Citation :

Hanut Lal, "Investigating the Interactive Effects of Training-Efficiency Patterns on Model Accuracy and CO₂ Emissions," International Journal of Computer Science and Engineering, vol. 13, no. 8, pp. 16-23, 2026. Crossref, https://doi.org/10.14445/23488387/IJCSE-V13I8P103

Abstract

The more energy uses a training machine-learning model, the more energy it consumes, but most of the efficiency techniques are only tested individually. The study explores the feasibility of using the three methods — sample triage, patient early exit, and delayed precision control — in order to lower CO₂ emission without compromising accuracy. The synthetic binary-classification data was used in a controlled experiment, employing CodeCarbon, three architectures (MLP-Small, MLP-Large, XGBoost) and four conditions (n = 10 runs per condition). For XGBoost, the composite condition showed a 63.4% reduction in emissions and a 18.8% reduction in accuracy (p < 0.001), which was not statistically different from the early-exit-only condition (p = 0.970). The increase in emissions due to sample triage was 82.1%. The combination of the two conditions resulted in 38% emission reductions for MLP-Large while preserving the accuracy considerably. The interaction of techniques is very model-dependent: some techniques interact in a beneficial way, some in a destructive way, and one technique (triage) even made emissions worse in the XGBoost setting. The results validate the benefits of testing efficiency patterns on a per-model basis before deployment and highlight the pitfalls of assuming that efficiency patterns can be stacked to improve their efficiency further.

Keywords

Machine learning, Training efficiency, CO₂ emissions, Green AI, Early stopping, Sample triage, XGBoost, Reproducibility, Ethical AI.

References

1. Roy Schwartz et al., “Green AI,” Communications of the ACM, vol. 63, no. 12, pp. 54-63, 2020.
   [CrossRef] [Google Scholar] [Publisher Link]
2. Emma Strubell, Ananya Ganesh, and Andrew McCallum, “Energy and Policy Considerations for Deep Learning in NLP,” Proceedings of the 57^th Annual Meeting of the Association for Computational Linguistics, Florence, Italy, pp. 3645-3650, 2019.
   [CrossRef] [Google Scholar] [Publisher Link]
3. David Patterson et al., “Carbon Emissions and Large Neural Network Training,” arXiv preprint, pp. 1-22, 2021.
   [CrossRef] [Google Scholar] [Publisher Link]
4. Yoshua Bengio et al., “Curriculum Learning,” Proceedings of the 26^th Annual International Conference on Machine Learning (ICML '09), Montreal, Quebec, Canada, pp. 41-48, 2009.
   [CrossRef] [Google Scholar] [Publisher Link]
5. Garvesh Raskutti, Martin J. Wainwright, and Bin Yu, “Early Stopping and Non-parametric Regression: An Optimal Data-Dependent Stopping Rule,” Journal of Machine Learning Research, vol. 15, no. 11, pp. 335-366, 2014.
   [Google Scholar] [Publisher Link]
6. Nesrine Bannour et al., “Evaluating the Carbon Footprint of NLP Methods: A Survey and Analysis of Existing Tools,” Proceedings of the Second Workshop on Simple and Efficient Natural Language Processing, pp. 11-21, 2021.
   [CrossRef] [Google Scholar] [Publisher Link]
7. Tianqi Chen, and Carlos Guestrin, “XGBoost: A Scalable Tree Boosting System,” Proceedings of the 22^nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Francisco, California, USA, pp. 785-794, 2016.
   [CrossRef] [Google Scholar] [Publisher Link]
8. Yuan Yao, Lorenzo Rosasco, and Andrea Caponnetto, “On Early Stopping in Gradient Descent Learning,” Constructive Approximation, vol. 26, no. 2, pp. 289-315, 2007.
   [CrossRef] [Google Scholar] [Publisher Link]
9. GuyHacohen, and Daphna Weinshall, “On the Power of Curriculum Learning in Training Deep Networks,” Proceedings of the 36^th International Conference on Machine Learning (ICML), pp. 2535-2544, 2019.
   [Google Scholar] [Publisher Link]
10. Alfredo Canziani, Adam Paszke, and Eugenio Culurciello, “An Analysis of Deep Neural Network Models for Practical Applications,” arXiv preprint, pp. 1-7, 2017.
    [CrossRef] [Google Scholar] [Publisher Link]
11. Marion Dörrich, Mingcheng Fan, and Andreas M. Kist, “Impact of Mixed Precision Techniques on Training and Inference Efficiency of Deep Neural Networks,” IEEE Access, vol. 11, pp. 57627-57634, 2023.
    [CrossRef] [Google Scholar] [Publisher Link]
12. Victor Schmidt et al., “CodeCarbon: Estimate and Track Carbon Emissions from Machine Learning Computing,” Zenodo, 2021.
    [CrossRef] [Publisher Link]
13. Norman Cliff, “Dominance Statistics: Ordinal Analyses to Answer Ordinal Questions,” Psychological Bulletin, vol. 114, no. 3, pp. 494-509, 1993.
    [CrossRef] [Google Scholar] [Publisher Link]
14. Andras Vargha, and Harold D. Delaney, “A Critique and Improvement of the CL Common Language Effect Size Statistics of McGraw and Wong,” Journal of Educational and Behavioral Statistics, vol. 25, no. 2, pp. 101-132, 2000.
    [CrossRef] [Google Scholar] [Publisher Link]
15. Lasse F. Wolff Anthony, Benjamin Kanding, and Raghavendra Selvan, “Carbontracker: Tracking and Predicting the Carbon Footprint of Training Deep Learning Models,” arXiv preprint, pp. 1-11, 2020.
    [CrossRef] [Google Scholar] [Publisher Link]
16. Alexandre Lacoste et al., “Quantifying the Carbon Emissions of Machine Learning,” arXiv Preprint, pp. 1-8, 2019.
    [CrossRef] [Google Scholar] [Publisher Link]
17. Alexandra Sasha Luccioni, and Alex Hernandez-Garcia, “Counting Carbon: A Survey of Factors Influencing the Emissions of Machine Learning,” arXiv preprint, pp. 1-19, 2023.
    [CrossRef] [Google Scholar] [Publisher Link]
18. Sasha Luccioni, Yacine Jernite, and Emma Strubell, “Power Hungry Processing: Watts Driving the Cost of AI Deployment?,” Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency, Rio de Janeiro, Brazil, pp. 85-99, 2024.
    [CrossRef] [Google Scholar] [Publisher Link]