# Energy and Policy Considerations for Deep Learning in NLP **Authors**: Emma Strubell, Ananya Ganesh, Andrew McCallum **Year**: 2019 **Source**: arXiv:1906.02243 **Published at**: ACL 2019 **URL**: https://arxiv.org/abs/1906.02243 ## Summary One of the first papers to draw systematic attention to the energy and carbon cost of training large deep learning models for [[NLP|natural language processing]]. By measuring the electricity consumed during training runs for several common architectures, the authors quantified costs that had previously gone unreported, establishing the vocabulary of training vs. inference as distinct cost categories and prompting the broader field to treat energy as a first-class metric. Often cited alongside [[Luccioni et al. (2024)]], which extended the analysis to the inference phase.