Eco2AI: carbon emissions tracking of machine learning models as the first step towards sustainable AI

Budennyy, Semen; Lazarev, Vladimir A.; Zakharenko, Nikita; Korovin, Alexey V.; Plosskaya, Olga; Denis, Dimitrov,; Arkhipkin, Vladimir; Oseledets, Ivan; Barsola, Ivan; Egorov, Ilya N.; Kosterina, Aleksandra; Zhukov, Leonid

doi:10.48550/arxiv.2208.00406

Cited by 2 publications

(2 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…All models (except MOFTransformer) were trained for 1000 (300) epochs with early stopping at 100. GHG emissions of model training and inference were measured by means of the Eco2AI package [56], suggesting the emission intensity coefficient of 240.56 kgCO 2 e per MWh (Moscow). All calculations were carried out on the workstation equipped with two Intel® Xeon® CPUs E5-2695 v4 @ 2.10GHz, 144 GB RAM, and NVIDIA GeForce RTX 3090 Ti.…”

Section: Modelsmentioning

confidence: 99%

Carbon Footprint of Artificial Intelligence in Materials Science: Should We Be Concerned?

Korolev,

Mitrofanov

2023

Preprint

View full text Add to dashboard Cite

While artificial intelligence is promoting remarkable progress in natural sciences, its broader societal implications are mostly disregarded. In this study, we evaluate the environmental impacts of deep learning in materials science through extensive benchmarking. Specifically, a diverse set of neural networks is trained for a particular supervised learning task to assess greenhouse gas (GHG) emissions during training and inference phases. A chronological perspective shows diminishing returns, which are expressed in a 28% decrease in mean absolute error and nearly a 15000% increase in carbon footprint of model training in 2016–2022. By utilizing up-to-date graphics processing units, it is possible to partially offset the immense growth in GHG emissions. Nonetheless, the practice to employ energy-efficient hardware is unaddressed by the materials informatics community, as it follows from the literature analysis in the field. Based on our findings, we encourage researchers to report GHG emissions on par with standard performance metrics.

show abstract

Section: Modelsmentioning

confidence: 99%

Carbon Footprint of Artificial Intelligence in Materials Science: Should We Be Concerned?

Korolev,

Mitrofanov

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Comparison of models' sizes and CO2 emission amounts as furnished by Eco2AI library. The CO2 emission related values allow for comparative analysis of the tested architecture[8]. Power consumption, CO2 emissions and amount of epochs are provided for the state-space synthetic dataset with pink noise (see section 2.4.1).…”

mentioning

confidence: 99%

Real-time low latency estimation of brain rhythms with deep neural networks

Semenkov,

Fedosov,

Makarov

et al. 2023

J. Neural Eng.

View full text Add to dashboard Cite

Objective. Neurofeedback and brain-computer interfacing technology open the exciting opportunity for establishing interactive closed-loop real-time communication with the human brain. This requires interpreting brain's rhythmic activity and generating timely feedback to the brain. Lower delay between neuronal events and the appropriate feedback increase the efficacy of such interaction. Novel more efficient approaches capable of tracking brain rhythm's phase and envelope are needed for scenarios that entail instantaneous interaction with the brain circuits. Approach. Isolating narrow-band signals incurs fundamental delays. To some extent they can be compensated using forecasting models. Given the high quality of modern time series forecasting neural networks we explored their utility for low-latency extraction of brain rhythm parameters. We tested five neural networks with conceptually distinct architectures in forecasting synthetic EEG rhythms. The strongest architecture was trained to simultaneously filter and forecast EEG data. We compared it against state-of-the-art techniques using synthetic and real data from 25 subjects.Main results. The Temporal Convolutional Network (TCN) remained the strongest forecasting model that achieved in the majority of testing scenarios >90% rhythm's envelope correlation with <10 ms effective delay and <20° circular standard deviation of phase estimates. It also remained stable enough to noise level perturbations. Trained to filter and predict the TCN outperformed the cFIR, the Kalman filter based state-space estimation technique and remained on par with the larger Conv-TasNet architecture. Significance. Here we have for the first time demonstrated the utility of the neural network approach for low-latency narrow-band filtering of brain activity signals. Our proposed approach coupled with efficient implementation enhances the effectiveness of brain-state dependent paradigms across various applications. Moreover, our framework for forecasting EEG signals holds promise for investigating the predictability of brain activity, providing valuable insights into the fundamental questions surrounding the functional organization and hierarchical information processing properties of the brain.

show abstract

Eco2AI: carbon emissions tracking of machine learning models as the first step towards sustainable AI

Cited by 2 publications

References 14 publications

Carbon Footprint of Artificial Intelligence in Materials Science: Should We Be Concerned?

Carbon Footprint of Artificial Intelligence in Materials Science: Should We Be Concerned?

Real-time low latency estimation of brain rhythms with deep neural networks

Contact Info

Product

Resources

About