The machine learning community is increasingly aware of the environmental impact of data processing, especially since the emergence
of deep learning models. The increase power consumption to train and deploy deep leanring models has set the machine learning community
to consider efficiency as a key aspect of their designs. However, recent studies [1, 2, 3] and regulations [4] tend to shift the focus
towards a different concept: frugality.
This dashboard implements a Mamdani fuzzy inference system (FIS) [5] to compute a frugality score based on performance
and energy consumption. Fuzzy logic is particularly suitable for multi-criteria decision-making tasks involving uncertainty and
human subjectivity [6, 7].
As the performance and energy consumption of a system can be considered as human subjective concepts, a fuzzy-based approach
permits to formalize the reasoning process of the user and to compute an interpretable frugality score.
Context-aware sustainability
Frugality is context specific and aims to optimize the performance and energy consumption based on real-life interpretable rules. As such, frugality questions the objectives of an application, and can lead to more sustainable, realistic, and user-centric designs.
Performance-to-energy ratio
Efficiency focuses on the trade-off between performance and resource use, here energy consumption. Widely present in the literature, this concept takes little to no consideration of the context of use.
How to use this dashboard?
You can toggle DEMO MODE in the sidebar to choose an execution scenario applied on the data analysed within the linked paper.
1. Training of non-neural-network based classification methods on MNIST
2. Training of benchmark neural networks on ImageNet from scratch
3. Training of benchmark neural networks on CIFAR100 from scratch
4. Training of benchmark neural networks on CIFAR100 pretrained on ImageNet
5. Training of rectangular-shaped multi-layer perceptrons with different depths on MNIST (from BUTTER-E [8])
[1] N. Girdhar, A. Raj, D. Sharma, V. Singh, A. Doucet, and M. Renz, “A comprehensive review of frugal artificial intelligence: challenges,
applications, and the road to sustainable AI,” Soft Computing, vol. 29, pp. 4823–4856, July 2025.
[2] R. Basu, P. Banerjee, and E. Sweeny, “Frugal Innovation: Core Competencies to Address Global Sustainability,” Journal of Management for
Global Sustainability, vol. 1, Dec. 2013.
[3] IPCC, “Summary for policymakers,” in Climate Change 2022: Mitigation of Climate Change. Contribution of Working Group III to the Sixth
Assessment Report of the Intergovernmental Panel on Climate Change (P. Shukla, J. Skea, R. Slade, A. A. Khourdajie, R. van Diemen, D. McCollum,
M. Pathak, S. Some, P. Vyas, R. Fradera, M. Belkacemi, A. Hasija, G. Lisboa, S. Luz, and J. Malley, eds.), Cambridge, UK and New York, NY,
USA: Cambridge University Press, 2022.
[4] AFNOR, “AFNOR SPEC 2314: Référentiel général pour l'IA frugale – Mesurer et réduire l'impact environnemental de l'IA,” AFNOR Éditions, 2024.
[5] E. H. Mamdani and S. Assilian, “An experiment in linguistic synthesis with a fuzzy logic controller,” International Journal of Man-Machine
Studies, vol. 7, pp. 1–13, Jan. 1975.
[6] F. A. Lootsma, Fuzzy Logic for Planning and Decision Making. Springer Science & Business Media, Mar. 2013. Google-Books-ID: 7GXmB-wAAQBAJ.
[7] L. Chen, G. Duan, S. Wang, and J. Ma, “A Choquet integral based fuzzy logic approach to solve uncertain multi-criteria decision making
problem,” Expert Systems with Applications, vol. 149, p. 113303, July 2020.s
[8] C. Tripp, J. Perr-Sauer, E. Bensen, J. Gafur, A. Nag, and A. Purkayastha, “Butter-e - energy consumption data for the butter empirical deep
learning dataset.” Open Energy Data Initiative (OEDI), National RenewableEnergy Laboratory, https://doi.org/10.25984/2329316, 2022. Accessed:
2026-01-13.
| E ↓ P → | P = low | P = medium | P = high |
|---|