Towards the Systematic Reporting of the Energy and Carbon Footprints of Machine Learning

By Peter Henderson, Jie Hu, Joshua Romoff, Emma Brunskill, Dan Jurafsky, Joelle Pineau. 2020

724 citations recorded by the source corpus.

Accurate reporting of energy and carbon usage is essential for understanding the potential climate impacts of machine learning research. We introduce a framework that makes this easier by providing a simple interface for tracking realtime energy consumption and carbon emissions, as well as generating standardized online appendices. Utilizing this framework, we create a leaderboard for energy efficient reinforcement learning algorithms to incentivize responsible research in this area as an example for other areas of machine learning. Finally, based on case studies using our framework, we propose strategies for mitigation of carbon emissions and reduction of energy consumption. By making accounting easier, we hope to further the sustainable development of machine learning experiments and spur more research into energy efficient algorithms.

Subfields: Computer Science Applications, Electrical and Electronic Engineering, Renewable Energy, Sustainability and the Environment, Efficiency & Compression, AI Ethics, Fairness & Society

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Science Applications

In Electrical and Electronic Engineering

In Renewable Energy, Sustainability and the Environment

In Efficiency & Compression

In AI Ethics, Fairness & Society

Citations in the corpus

AI and the Everything in the Whole Wide World Benchmark (2021, methodology): “• Techniques for analyzing model properties (see Appendix A.4) that are orthogonal to system outputs such as proﬁling energy consumption [e.g. Henderson et al., 2020, Schwartz et al., 2020], memory requirements [e.g. Ethayarajh and Jurafsky, 2020], and stability in the face of perturbations to…”
Carbon Emissions and Large Neural Network Training (2021, methodology): “An example of an open source tool to record such information is experiment-impact-tracker [Hen20].”
Evaluating the Social Impact of Generative AI Systems in Systems and Society (2023, background): “For instance, [132] proposes both a web-based and programmatic approach for quantifying the carbon emissions of models, meanwhile [104] proposes an experiment-impact-tracker, for energy and carbon usage reporting research.”
Chasing Carbon: The Elusive Environmental Footprint of Computing (2020, methodology): “Powering hardware with renewable energy reduces emissions from operational energy consumption; consequently, [14], [76], [77]. hardware manufacturing begins to dominate the carbon foot-print.”
When BERT Plays the Lottery, All Tickets Are Winning (2020, neutral): “, 2018) and this statement and emissions information was generated with experiment-impact-tracker (Henderson et al., 2020).”
FastFormers: Highly Efficient Transformer Models for Natural Language Understanding (2020, methodology): “The energy efﬁciency is measured by an open source python library called experiment-impact-tracker proposed in (Henderson et al., 2020).”
Sponge Examples: Energy-Latency Attacks on Neural Networks (2020, background): “Most of the prior research on the carbon footprint of machine learning focuses on the energy required to train large neural network models and its contribution to carbon emissions [53, 54, 55].”
The Carbon Footprint of Machine Learning Training Will Plateau, Then Shrink (2022, background): “More than 1000 papers cite this paper as the source for the impact on carbon emissions of ML models, e.g., [1,6,7,17,18,19,20].”
Unraveling the hidden environmental impacts of AI solutions for environment (2021, neutral): “• Integrated tools, such as Experiment Impact Tracker 1 [17], Carbon Tracker 2 [2] and CodeCarbon 3, which are all Python packages reporting measured energy consumption and the associated carbon footprint.”
Computational bioacoustics with deep learning: a review and roadmap (2021, background): “For DL, impacts are incurred while training a NN, and while applying it in practice: their relative signiﬁcance depends on whether training or inference will be run many times (Henderson et al., 2020).”
Carbontracker: Tracking and Predicting the Carbon Footprint of Training Deep Learning Models (2020, methodology): “The methodology is similar to that of Henderson et al. (2020) but differs from prior art in two major ways: (1) We allow for a further proactive and intervention-driven approach to reducing carbon emissions by supporting predictions.”
Dynaboard: An Evaluation-As-A-Service Platform for Holistic Next-Generation Benchmarking (2021, positive): “Second, as a ﬁeld, we need to focus more on Green AI and explicitly account for a model’s energy/carbon footprint in our model ranking [63, 68, 28].”
Recovering Petaflops in Contrastive Semi-Supervised Learning of Visual Representations (2020, background): “Burdensome computational requirements not only make training laborious and particularly timeand energy-consuming; they also exacerbate other issues, making it more difficult to scale to more complex models and problems, and potentially inducing significant carbon footprints depending on the infrastr”
Sustainable AI: AI for sustainability and the sustainability of AI (2021, methodology): “…footprint of GPU compute through specifying hardware type, hours used, cloud provider, and region [12], Anthony et al. [3]; and, the ‘experiment-impact-tracker’ framework “for tracking real-time energy consumption and carbon emissions, as well as generating standardized online appendices” [10, 1].”
Large Language Models for Software Engineering: Survey and Open Problems (2023, background): “It has been pointed out that the training and developing of larger models may have direct environmental impact [219].”
Pile of Law: Learning Responsible Data Filtering from the Law and a 256GB Open-Source Legal Dataset (2022, background): “[54] Peter Henderson, Jieru Hu, Joshua Romoff, Emma Brunskill, Dan Jurafsky, and Joelle Pineau.”