Pith. sign in

REVIEW 5 major objections 5 minor 31 references

Federated learning's carbon footprint can only be compared across studies if researchers adopt a standardized, phase-aware accounting method that includes both compute and communication emissions.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

A phase-aware carbon-accounting method for federated learning shows that client efficiency tiers and coordination idle time can inflate total CO2e by 8-22x, while GPU choice changes runtime more than energy.

T0 review reviewed 2026-08-03 challenge →

load-bearing objection A practical NVFlare+CodeCarbon integration for standardized FL carbon accounting, but the empirical validation is under-specified and the measurement accuracy is uncalibrated. the 5 major comments →

arxiv 2602.00343 v2 pith:PNY3MDYZ submitted 2026-01-30 cs.DC cs.AIcs.PF

Standardized Methods and Recommendations for Green Federated Learning

Austin Tapp , Holger R. Roth , Ziyue Xu , Abhijeet Parida , Hareem Nisar , Marius George Linguraru This is my paper
classification cs.DC cs.AIcs.PF
keywords green federated learningcarbon accountingCodeCarbonNVFlareenergy measurementcommunication emissionscarbon intensityreproducibility
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 paper argues that federated learning's environmental cost cannot be meaningfully compared across studies until researchers agree on how to measure it. It offers a concrete protocol: attribute emissions to initialization, per-round training, evaluation, and idle/coordination phases at each client, and add an estimate for communication based on model-update size. Using this protocol on CIFAR-10, it finds that slower client hardware and coordination idle increase total CO2e by 8.34x and 21.73x over a high-efficiency baseline, costs that a fixed FL protocol hides. On retinal segmentation, swapping H100 for V100 GPUs changes runtime by roughly 1.7x but shifts energy and emissions unevenly across sites, arguing for per-site and per-round reporting. The paper's claim is that reproducible 'green' FL evaluation requires a standardized measurement boundary.

Core claim

The discovery is that a standardized, phase-aware carbon accounting layer for federated learning is feasible and necessary. By wrapping each NVFlare client with CodeCarbon's process-level tracker and logging explicit tasks for initialization, idle time, per-round training, and evaluation, while additionally estimating communication emissions from transmitted update bytes, the authors obtain a comparable carbon footprint for FL workloads. Their measurements show that system-level inefficiency and coordination idle can dominate the carbon cost of an otherwise identical FL protocol, and that heterogeneous client hardware changes both runtime and the distribution of emissions across sites. The a

What carries the argument

The central mechanism is the phase-aware task boundary: CodeCarbon's EmissionsTracker with tracking_mode='process' and measure_power_secs=1, wrapped around explicit NVFlare client tasks (init, idle_time, round_k, evaluate), plus a communication-emissions estimator E_comm = 2 * D_GB * I_net, scaled by grid carbon intensity F_grid. This instrumentation converts an opaque FL run into a stream of per-round, per-site energy and CO2e numbers that can be aggregated and compared. The measurement boundary—including idle and one-time initialization—is what makes the accounting comparable across deployments.

Load-bearing premise

The entire empirical payload assumes CodeCarbon's process-level energy estimator faithfully captures each client's real CPU, GPU, and RAM draw without calibration against a physical meter.

What would settle it

Run the same CIFAR-10 FL experiment while logging each client's wall power with a calibrated meter; if CodeCarbon's process-level total energy differs systematically from the meter, or if the meter shows idle-phase energy the tracker misses, then the reported CO2e totals and the 8.34x/21.73x ratios will not survive.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Green-FL studies that report only compute energy will miss coordination idle and communication, which for slow clients can dominate the total.
  • A fixed FL protocol on less efficient client machines can produce more than 20 times the CO2e of the same protocol on efficient machines, so hardware tier must be reported as a covariate.
  • GPU-tier choices such as H100 vs V100 change both runtime and the spatial distribution of emissions across sites, so per-site and per-round reporting is necessary for fair comparisons.
  • A lightweight instrumentation layer around existing FL orchestration can yield standardized carbon numbers without requiring new hardware.
  • Because grid carbon intensity varies by location, the same energy use can correspond to very different CO2e; standardized reporting must include location or grid-intensity metadata.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Left implicit in the paper is that the protocol can be extended to server-side aggregation and orchestration energy; the paper measures client-side and communication only, and adding server energy would likely raise the coordination share further.
  • The efficiency-tier ratios (8.34x, 21.73x) are tied to the specific injected slowdown schedules; a useful next test is to calibrate the protocol against physical power meters and to define a standard slowdown taxonomy so ratios transfer across studies.
  • The communication-emissions model uses a constant network energy intensity; plugging in time-varying or per-link intensity would let the same framework produce location- and time-aware footprints, which the paper identifies as a driver but does not implement.
  • Because the tracker's absolute values are estimates, the strongest use of this method is relative comparison across FL configurations rather than carbon certification; future work could add uncertainty bounds to each reported number.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes a standardized carbon-accounting methodology for federated learning, integrating NVIDIA NVFlare with CodeCarbon process-level energy tracking to attribute CO2e to explicit phases (initialization, per-round training, idle/coordination, evaluation) and supplementing measured compute energy with an estimated communication-energy term. The authors report experiments on CIFAR-10 classification and retinal optic disk segmentation, including controlled client-efficiency tiers (high/medium/low) and an H100 versus V100 GPU comparison, and claim that their method supports reproducible green-FL evaluation. The main empirical findings are that medium- and low-efficiency tiers increase total CO2e by 8.34x and 21.73x relative to the high-efficiency baseline, and that H100/V100 runs show a ~1.7x runtime gap with non-uniform energy changes across sites.

Significance. If the method were soundly validated, it would provide a valuable measurement boundary and logging schema for the green-FL community, addressing a real need for comparable carbon accounting across FL studies. The paper also ships code and uses a configuration-driven workflow, which are strengths for reproducibility. However, the current manuscript does not substantiate the validation: the experimental sections are empty, and the measurement assumptions are not calibrated or uncertainty-quantified. The significance is therefore potential rather than established.

major comments (5)
  1. [§3.1–§3.2, §4, §5] The core validation sections are missing from the submitted text. Headings '3.1 CIFAR-10 with Efficiency Testing', '3.2 Retinal Optic Disk Segmentation', '4 Discussion', and '5 Conclusion' appear with no body text. Tables 1–2 and Figures 2–5 are referenced but never described or interpreted. The central claim that the methodology is 'validated' on two workloads cannot be assessed. This is a load-bearing omission that must be repaired before any decision.
  2. [§2.2, 'Client-side emissions tracking'] The entire empirical basis rests on CodeCarbon's process-level EmissionsTracker (tracking_mode='process', measure_power_secs=1) for per-client energy attribution. The manuscript reports no calibration against physical power meters, no repeated runs, and no uncertainty analysis. Since six clients run as separate processes on the same host, per-process GPU power attribution can be materially biased when processes contend for the same GPU. All headline CO2e ratios (8.34x, 21.73x) and the per-site comparisons derive from these per-client estimates; without a validation of the measurement, the empirical conclusions are unsupported.
  3. [§2.4, Eq. (1)] The communication emission model is specified as E_comm = 2 * D_GB * I_net and C_comm = E_comm * F_grid, but neither I_net (network energy intensity) nor F_grid (grid carbon factor) is given a numerical value or a source. Absolute emission totals are therefore not reproducible, undermining the paper's stated goal of a standardized, reproducible method. Additionally, the linear relationship between transmitted bytes and energy is an assumption with no cited basis or experimental justification.
  4. [§2.5 and Table 1] The efficiency-tier experiments are produced by 'injecting slowdowns' into the client training loop. The resulting ratios (8.34x and 21.73x) are arithmetic consequences of the chosen slowdown factors, not independent empirical discoveries about naturally occurring FL inefficiencies. This is acceptable as a demonstration of the measurement method, but the paper should explicitly frame it as such and report the actual slowdown-injection parameters. Without that, the claims risk being circular and are not reproducible.
  5. [Table 2, Site 1 row] Site 1 reports identical energy (0.32 kWh) and identical CO2e (0.13 kg) for H100 and V100 runs, despite a 503-minute versus 290-minute runtime gap. This implies very different average power draws (≈38 W for V100 vs. ≈66 W for H100), which is surprising and may indicate a measurement artifact or a reporting error. The row undermines confidence in the H100/V100 comparison and needs a specific explanation or verification.
minor comments (5)
  1. [Table 1] The columns 'Mean energy (kWh/round)' and 'Mean CO2e (kg/round)' appear to be per-client-per-round averages, but this is not stated. The relationship between these values and 'Total CO2e' is not evident without assuming six clients and ten round. Clarify the aggregation.
  2. [References] Several references have incorrect or inconsistent years (e.g., McMahan et al. is listed as 2023 despite the arXiv 1602.05629 reference; ACM copyright notice says 2018). Also, reference [15] and [20] appear to be the same work cited twice with different formatting. These need cleanup.
  3. [Figures] Figures 2–5 are referenced but not described in the text, and their captions are minimal. If the validation sections are restored, each figure needs an explicit discussion.
  4. [§2.1] The measurement boundary is described as 'comparable across FL deployments,' but the list mixes measured and estimated components without a clear statement of which quantities enter the reported totals. A formula with explicit terms would help.
  5. [General] The paper contains several typographical and formatting issues, including inconsistent use of 'CO2e' vs. 'CO 2e', a footnote number without a corresponding footnote, and a dangling URL in Figure 5's caption. These should be corrected.

Circularity Check

0 steps flagged

No significant circularity: the paper is an empirical measurement-and-recommendation study with no fitted parameters, no self-citation chain bearing the central argument, and no equation that reduces to its own inputs.

full rationale

The paper's central contribution is a measurement methodology (phase-aware CodeCarbon tracking inside NVFlare) plus observed CO2e figures from two workloads. The 8.34x and 21.73x ratios are measured outcomes of deliberately injected slowdowns, not predictions derived from a fitted model; the direction of causation is physical (more runtime and idle time consume more energy), and the values are not used to reverse-engineer any parameter. The communication-emissions equation E_comm = 2*D_GB*I_net is an estimate with unspecified I_net and F_grid, but it is not fitted to the reported totals and does not constitute a circular reduction. Citations to NVFlare and CodeCarbon are tool/framework references, not self-citations used to justify the paper's empirical claims. Concerns about CodeCarbon's per-process energy attribution being uncalibrated are external-validity or correctness risks, which the instructions explicitly separate from circularity. No equation is shown to equal another by construction, and no fitted input is renamed as a prediction. Therefore the derivation chain is self-contained in the relevant sense and no significant circularity is present.

Axiom & Free-Parameter Ledger

2 free parameters · 5 axioms · 0 invented entities

The paper rests on standard software components (CodeCarbon, NVFlare) and a linear communication-energy model; no new physical entities. Primary uncharged inputs are the network energy intensity I_net and slowdown injection parameters, neither of which are given numerically. Several domain assumptions about measurement fidelity are untested.

free parameters (2)
  • I_net (network energy intensity) = not specified
    In E_comm = 2 * D_GB * I_net, the energy per GB to transmit model updates is a configurable value with no default or source given; communication totals in Table 1 depend on it.
  • Slowdown injection factors for medium/low efficiency tiers = not specified
    The headline 8.34x and 21.73x CO2e increases are relative ratios produced by deliberately injected client-loop delays; the injection magnitude/duration is not reported, so these are hand-chosen experimental conditions rather than independently grounded measurements.
axioms (5)
  • domain assumption CodeCarbon process-level tracking accurately measures CPU/GPU/RAM energy consumption
    Central measurements rely on CodeCarbon EmissionsTracker with tracking_mode='process'; no calibration against hardware power meters is reported (Section 2.2).
  • domain assumption Grid carbon intensity for country_iso_code represents actual emissions
    CO2e conversion uses CodeCarbon's grid intensity (default USA); no per-site/time-of-day verification is shown (Section 2.2, Figure 4).
  • ad hoc to paper Communication energy is linear in transmitted bytes (E = 2 * D * I_net)
    The communication model is stated in Section 2.4 without empirical basis or reference; the factor 2 (uplink+downlink) and constant I_net are adopted for this paper.
  • domain assumption Simulated NVFlare execution with six clients and fixed rounds is representative of real FL deployments
    Results come from a local simulation, not distributed devices; extrapolation to at-scale FL is assumed (Section 2.2).
  • domain assumption The measurement boundary (init, idle, round, evaluate) captures all relevant FL emissions
    The boundary definition in Section 2.1 is a design choice; server-side compute, networking hardware, and device embodied carbon are excluded without discussion.

reviewed 2026-08-03 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Standardized Methods and Recommendations for Green Federated Learning." pith.science (2026). https://pith.science/paper/PNY3MDYZ

@misc{pith2026260200343,
  author       = {Pith},
  title        = {Pith review of: Standardized Methods and Recommendations for Green Federated Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PNY3MDYZ}},
  note         = {Machine review of arXiv:2602.00343}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Federated learning (FL) enables collaborative model training over privacy-sensitive, distributed data, but its environmental impact is difficult to compare across studies due to inconsistent measurement boundaries and heterogeneous reporting. We present a practical carbon-accounting methodology for FL CO2e tracking using NVIDIA NVFlare and CodeCarbon for explicit, phase-aware tasks (initialization, per-round training, evaluation, and idle/coordination). To capture non-compute effects, we additionally estimate communication emissions from transmitted model-update sizes under a network-configurable energy model. We validate the proposed approach on two representative workloads: CIFAR-10 image classification and retinal optic disk segmentation. In CIFAR-10, controlled client-efficiency scenarios show that system-level slowdowns and coordination effects can contribute meaningfully to carbon footprint under an otherwise fixed FL protocol, increasing total CO2e by 8.34x (medium) and 21.73x (low) relative to the high-efficiency baseline. In retinal segmentation, swapping GPU tiers (H100 vs.\ V100) yields a consistent 1.7x runtime gap (290 vs. 503 minutes) while producing non-uniform changes in total energy and CO2e across sites, underscoring the need for per-site and per-round reporting. Overall, our results support a standardized carbon accounting method that acts as a prerequisite for reproducible 'green' FL evaluation. Our code is available at https://github.com/Pediatric-Accelerated-Intelligence-Lab/carbon_footprint.

Figures

Figures reproduced from arXiv: 2602.00343 by Abhijeet Parida, Austin Tapp, Hareem Nisar, Holger R. Roth, Marius George Linguraru, Ziyue Xu.

Figure 1
Figure 1. Figure 1: Green Federated Learning simulated through NVFlare uses a central server to coordinate training and aggregate [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Mean emissions over time by client-performance [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 5
Figure 5. Figure 5: Global variation in grid carbon intensity, motivating [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 4
Figure 4. Figure 4: Energy vs. emissions under alternative grid carbon [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

31 extracted references · 2 canonical work pages

  1. [1]

    Abdullatif Albaseer, Abegaz Mohammed Seid, Mohamed Abdallah, Ala Al-Fuqaha, and Aiman Erbad. 2024. Novel Approach for Curbing Unfair Energy Consumption and Biased Model in Federated Edge Learning.IEEE Transactions on Green Communications and Networking8, 2 (June 2024), 865–877. doi:10.1109/TGCN. 2024.3350735 1https://app.electricitymaps.com/map Standardiz...

    arXiv 2024

  2. [2]

    Daniel Richards Arputharaj, Charlotte Rodriguez, Angelo Rodio, and Giovanni Neglia. 2025. Green Federated Learning via Carbon-Aware Client and Time Slot Scheduling. arXiv:2509.08980 [cs.LG] https://arxiv.org/abs/2509.08980

    Pith/arXiv arXiv 2025

  3. [3]

    Luca Barbieri, Stefano Savazzi, Sanaz Kianoush, Monica Nicoli, and Luigi Serio

  4. [4]

    Beutel, Taner Topal, Akhil Mathur, Xinchi Qiu, Javier Fernandez- Marques, Yan Gao, Lorenzo Sani, Kwing Hei Li, Titouan Parcollet, Pedro Porto Buarque de Gusmão, and Nicholas D

    Daniel J. Beutel, Taner Topal, Akhil Mathur, Xinchi Qiu, Javier Fernandez- Marques, Yan Gao, Lorenzo Sani, Kwing Hei Li, Titouan Parcollet, Pedro Porto Buarque de Gusmão, and Nicholas D. Lane. 2022. Flower: A Friendly Federated Learning Research Framework. arXiv:2007.14390 [cs.LG] https: //arxiv.org/abs/2007.14390

    Pith/arXiv arXiv 2022

  5. [5]

    Verónica Bolón-Canedo, Laura Morán-Fernández, Brais Cancela, and Amparo Alonso-Betanzos. 2024. A review of green artificial intelligence: Towards a more sustainable future.Neurocomputing599 (2024), 128096. doi:10.1016/j.neucom. 2024.128096

    arXiv 2024

  6. [6]

    Gökcan Cantali, Gürkan Gür, and Burkhard Stiller. 2026. FedSynthesis: A Flower- based Framework for Carbon-Reduced Federated Learning. InProceedings of the Figure 5: Global variation in grid carbon intensity, motivating location-aware and time-aware carbon accounting. Source: Electricity Maps live map. 1 18th IEEE/ACM International Conference on Utility a...

    arXiv 2026

  7. [7]

    Chao Feng, Alberto Huertas Celdrán, Xi Cheng, Gérôme Bovet, and Burkhard Stiller. 2026. GreenDFL: A framework for assessing the sustainability of Decen- tralized Federated Learning systems.Information and Software Technology190 (2026), 107937. doi:10.1016/j.infsof.2025.107937

    arXiv 2026

  8. [8]

    Oscar Fontenla-Romero, Blanca Guijarro-Berdiñas, Elena Hernández-Pereira, and Blanca Pérez-Sánchez. 2024. An Effective and Efficient Green Federated Learning Method for One-Layer Neural Networks. InSAC’24: Proceedings of the 39th ACM/SIGAPP Symposium on Applied Computing. ACM, New York, NY, USA, 1050–1052. doi:10.1145/3605098.3636144

    arXiv 2024

  9. [9]

    Yuan-Lin Hsu, Chia-Feng Liu, Hung-Yu Wei, and Mehdi Bennis. 2022. Optimized Data Sampling and Energy Consumption in IIoT: A Federated Learning Approach. IEEE Transactions on Communications70, 12 (Dec. 2022), 7915–7931. doi:10.1109/ TCOMM.2022.3216353

    arXiv 2022

  10. [10]

    Yan Hu, Haibao Huang, and Nuo Yu. 2022. Resource Optimization and Device Scheduling for Flexible Federated Edge Learning with Tradeoff Between Energy Consumption and Model Performance.Mobile Networks & Applications27 (2022), 2118–2137. doi:10.1007/s11036-022-02009-2

    work page doi:10.1007/s11036-022-02009-2 2022

  11. [11]

    Minsu Kim, Walid Saad, Mohammad Mozaffari, and Mérouane Debbah

  12. [12]

    Primatar Kuswiradyo, Binayak Kar, and Shan-Hsiang Shen. 2024. Optimizing the Energy Consumption in Three -Tier Cloud–Edge–Fog Federated Systems with Omnidirectional Offloading.Computer Networks250, Article 110578 (2024), 1 pages. doi:10.1016/j.comnet.2024.110578

    arXiv 2024

  13. [13]

    Peichun Li, Xumin Huang, Miao Pan, and Rong Yu. 2021. FedGreen: Federated Learning with Fine-Grained Gradient Compression for Green Mobile Edge Com- puting. In2021 IEEE Global Communications Conference (GLOBECOM). IEEE, Madrid, Spain, 1–6. doi:10.1109/GLOBECOM46510.2021.9685582

    arXiv 2021

  14. [14]

    Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Agüera y Arcas

    H. Brendan McMahan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Agüera y Arcas. 2023. Communication-Efficient Learning of Deep Net- works from Decentralized Data. arXiv:1602.05629 [cs.LG] https://arxiv.org/abs/ 1602.05629

    Pith/arXiv arXiv 2023

  15. [15]

    Talha Mehboob, Noman Bashir, Jesus Omana Iglesias, Michael Zink, and David Irwin. 2023. EcoLearn: Optimizing the Carbon Footprint of Federated Learning. arXiv:2310.17972 [cs.LG] https://arxiv.org/abs/2310.17972 arXiv preprint, version updated in 2025

    Pith/arXiv arXiv 2023

  16. [16]

    mlco2. 2025. CodeCarbon: Track Emissions from Compute and Recommend Ways to Reduce Their Impact on the Environment. Software repository. https: //github.com/mlco2/codecarbon Accessed 2025

    2025

  17. [17]

    NVIDIA. 2025. NVFlare: Federated Learning from Simulation to Real World. Software framework and documentation. https://github.com/NVIDIA/NVFlare Accessed 2025

    2025

  18. [18]

    Shamim Hossain

    Yuanhang Qi and M. Shamim Hossain. 2024. Harnessing Federated Generative Learning for Green and Sustainable Internet of Things.Journal of Network and Computer Applications222, Article 103812 (2024), 1 pages. doi:10.1016/j.jnca.2023. 103812 ACM e-Energy ’26, 2026, Banff, Canada Tapp et al

    work page doi:10.1016/j.jnca.2023 2024

  19. [19]

    Xinchi Qiu, Titouan Parcollet, Javier Fernandez-Marques, Pedro P. B. Gusmao, Yan Gao, Daniel J. Beutel, Taner Topal, Akhil Mathur, and Nicholas D. Lane. 2023. A first look into the carbon footprint of federated learning.J. Mach. Learn. Res. 24, 1, Article 129 (Jan. 2023), 23 pages

    2023

  20. [20]

    Gusmao, Yan Gao, Daniel J

    Xinchi Qiu, Titouan Parcollet, Javier Fernandez-Marques, Pedro P.B. Gusmao, Yan Gao, Daniel J. Beutel, Taner Topal, Akhil Mathur, and Nicholas D. Lane. 2023. A First Look into the Carbon Footprint of Federated Learning.Journal of Machine Learning Research24, 129 (2023), 1–23. http://jmlr.org/papers/v24/21-0445.html

    2023

  21. [21]

    Al -Moliki, Khaled A

    Adeb Salh, Razali Ngah, Lukman Audah, Kwang Soon Kim, Qazwan Abdullah, Yahya M. Al -Moliki, Khaled A. Aljaloud, and Md. Hairul Nizam Talib. 2023. Energy-Efficient Federated Learning with Resource Allocation for Green IoT Edge Intelligence in B5G.IEEE Access11 (2023), 16353–16367. doi:10.1109/ ACCESS.2023.3244099

    arXiv 2023

  22. [22]

    Raffaele Savazzi, Leo Loven, and Allan B. Poulsen. 2022. Energy and Carbon Footprint Analysis of Distributed and Federated Learning: An Overview and Experimental Evaluation. arXiv:2209.00050 [cs.LG] https://arxiv.org/abs/2209. 00050

    Pith/arXiv arXiv 2022

  23. [23]

    Dipanwita Thakur, Antonella Guzzo, Giancarlo Fortino, and Francesco Piccialli

  24. [24]

    Jiali Wang, Yijie Mao, Ting Wang, and Yuanming Shi. 2024. Green Federated Learning over Cloud-RAN with Limited Fronthaul Capacity and Quantized Neural Networks.IEEE Transactions on Wireless Communications23, 5 (May 2024), 4300–

    2024

  25. [25]

    Zhaohui Yang, Mingzhe Chen, Walid Saad, Chong Sang Hong, and Mohammad Shikh-Bahaei. 2021. Energy Efficient Federated Learning over Wireless Commu- nication Networks.IEEE Transactions on Wireless Communications20, 3 (March 2021), 1935–1949. doi:10.1109/TWC.2020.3037554

    arXiv 2021

  26. [26]

    Ashkan Yousefpour, Shen Guo, Ashish Shenoy, Sayan Ghosh, Pierre Stock, Kiwan Maeng, Schalk-Willem Krüger, Michael Rabbat, Carole-Jean Wu, and Ilya Mironov

  27. [31]

    arXiv:2303.14604 [cs.LG] https://arxiv.org/abs/ 2303.14604

    Green Federated Learning. arXiv:2303.14604 [cs.LG] https://arxiv.org/abs/ 2303.14604

    Pith/arXiv arXiv

  28. [2023]

    In2023 IEEE 28th International Workshop on Computer Aided Modeling and Design of Communication Links and Networks (CAMAD)

    A Carbon Tracking Model for Federated Learning: Impact of Quantization and Sparsification. In2023 IEEE 28th International Workshop on Computer Aided Modeling and Design of Communication Links and Networks (CAMAD). IEEE, Edinburgh, United Kingdom, 213–218. doi:10.1109/CAMAD59638.2023.10478391

    arXiv 2023

  29. [2024]

    2024), 1386–1402

    Green, Quantized Federated Learning over Wireless Networks: An Energy-Efficient Design.IEEE Transactions on Wireless Communications23, 2 (Feb. 2024), 1386–1402. doi:10.1109/TWC.2023.3289177

    arXiv 2024

  30. [2025]

    Surv.57, 8, Article 194 (March 2025), 36 pages

    Green Federated Learning: A New Era of Green Aware AI.ACM Comput. Surv.57, 8, Article 194 (March 2025), 36 pages. doi:10.1145/3718363

    doi:10.1145/3718363 2025

  31. [4314]

    doi:10.1109/TWC.2023.3317129

    arXiv 2023

This paper was first reviewed by deepseek-v4-flash on August 3, 2026.