man-made greenhouse gas emissions, focus is now being brought to bear on the amount of energy that artificial intelligence (AI) applications consume. Research has highlighted the immense carbon footprint of machine learning (ML) driven applications, due to the extraordinary growth in the size of deep learning models, which are estimated to have grown by a factor of 300,000 over the last six years. This is a concern, so we wish to add our voice to a growing community of responsible AI researchers and practitioners and help highlight how energy awareness and responsible best practices can be used to enhance the environmental sustainability of AI. Hence, we provide a preliminary exploration of the energy use profile of ML training in the cloud and demonstrate how transfer learning can be used to reduce this energy consumption.