Page 32 - SMILESENG

Intl. Summer School on Search- and Machine Learning-based Software Engineering
correct predictions for the models trained on the synthetic data was less than 10% lower than for the models trained with real data.
B. Choice of model
It is known that some ML models are more suited to perform certain tasks, while being less apt for others. For this case, we propose that the choice of a ML model that is adequate for the task at hand is also a relevant choice for its energy efficiency, and not just for its performance.
REFERENCES
[1] Albert Bifet et al. “MOA: Massive Online Analysis”. In: The Journal of Machine Learning Research 11 (Aug. 2010), pp. 1601–1604. ISSN: 1532-4435.
[2] Eva Garc´ıa-Mart´ın et al. “Estimation of energy consump- tion in machine learning”. en. In: Journal of Parallel and Distributed Computing 134 (Dec. 2019), pp. 75–88. ISSN: 07437315. DOI: 10.1016/j.jpdc.2019.07.007. URL: https://linkinghub.elsevier.com/retrieve/pii/ S0743731518308773 (visited on 01/26/2022).
[3] Javier Mancebo et al. “FEETINGS: Framework for En- ergy Efficiency Testing to Improve Environmental Goal of the Software”. en. In: Sustainable Computing: Infor- matics and Systems 30 (June 2021), p. 100558. ISSN: 2210-5379. DOI: 10.1016/j.suscom.2021.100558. URL: https : / / www . sciencedirect . com / science / article / pii / S2210537921000494 (visited on 02/04/2022).
[4] Parallel File System Products — WEKA. URL: https: / / www . weka . io / parallel - file - system/ (visited on 06/10/2022).
[5] Roy Schwartz et al. “Green AI”. en. In: Communications of the ACM 63.12 (Nov. 2020), pp. 54–63. ISSN: 0001- 0782, 1557-7317. DOI: 10.1145/3381831. URL: https:// dl.acm.org/doi/10.1145/3381831 (visited on 01/25/2022).
20