Pith. sign in

REVIEW 6 minor 48 references

Greening AI-enabled Systems with Software Engineering: A Research Agenda for Environmentally Sustainable AI Practices

T0 review · 0 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Software engineering methods are the missing toolkit for environmentally sustainable AI.

desk verdict A solid, honest workshop synthesis that usefully frames Green AI as a software engineering problem; the only real soft spot is the 'key challenges for the field' wording, which overstates the representativeness of 29 participants. read the letter →

arxiv 2506.01774 v2 pith:G4VQAHDQ submitted 2025-06-02 cs.SE cs.AI

Luís Cruz , João Paulo Fernandes , Maja H. Kirkeby , Silverio Martínez-Fernández , June Sallou , Hina Anwar , Enrique Barba Roque , Justus Bogner
show 17 more authors
Joel Castaño Fernando Castor Aadil Chasmawala Simão Cunha Daniel Feitosa Alexandra González Andreas Jedlitschka Patricia Lago Henry Muccini Ana Oprescu Pooja Rani João Saraiva Federica Sarro Raghavendra Selvan Karthik Vaidhyanathan Roberto Verdecchia Ivan P. Yamshchikov
This is my paper
classification cs.SEcs.AI
keywords greenAIsoftwareengineeringenergyassessmentbenchmarkingsustainabilityarchitectureMLOpsempiricalmethodsresearchagenda
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This report argues that the environmental cost of AI-enabled systems will shrink only when software engineering methods are applied across the whole system, not just to individual models. Based on a week-long workshop with 29 academics and practitioners, it identifies five focus areas as the field's open research agenda: energy assessment and standardization, evaluation and benchmarking, system architecture and lifecycle, empirical methods, and education. Three cross-cutting concerns—standardization, metric design, and holistic longitudinal thinking—recur in all five areas and tie the agenda together. A sympathetic reader takes the paper's claim to be that making progress in these five areas, guided by software engineering practice, is what would turn sustainable AI into an engineering reality.

What carries the argument

The five focus areas are the load-bearing structure of the argument: energy assessment and standardization, evaluation and benchmarking, software architecture and lifecycle, empirical methods and reproducibility, and education and awareness. Each chapter states a challenge, reviews existing work, and ends with a 'Positioning Software Engineering' subsection that proposes how SE techniques—such as a meta-methodology for choosing assessment methods, extension of benchmark suites like MLPerf with energy and carbon metrics, architectural tactics and runtime self-adaptation, standardized metric taxonomies, and embedding sustainability in required courses—would move that area forward. The three cross-cutting concerns of standardization, metrics, and holistic/longitudinal thinking act as the connective tissue that binds the five areas into a single agenda.

What would settle it

A systematic review or broad survey of the Green AI and software engineering community that produces a materially different set of top-priority challenges—for example, adding hardware-software co-design or energy-aware data management and dropping education—would undercut the claim that these five areas define the field's research agenda. A second workshop with a different participant mix that yields a different set of focus areas would show the agenda reflects the group rather than the field.

Watch

Extended reading notes

Core claim

The central claim is that environmentally sustainable AI must be engineered, not just trained: AI-enabled systems consume energy through architectural choices, lifecycle decisions, and deployment patterns that software engineering already has tools to shape. The report deliberately does not prescribe a single framework; it claims that a research agenda organized around energy assessment and standardization, sustainability-aware evaluation and benchmarking, architecture and lifecycle, empirical methods, and education captures where the field's efforts should concentrate. Within this agenda, standardization of energy metrics, well-designed metrics that respect multi-dimensional trade-offs, and holistic thinking across system boundaries and time scales are the cross-cutting themes that make the five areas one coherent program. In the paper's own words, what emerges is a conceptual research agenda, not a finished blueprint.

Load-bearing premise

The agenda's validity rests on the 29 workshop participants being representative enough of the wider research and practice community that the challenges they prioritized match the field's actual priorities; if that group had blind spots, the agenda could systematically miss important challenges.

Editorial extensions

If this is right

  • Energy and carbon metrics become first-class values in standard benchmark suites, so model comparisons routinely report environmental cost alongside accuracy and latency.
  • Architects gain concrete tactics, decision maps, and runtime self-adaptation mechanisms that let them trade off energy against other quality attributes across edge, fog, and cloud deployment.
  • A shared empirical body of knowledge, built on standardized metric taxonomies and replication, makes Green AI findings comparable and supports evidence-based regulation such as the EU AI Act.
  • Sustainability enters mandatory software engineering and AI curricula, so future practitioners treat energy awareness as a default skill rather than a niche elective.
  • Interoperable, adaptable energy assessment methodologies, documented in a public repository, let practitioners choose the right measurement approach for their system instead of relying on one rigid standard.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the standardization pillar succeeds, the accountability ambiguity the report flags in MLOps may resolve itself: with comparable metrics, regulators and customers can compare vendors directly, shifting responsibility from infrastructure providers to the whole supply chain.
  • The report's emphasis on longitudinal thinking invites a concrete test of the Jevons paradox in AI: longitudinal field studies that track whether efficiency gains in models lead to increased total deployment and energy use would extend the agenda beyond lab benchmarks.
  • The meta-methodology concept could be operationalized as a decision-support tool that recommends an energy assessment methodology from a catalogue based on system characteristics; a community-maintained repository of such methodologies, similar to design pattern catalogs, would be a direct next step the paper leaves implicit.
  • Re-running the same co-creation workshop with a deliberately more industry-heavy or hardware-vendor-diverse participant group would reveal which agenda items are robust and which depend on the original group's composition.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

0 major / 6 minor

Summary. This paper reports on the 'Greening AI with Software Engineering' workshop (CECAM-Lorentz, February 2025) and synthesizes its outcomes into a research agenda for environmentally sustainable AI from a software engineering perspective. After describing the workshop's co-creation process in a footnote, the paper organizes the agenda into five focus areas: energy assessment and standardization, evaluation and benchmarking of AI sustainability, software architecture and lifecycle, empirical methods and reproducibility, and education and awareness. Each area is presented through a set of open challenges, selected references to prior green AI/SE work, and a short 'Positioning Software Engineering' subsection that speculates how SE methods could contribute. Three cross-cutting themes are identified in a concluding synthesis: standardization, metric design, and holistic/longitudinal thinking.

Significance. If the agenda is taken up by the green AI and software engineering communities, it could serve a useful coordination function: it provides a shared vocabulary, a structured list of open problems, and concrete targets such as standardized energy benchmarks, metric taxonomies, and design-time energy prediction tools. The paper's strengths are its grounding in a described participatory process, its integration of recent empirical literature (e.g., Carbontracker, MLPerf, BLOOM footprint, green architectural tactics), and its honest use of speculative phrasing in the 'Positioning Software Engineering' subsections. The paper makes no quantitative claims and should be assessed as a vision/agenda contribution rather than an empirical study. Its main weakness is the untested generalizability of the 29-participant workshop, which the authors should acknowledge more explicitly.

minor comments (6)
  1. [Abstract and Section 1] The phrase 'identified and prioritized key challenges for the field' generalizes from a 29-participant, predominantly academic workshop to the entire field. Please replace it with 'challenges identified by the workshop participants' or 'candidate key challenges' and add a sentence in Section 1 or a new Limitations paragraph stating that the agenda reflects the participants' composition and is not claimed to be exhaustive.
  2. [Footnote 2] The statement that the coordinators 'removed underdeveloped topics' is an editorial filter that goes beyond participant consensus; this deserves a place in the main text. The authors should acknowledge that the final agenda is a consensus-then-curation outcome, and ideally describe the criteria for removing topics or list the removed topics so readers can judge the agenda's completeness.
  3. [Section 1] The paper does not document participant demographics (e.g., academic vs. industry, geographic distribution, research areas). A short paragraph on participant composition would help readers calibrate the generalizability of the agenda, especially since the paper claims 'practitioners' were included without naming specific practitioner affiliations.
  4. [Sections 2.3, 3.4, 4.4, 5.3, 6.1] The 'Positioning Software Engineering' subsections are repetitive in structure and use 'We imagine' repeatedly; a single consolidated 'Role of Software Engineering' discussion would reduce redundancy and sharpen the agenda's recommendations.
  5. [References] The reference list contains inconsistent publication status annotations (e.g., 'To appear' in [1] and [11], 'Presented at' in [3] and [41]); please standardize these entries.
  6. [Introduction] The paper lacks an explicit related-work positioning against existing Green AI roadmaps and surveys (e.g., the education roadmap in [30]); a brief paragraph in the introduction would strengthen the novelty claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the report is a self-described synthesis of a workshop, not a derivation forced by its inputs.

full rationale

This paper contains no equations, no fitted parameters, and no quantitative predictions; its central claim is that five focus areas emerged from a co-creation workshop. Footnote 2 documents the iterative elicitation process (sticky notes, rotating discussions, anchors, coordinator consolidation), and the abstract and conclusion explicitly frame the output as a 'research agenda emerging from the workshop.' The identity between the workshop discussion and the reported agenda is a matter of provenance, not circular derivation: the paper does not claim to derive the agenda from first principles or external data, and it repeatedly hedges with phrases such as 'we imagine' and 'emerging from the workshop.' Each focus area is independently grounded in external literature, including work by non-participants (e.g., MLPerf, Green AI by Schwartz et al., Luccioni et al.), so the agenda is not self-supporting by construction. The numerous self-citations by participants (e.g., Carbontracker, PePR, EcoMLS, Green Architectural Tactics) are used as concrete examples of existing tools or studies, not as the authority that compels the agenda's validity. No uniqueness theorem is imported, no parameter is fitted and renamed as a prediction, and no known result is merely renamed. The skeptic's concern about the representativeness of the 29 participants is a legitimate question about external validity and generalizability, but it is an empirical/correctness issue, not a circularity of reasoning. Accordingly, the circularity score is 0.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

The paper introduces no free parameters and no invented entities. It relies on two domain assumptions: that AI's environmental impact is a pressing problem, and that software engineering can influence it. Both are reasonable and supported by cited literature.

assumptions (2)
  • domain assumption The environmental impact of AI is large and growing, making sustainable AI a pressing problem.
    Stated in the introduction with citations to Luccioni et al. and Menzies and Johnson; it motivates the entire agenda.
  • domain assumption Software engineering practices can materially influence the environmental footprint of AI-enabled systems.
    This premise underlies the 'Positioning Software Engineering' framing throughout the paper, especially in Sections 2.3, 3.4, and 4.4.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Greening AI-enabled Systems with Software Engineering: A Research Agenda for Environmentally Sustainable AI Practices." pith.science (2026). https://pith.science/paper/G4VQAHDQ

@misc{pith2026250601774,
  author       = {Pith},
  title        = {Pith review of: Greening AI-enabled Systems with Software Engineering: A Research Agenda for Environmentally Sustainable AI Practices},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G4VQAHDQ}},
  note         = {Machine review of arXiv:2506.01774}
}
read the original abstract

The environmental impact of Artificial Intelligence (AI)-enabled systems is increasing rapidly, and software engineering plays a critical role in developing sustainable solutions. The "Greening AI with Software Engineering" CECAM-Lorentz workshop (no. 1358, 2025) funded by the Centre Europ\'een de Calcul Atomique et Mol\'eculaire and the Lorentz Center, provided an interdisciplinary forum for 29 participants, from practitioners to academics, to share knowledge, ideas, practices, and current results dedicated to advancing green software and AI research. The workshop was held February 3-7, 2025, in Lausanne, Switzerland. Through keynotes, flash talks, and collaborative discussions, participants identified and prioritized key challenges for the field. These included energy assessment and standardization, benchmarking practices, sustainability-aware architectures, runtime adaptation, empirical methodologies, and education. This report presents a research agenda emerging from the workshop, outlining open research directions and practical recommendations to guide the development of environmentally sustainable AI-enabled systems rooted in software engineering principles.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

48 extracted references · 24 canonical work pages

  1. [1]

    Negar Alizadeh, Boris Belchev, Nishant Saurabh, Patricia Kelbert, and Fernando Castor. 2025. Language Models in Software Development Tasks: An Experimental Analysis of Energy and Accuracy. (2025). arXiv:2412.00329https: //arxiv.org/abs/2412.00329To appear in the Proceed- ings of the 22nd International Conference on Mining Soft- ware Repositories, MSR 2025...

    arXiv 2025

  2. [2]

    Negar Alizadeh and Fernando Castor. 2024. Green AI: a Preliminary Empirical Study on Energy Consumption in DL Models Across Different Runtime Infrastructures. InPro- ceedings of the IEEE/ACM 3rd International Conference on AI Engineering - Software Engineering for AI(Lisbon, Por- tugal)(CAIN ’24). ACM, New York, NY, USA, 134–139. doi:10.1145/3644815.3644967

    arXiv 2024

  3. [3]

    Wolff Anthony, Benjamin Kanding, and Raghaven- dra Selvan

    Lasse F. Wolff Anthony, Benjamin Kanding, and Raghaven- dra Selvan. 2020. Carbontracker: Tracking and Predicting the Carbon Footprint of Training Deep Learning Models. (2020). arXiv:2007.03051 [cs.CY]https://arxiv.org/abs/ 2007.03051Presented at ICML Workshop on Challenges in Deploying and Monitoring Machine Learning Systems

    arXiv 2020

  4. [4]

    Timur Babakol and Yu David Liu. 2024. Tensor-Aware Energy Accounting. InProceedings of the IEEE/ACM 46th International Conference on Software Engineering(Lisbon, Portugal)(ICSE ’24). ACM, New York, NY, USA, Article 93, 12 pages. doi:10.1145/3597503.3639156

    arXiv 2024

  5. [5]

    Pedram Bakhtiarifard, Christian Igel, and Raghavendra Sel- van. 2024. EC-NAS: Energy Consumption Aware Tabular Benchmarks for Neural Architecture Search. InIEEE Inter- national Conference on Acoustics, Speech and Signal Process- ing, ICASSP 2024, Seoul, Republic of Korea, April 14-19,

    work page 2024

  6. [6]

    2012.Software Architecture in Practice(3rd ed.)

    Len Bass, Paul Clements, and Rick Kazman. 2012.Software Architecture in Practice(3rd ed.). Addison-Wesley Profes- sional, Westford, MA, USA. Publication Title: The SEI Series in Software Engineering

    work page 2012

  7. [8]

    Luc ´ ıa Bouza, Aur´ elie Bugeau, and Lo¨ıc Lannelongue. 2023. How to estimate carbon footprint when training deep learn- ing models? A guide and review.Environmental Research Communications5, 11 (nov 2023), 115014. doi:10.1088/ 2515-7620/acf81b

    work page 2023

  8. [9]

    Han Cai, Chuang Gan, Tianzhe Wang, Zhekai Zhang, and Song Han. 2020. Once-for-All: Train One Net- work and Specialize it for Efficient Deployment. (2020). arXiv:1908.09791 [cs.LG]https://arxiv.org/abs/1908. 09791

    arXiv 2020

Show all 48 references
  1. [10]

    Qingqing Cao, Yash Kumar Lal, Harsh Trivedi, Aruna Bala- subramanian, and Niranjan Balasubramanian. 2021. IrEne: Interpretable Energy Prediction for Transformers. (2021). arXiv:2106.01199 [cs.CL]https://arxiv.org/abs/2106. 01199

    2021 arXiv

  2. [11]

    Iffat Fatima, Patricia Lago, Vasilios Andrikopoulos, and Bram van der Waaij. 2025. Using Sustainability Im- pact Scores for Software Architecture Evaluation. (2025). arXiv:2501.17004 [cs.SE]https://arxiv.org/abs/2501. 17004To appear in the Proceedings of the 22nd Interna- tion...

    2025 arXiv

  3. [12]

    Stefano Forti, Uwe Breitenb ¨ucher, and Jacopo Soldani. 2022. Trending Topics in Software Engineering.SIGSOFT Softw. Eng. Notes47, 3 (July 2022), 20–21. doi:10.1145/3539814. 3539820

    2022 doi

  4. [13]

    Ay- ala, and Cristina G´ omez

    Xavier Franch, Silverio Mart ´ ınez-Fern´ andez, Claudia P. Ay- ala, and Cristina G´ omez. 2022. Architectural Decisions in AI-Based Systems: An Ontological View. InQuality of Information and Communications Technology, Antonio Vallecillo, Joost Visser, and Ricardo P´ erez-Cast...

    2022 doi

  5. [14]

    1995.Design patterns: elements of reusable object- oriented software

    Erich Gamma, Richard Helm, Ralph Johnson, and John Vlis- sides. 1995.Design patterns: elements of reusable object- oriented software. Addison-Wesley Longman Publishing Co., Inc., USA

    1995

  6. [15]

    Stefanos Georgiou, Maria Kechagia, Tushar Sharma, Feder- ica Sarro, and Ying Zou. 2022. Green AI: do deep learning frameworks have different costs?. InProceedings of the 44th International Conference on Software Engineering(Pitts- burgh, Pennsylvania)(ICSE ’22). ACM, New York,...

    2022

  7. [16]

    Langdon, and Federica Sarro

    Jingzhi Gong, Sisi Li, Giordano d’Aloisio, Zishuo Ding, Yu- long Ye, William B. Langdon, and Federica Sarro. 2024. GreenStableYolo: Optimizing Inference Time and Image Quality of Text-to-Image Generation. InSearch-Based Soft- ware Engineering, Gunel Jahangirova and Foutse Khom...

    2024

  8. [17]

    Shuai Hao, Ding Li, William G. J. Halfond, and Ramesh Govindan. 2013. Estimating mobile application energy con- sumption using program analysis . In2013 35th Interna- tional Conference on Software Engineering (ICSE). IEEE, Los Alamitos, CA, USA, 92–101. doi:10.1109/ICSE.2013. 6606555

    2013 doi

  9. [18]

    Hillary Hart and Craig Baehr. 2013. Sustainable Practices for Developing a Body of Knowledge.Technical Communication 60, 4 (November 2013), 259–266.https://www.jstor.org/ stable/26464355

    2013

  10. [19]

    Peter Henderson, Jieru Hu, Joshua Romoff, Emma Brun- skill, Dan Jurafsky, and Joelle Pineau. 2020. Towards the systematic reporting of the energy and carbon footprints of machine learning.Journal of Machine Learning Research21, 248 (2020), 1–43

    2020

  11. [20]

    Vitor Maciel Fontes Jacques, Negar Alizadeh, and Fer- nando Castor. 2024. A Study on the Battery Usage of Deep Learning Frameworks on iOS Devices. InProceed- ings of the IEEE/ACM 11th International Conference on Mobile Software Engineering and Systems(Lisbon, Portu- gal)(MOBIL...

    2024

  12. [21]

    Heli J ¨arvenp¨a¨a, Patricia Lago, Justus Bogner, Grace Lewis, Henry Muccini, and Ipek Ozkaya. 2024. A Synthesis of Green Architectural Tactics for ML-Enabled Systems. InProceed- ings of the 46th International Conference on Software Engi- neering: Software Engineering in Socie...

    2024

  13. [22]

    Patricia Lago. 2019. Architecture Design Decision Maps for Software Sustainability . In2019 IEEE/ACM 41st Interna- tional Conference on Software Engineering: Software Engi- neering in Society (ICSE-SEIS). IEEE, Los Alamitos, CA, USA, 61–64. doi:10.1109/ICSE-SEIS.2019.00015

    2019

  14. [23]

    Lewis, Ipek Ozkaya, and Xiwei Xu

    Grace A. Lewis, Ipek Ozkaya, and Xiwei Xu. 2021. Software Architecture Challenges for ML Systems . In2021 IEEE In- ternational Conference on Software Maintenance and Evo- lution (ICSME). IEEE, Los Alamitos, CA, USA, 634–638. doi:10.1109/ICSME52107.2021.00071

    2021

  15. [24]

    Francisco L´ opez, Lars Karlsson, and Paolo Bientinesi. 2023. FLOPs as a Discriminant for Dense Linear Algebra Al- gorithms. InProceedings of the 51st International Confer- ence on Parallel Processing(Bordeaux, France)(ICPP ’22). ACM, New York, NY, USA, Article 11, 10 pages. d...

    2023

  16. [25]

    Alexandra Sasha Luccioni, Sylvain Viguier, and Anne-Laure Ligozat. 2023. Estimating the Carbon Footprint of BLOOM, a 176B Parameter Language Model.Journal of Machine Learning Research24, 253 (2023), 1–15.http://jmlr.org/ papers/v24/23-0069.html

    2023

  17. [26]

    Sasha Luccioni, Boris Gamazaychikov, Emma Strubell, Sara Hooker, Yacine Jernite, Carole-Jean Wu, and Margaret Mitchell. 2025. AI Energy Score Leaderboard - February 2025.https://huggingface.co/spaces/AIEnergyScore/ Leaderboard

    2025

  18. [27]

    Sasha Luccioni, Yacine Jernite, and Emma Strubell. 2024. Power Hungry Processing: Watts Driving the Cost of AI De- ployment?. InProceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency(Rio de Janeiro, Brazil)(F AccT ’24). ACM, New York, NY, USA, 85–...

    2024

  19. [28]

    Diamos, Cody Coleman, Paulius Micikevicius, David A

    Peter Mattson, Christine Cheng, Gregory F. Diamos, Cody Coleman, Paulius Micikevicius, David A. Patter- son, Hanlin Tang, Gu-Yeon Wei, Peter Bailis, Victor Bittorf, David Brooks, Dehao Chen, Debo Dutta, Udit Gupta, Kim M. Hazelwood, Andy Hock, Xinyuan Huang, Daniel Kang, David...

    2020

  20. [29]

    Tim Menzies and Brittany Johnson. 2024. Powering Down: An Interview With Federica Sarro on Tackling Energy Con- sumption in AI-Powered Software Systems .IEEE Software 41, 05 (Sept. 2024), 89–92. doi:10.1109/MS.2024.3410011

    2024

  21. [30]

    Ana Moreira, Patricia Lago, Rogardt Heldal, Stefanie Betz, Ian Brooks, Rafael Capilla, Vlad Constantin Coroam˘ a, Leti- cia Duboc, Jo˜ ao Paulo Fernandes, Ola Leifler, Ngoc-Thanh Nguyen, Shola Oyedeji, Birgit Penzenstadler, Anne-Kathrin Peters, Jari Porras, and Colin C. Venter...

    2025 doi

  22. [31]

    Nienke Nijkamp, June Sallou, Niels Van Der Heijden, and Lu ´ ıs Cruz. 2024. Green AI in Action: Strategic Model Selection for Ensembles in Production. InProceedings of the 1st ACM International Conference on AI-Powered Soft- ware. ACM, Porto de Galinhas Brazil, 50–58. doi:10.1...

    2024

  23. [32]

    Wellington Oliveira, Renato Oliveira, Fernando Castor, Gus- tavo Pinto, and Jo˜ ao Paulo Fernandes. 2021. Improving energy-efficiency by recommending Java collections.Empir- ical Softw. Engg.26, 3 (May 2021), 45 pages. doi:10.1007/ s10664-021-09950-y

    2021

  24. [33]

    Anne-Kathrin Peters, Rafael Capilla, Vlad Constantin Coroam˘ a, Rogardt Heldal, Patricia Lago, Ola Leifler, Ana Moreira, Jo˜ ao Paulo Fernandes, Birgit Penzenstadler, Jari Porras, and Colin C Venters. 2024. Sustainability in Com- puting Education: A Systematic Literature Revie...

    2024 doi

  25. [34]

    Kalbarczyk, Tamer Ba¸ sar, and Ravishankar K

    Haoran Qiu, Weichao Mao, Archit Patke, Shengkun Cui, Saurabh Jha, Chen Wang, Hubertus Franke, Zbigniew T. Kalbarczyk, Tamer Ba¸ sar, and Ravishankar K. Iyer. 2024. Power-aware deep learning model serving withµ-serve. In Proceedings of the 2024 USENIX Conference on Usenix Annua...

    2024

  26. [35]

    Saurabhsingh Rajput, Maria Kechagia, Federica Sarro, and Tushar Sharma. 2024. Greenlight: Highlighting TensorFlow APIs Energy Footprint. InProceedings of the 21st Interna- tional Conference on Mining Software Repositories(Lisbon, Portugal)(MSR ’24). ACM, New York, NY, USA, 304...

    2024

  27. [37]

    Saurabhsingh Rajput, Tim Widmayer, Ziyuan Shang, Maria Kechagia, Federica Sarro, and Tushar Sharma. 2024. En- hancing Energy-Awareness in Deep Learning through Fine- Grained Energy Measurement.ACM Trans. Softw. Eng. Methodol.33, 8, Article 211 (Dec. 2024), 34 pages. doi:10. 11...

    2024

  28. [38]

    Scott Gardner, Itay Hubara, Sachin Idgunji, Thomas B

    Vijay Janapa Reddi, Christine Cheng, David Kanter, Pe- ter Mattson, Guenther Schmuelling, Carole-Jean Wu, Brian Anderson, Maximilien Breughe, Mark Charlebois, William Chou, Ramesh Chukka, Cody Coleman, Sam Davis, Pan Deng, Greg Diamos, Jared Duke, Dave Fick, J. Scott Gardner, ...

  29. [39]

    Federica Sarro. 2023. Search-Based Software Engineering in the Era of Modern Software Systems . In2023 IEEE 31st International Requirements Engineering Conference (RE). IEEE, Los Alamitos, CA, USA, 3–5. doi:10.1109/RE57278. 2023.00010

    2023

  30. [40]

    Smith, and Oren Et- zioni

    Roy Schwartz, Jesse Dodge, Noah A. Smith, and Oren Et- zioni. 2020. Green AI.Commun. ACM63, 12 (Nov. 2020), 54–63. doi:10.1145/3381831

    2020 doi

  31. [41]

    Raghavendra Selvan, Bob Pepin, Christian Igel, Gabrielle Samuel, and Erik B Dam. 2024. PePR: Performance Per Re- source Unit as a Metric to Promote Small-Scale Deep Learn- ing in Medical Image Analysis. (2024). doi:10.48550/arXiv. 2403.12562arXiv:2403.12562 Presented at Northe...

    2024 doi arXiv

  32. [42]

    Meghana Tedla, Shubham Kulkarni, and Karthik Vaid- hyanathan. 2024. EcoMLS: A Self-Adaptation Approach for Architecting Green ML-Enabled Systems. In2024 IEEE 21st International Conference on Software Architecture Compan- ion (ICSA-C). IEEE, Hyderabad, India, 230–237. doi:10. 1...

    2024

  33. [43]

    Roberto Verdecchia, Luis Cruz, June Sallou, Michelle Lin, James Wickenden, and Estelle Hotellier. 2022. Data-Centric Green AI An Exploratory Empirical Study. In2022 Interna- tional Conference on ICT for Sustainability (ICT4S). IEEE, Plovdiv, Bulgaria, 35–45. doi:10.1109/ICT4S5...

    2022

  34. [44]

    Ohls- son, Bj ¨orn Regnell, and Anders Wessl´ en

    Claes Wohlin, Per Runeson, Martin H ¨ost, Magnus C. Ohls- son, Bj ¨orn Regnell, and Anders Wessl´ en. 2024.Experimen- tation in Software Engineering. Springer Berlin Heidelberg, Berlin. doi:10.1007/978-3-662-69306-3

    2024 doi

  35. [45]

    Dustin Wright, Christian Igel, Gabrielle Samuel, and Raghavendra Selvan. 2024. Efficiency is Not Enough: A Crit- ical Perspective of Environmentally Sustainable AI. (2024). doi:10.1145/3724500arXiv:2403.12562 [cs.LG]

    2024 arXiv

  36. [46]

    Daliang Xu, Wangsong Yin, Hao Zhang, Xin Jin, Ying Zhang, Shiyun Wei, Mengwei Xu, and Xuanzhe Liu. 2025. EdgeLLM: Fast On-Device LLM Inference With Speculative Decoding.IEEE Transactions on Mobile Computing24, 4 (April 2025), 3256–3273. doi:10.1109/TMC.2024.3513457

    2025

  37. [47]

    Tim Yarally, Luis Cruz, Daniel Feitosa, June Sallou, and Arie van Deursen. 2023. Batching for Green AI - An Exploratory Study on Inference . In2023 49th Euromicro Conference on Software Engineering and Advanced Appli- cations (SEAA). IEEE, Los Alamitos, CA, USA, 112–119. doi:1...

    2023

  38. [48]

    Yiyang Zhao, Yunzhuo Liu, Bo Jiang, and Tian Guo. 2024. CE-NAS: An End-to-End Carbon-Efficient Neural Architec- ture Search Framework. (2024). arXiv:2406.01414 [cs.LG] https://arxiv.org/abs/2406.01414

    2024 arXiv

  39. [2020]

    In2020 ACM/IEEE 47th Annual International Symposium on Computer Archi- tecture (ISCA)

    MLPerf Inference Benchmark . In2020 ACM/IEEE 47th Annual International Symposium on Computer Archi- tecture (ISCA). IEEE, Los Alamitos, CA, USA, 446–459. doi:10.1109/ISCA45697.2020.00045

    2020

  40. [2024]

    doi:10.1109/ICASSP48485.2024

    IEEE, 5660–5664. doi:10.1109/ICASSP48485.2024. 10448303

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.