Pith. sign in

REVIEW 4 major objections 3 minor 43 references

HOSt3R: Keypoint-free Hand-Object 3D Reconstruction from RGB images

T0 review · 4 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper proposes HOSt3R, a keypoint-free system that estimates hand-object 3D transformations and shape from monocular RGB video without pre-scanned templates or camera intrinsics.

desk verdict The submission is a document-integrity failure: the abstract advertises a hand-object 3D reconstruction method while the body is an unrelated power-systems paper, so there is nothing to evaluate. read the letter →

arxiv 2508.16465 v2 pith:V5NIN45J submitted 2025-08-22 cs.CV cs.AIcs.HCcs.LGcs.RO

Anilkumar Swamy , Vincent Leroy , Philippe Weinzaepfel , Jean-Sébastien Franco , Grégory Rogez This is my paper · ORCID
classification cs.CVcs.AIcs.HCcs.LGcs.RO
keywords hand-object3Dreconstructionkeypoint-freeobject-agnosticmonocularRGBvideomulti-viewcameraintrinsics-freeSHOWMebenchmarkHO3Dgeneralization
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

The authors set out to show that hand-object 3D reconstruction from monocular RGB video need not rely on keypoint detection, pre-scanned object templates, or camera intrinsics. Their proposed system, HOSt3R, instead aligns pixels across views and feeds that alignment into a multi-view reconstruction stage to recover hand-object 3D transformation and shape. If true, this would remove two common failure sources of existing pipelines—weak textures and hand-object occlusion—and widen reconstruction to arbitrary, unseen objects. The abstract reports state-of-the-art performance on the SHOWMe benchmark and generalization on HO3D. The submitted full text, however, is an unrelated power-systems manuscript, so no method, derivation, or experimental evidence for these claims appears in the document.

What carries the argument

HOSt3R, a dense pixel-alignment mechanism that replaces keypoint detection: instead of detecting hand keypoints or running structure-from-motion, the system learns per-pixel correspondences across monocular views to recover the hand-object 3D transformation, then integrates these alignments in a multi-view reconstruction stage to estimate shape. This alignment is what is supposed to make the method independent of object templates, camera intrinsics, and keypoint detectors.

What would settle it

A direct test: on monocular RGB sequences with ground-truth hand-object poses, vary the camera focal length and occlude the hand; if pose error collapses under either change, the claim of intrinsics-free, keypoint-free alignment is refuted—and the full text currently offers no architecture to run such a test.

Watch

Extended reading notes

Core claim

The central claim is that HOSt3R can estimate the 3D transformation and shape of a hand and an arbitrary manipulated object directly from monocular RGB images or video, without hand-keypoint detection, pre-scanned object templates, or camera intrinsics. The method is presented as a keypoint detector-free alternative to the standard two-stage pipeline of 3D tracking followed by multi-view reconstruction: dense pixel alignment across views supplies the geometry, and a multi-view reconstruction stage recovers shape. The abstract reports state-of-the-art performance on SHOWMe for object-agnostic hand-object transformation and shape estimation, and experiments on HO3D demonstrating generalization

Load-bearing premise

The claim stands on the assumption that learned pixel alignment from monocular RGB alone can recover metric hand-object pose without keypoints, templates, or intrinsics; in this submission that assumption is unsupported because the body is an unrelated power-systems manuscript.

Editorial extensions

If this is right

  • Monocular RGB video would become sufficient for object-agnostic hand-object reconstruction, removing the need for multi-camera rigs or depth sensors.
  • Because no keypoint detector is used, textureless objects and hand-object occlusion would no longer break the first stage of tracking.
  • Without pre-scanned templates or camera intrinsics, the system could be applied to unseen objects in casual, non-intrusive capture settings.
  • If the SHOWMe results hold, HOSt3R would outperform existing two-stage tracking-plus-reconstruction baselines on both transformation and shape estimation.
  • The HO3D experiments suggest the learned alignment would generalize to object categories not seen during training.

Reading between the lines

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

  • If the core claim holds, the two-stage "track then reconstruct" paradigm may collapse into a single learned alignment, and errors that keypoint pipelines accumulate at detection could disappear.
  • The same dense alignment could plausibly transfer to other articulated interactions, such as tool use or hand-hand contact, since the object-agnostic alignment is not tied to hand keypoints.
  • A controlled experiment varying camera focal length while fixing the scene would test whether metric scale comes from a learned hand-size prior or from motion parallax; the paper's intrinsics-free claim predicts the latter is sufficient.
  • If dense alignment drives the results, pose accuracy should degrade gracefully under motion blur rather than fail abruptly, unlike keypoint detectors.
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

4 major / 3 minor

Summary. The submission, arXiv:2508.16465, presents an abstract claiming a new keypoint-free hand-object 3D reconstruction method, HOSt3R, that achieves state-of-the-art performance for object-agnostic hand-object 3D transformation and shape estimation on the SHOWMe benchmark and generalizes to unseen categories on HO3D. The full text, however, is an unrelated IEEE Transactions on Power Systems paper titled 'Wide-Area Power System Oscillations from Large-Scale AI Workloads' by Min-Seung Ko and Hao Zhu, with arXiv footer 2508.16457v2. The body contains no mention of HOSt3R, hand-object reconstruction, keypoint-free alignment, SHOWMe, HO3D, or any computer-vision method or experiment. The abstract's central claims are therefore entirely unsupported by the manuscript content.

Significance. If the claimed method existed and performed as stated, it would be a meaningful advance for monocular hand-object 3D reconstruction: removing keypoint detection, template dependence, and camera-intrinsic requirements would improve scalability and generalization. However, this submission provides no architecture, training objective, evaluation protocol, benchmark numbers, ablations, or theoretical justification. There is no reproducible code or falsifiable prediction beyond the abstract's assertion. The body text is a different paper from a different field, so there is no technical content to assess. The significance of the claimed result cannot be evaluated, and the manuscript in its current form does not meet the standards of a peer-reviewed publication.

major comments (4)
  1. [Full text (all sections)] The body is 'Wide-Area Power System Oscillations from Large-Scale AI Workloads' by Ko and Zhu, with arXiv footer 2508.16457v2. It does not once refer to HOSt3R, hand-object reconstruction, keypoint detection, SHOWMe, HO3D, or any computer-vision concept. The central claim in the abstract—keypoint-free, template-free, intrinsic-free hand-object 3D reconstruction—is therefore completely unsupported. A reader cannot check correctness, reproducibility, generalization, or state-of-the-art status because the relevant content is absent.
  2. [Abstract / Experiments] No experiments on SHOWMe or HO3D are reported anywhere in the manuscript. There are no comparison tables, no ablations, no evaluation metrics (e.g., transformation error, shape Chamfer distance, pose accuracy), no dataset splits, and no baselines. The sentences claiming state-of-the-art performance and generalization to unseen categories are assertions without any supporting evidence.
  3. [Method] No method is described. The abstract's load-bearing premise—that learned pixel alignment from monocular RGB alone can estimate hand-object relative 3D pose and metric scale without keypoints, object templates, or camera intrinsics—is not accompanied by any architecture, loss function, training procedure, or theoretical argument. This is not a case of insufficient ablation; there is no technical content to scrutinize.
  4. [Figures / internal consistency] Several figure captions in the unrelated body (e.g., Figs. 8(a), 9, 10(a)) contain LLM benchmark curves (GSM8k, HumanEval, BoolQ, MMLU) that are unrelated to both the power-systems text and the HOSt3R abstract. This indicates template corruption and makes it impossible to attribute any displayed result to the claimed HOSt3R method.
minor comments (3)
  1. [Title and abstract] The title and abstract describe HOSt3R, while the body is a different paper from the power-systems domain. The mismatch makes the manuscript internally inconsistent and unreadable as a coherent submission.
  2. [arXiv identifier] The paper is labeled arXiv:2508.16465, but the body footer shows arXiv:2508.16457v2 and a different submission date. The reader cannot locate the correct source for the claimed method.
  3. [Header and references] The IEEE Transactions on Power Systems header, author names, and all references pertain to the unrelated power-systems paper, not to the claimed HOSt3R contribution. This further confirms that the manuscript is a corrupted or mis-uploaded file.

Circularity Check

1 steps flagged · score 6.0 of 10

The HOSt3R abstract is unsupported by the unrelated power-systems body; within that body, the ~1 Hz oscillation 'confirmation' is preconditioned because the forcing spectrum was set to 0.5–1.5 Hz precisely because ~1 Hz oscillations were already reported.

  1. fitted input called prediction [Section III-A (Table I) and Section IV-B (Fig. 7 discussion)]
    "recent grid disturbance monitoring reports have linked AI datacenter operations to oscillations near 1 Hz [11], [12], [31]. Thus, in our tests, we will set the baseline range to be [0.5, 1.5] Hz. ... the coherence around 1 Hz exceeds 0.6, indicating strong similarity between the forcing signal and the frequency response. This confirms that the 1 Hz oscillation is primarily driven by the datacenter demand."

    The datacenter forcing model is built with a dominant frequency band of 0.5–1.5 Hz, chosen specifically because ~1 Hz oscillations were already reported as the phenomenon under study. The simulation then finds the grid response concentrated in 0.5–1.3 Hz with a prominent mode near 1.1 Hz, and the paper presents the coherence between this forcing and the response as 'confirmation' that the 1 Hz oscillation is driven by datacenter demand. That confirmation is a restatement of the construction: the output spectrum is the input spectrum propagated through a resonant linear system. The frequency-emergence claim is therefore forced by the fitted input, not an independent prediction.

full rationale

The submitted arXiv:2508.16465 document is internally inconsistent: the abstract claims HOSt3R, a keypoint-free hand-object 3D reconstruction method with SOTA results on SHOWMe and HO3D generalization, but the full text is an unrelated power-systems paper titled 'Wide-Area Power System Oscillations from Large-Scale AI Workloads' by Ko and Zhu. There is no architecture, training objective, benchmark, or experiment for HOSt3R anywhere in the body, so the central CV claim is not circular; it is simply unsupported and unassessable. Within the power-systems body, the one concrete circular reduction is the forcing-frequency construction: Table I and Section III-A set the workload fluctuation band to 0.5–1.5 Hz because disturbances near 1 Hz had already been attributed to AI datacenters, and Section IV-B then reports the resulting 1 Hz response as a confirmation of datacenter-driven oscillation. That step reduces to its own input. The rest of the power-systems study—comparisons of inertia, penetration, sizing, siting, mitigation, and workload ratio—has independent simulation content, though its validation relies on qualitative consistency and private endorsements rather than public measurements. Overall, the paper deserves a partial-circularity score of 6 rather than a higher score because the central HOSt3R claim is not derivable from the body at all, and the body's factor-based results are not all by-construction tautologies.

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

For the claimed HOSt3R method, the ledger cannot be populated: the manuscript contains no equations, architecture, or experimental procedure. The entries below therefore reflect (a) assumptions asserted in the abstract and (b) the unrelated body text, which is in-scope by policy but belongs to a different paper. The body's hand-set stochastic model parameters and its 'educated estimate' workload ratio are listed as free parameters because they are chosen by hand, not derived or measured; they directly shape the oscillation spectra the body then reports. No invented entities appear in either text.

free parameters (2)
  • Stochastic workload model parameters (sigma_xi=0.1, r_tr in [0.55,0.8], sigma_Delta=0.05, mu_Delta=0.3, sigma_eta ranges = Table I values
    From the mismatched body text, Section III and Table I: the values are 'selected based on reported measurements and profiling studies' rather than derived; the nominal frequency range is chosen to match reports of ~1 Hz AI-driven oscillations, preconditioning the resonance the case studies then find.
  • Workload demand ratio P_hat_tr0 : sum(P_hat_trj) : sum(P_hat_ftk) = 9 : 0.5 : 0.5 (90% / 5% / 5%)
    Section III-C footnote: 'These values have not been directly measured but represent an educated estimate'; sensitivity is checked in Section IV-D, which confirms the ratio changes the oscillation content, so the finding is conditional on this hand-set input.
assumptions (2)
  • domain assumption Existing object-agnostic hand-object reconstruction is keypoint-dependent (SfM, hand-keypoint optimization) and this is the main scalability bottleneck.
    Asserted in the abstract to motivate HOSt3R; the provided text has no related-work section, and the 'keypoint-free' framing is not positioned against keypoint-free dense 3D reconstruction methods, so the premise may overstate the field's dependence.
  • domain assumption AI workload power profiles can be modeled as periodic stochastic signals with Gaussian deviations and instantaneous phase transitions (Eqs. 1-12).
    From the mismatched body, Section III; the model form is assumed and informed by cited profiling studies, not by new measurements; phase transitions are assumed instantaneous.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HOSt3R: Keypoint-free Hand-Object 3D Reconstruction from RGB images." pith.science (2026). https://pith.science/paper/V5NIN45J

@misc{pith2026250816465,
  author       = {Pith},
  title        = {Pith review of: HOSt3R: Keypoint-free Hand-Object 3D Reconstruction from RGB images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/V5NIN45J}},
  note         = {Machine review of arXiv:2508.16465}
}
read the original abstract

Hand-object 3D reconstruction has become increasingly important for applications in human-robot interaction and immersive AR/VR experiences. A common approach for object-agnostic hand-object reconstruction from RGB sequences involves a two-stage pipeline: hand-object 3D tracking followed by multi-view 3D reconstruction. However, existing methods rely on keypoint detection techniques, such as Structure from Motion (SfM) and hand-keypoint optimization, which struggle with diverse object geometries, weak textures, and mutual hand-object occlusions, limiting scalability and generalization. As a key enabler to generic and seamless, non-intrusive applicability, we propose in this work a robust, keypoint detector-free approach to estimating hand-object 3D transformations from monocular motion video/images. We further integrate this with a multi-view reconstruction pipeline to accurately recover hand-object 3D shape. Our method, named HOSt3R, is unconstrained, does not rely on pre-scanned object templates or camera intrinsics, and reaches state-of-the-art performance for the tasks of object-agnostic hand-object 3D transformation and shape estimation on the SHOWMe benchmark. We also experiment on sequences from the HO3D dataset, demonstrating generalization to unseen object categories.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 24 canonical work pages

  1. [1]

    The unseen AI dis- ruptions for power grids: LLM-induced transients,

    Y . Li, M. Mughees, Y . Chen, and Y . R. Li, “The unseen AI dis- ruptions for power grids: LLM-induced transients,”arXiv preprint arXiv:2409.11416, 2024

    arXiv 2024

  2. [2]

    Practical guidance and considerations for large load interconnections,

    R. Quintet al., “Practical guidance and considerations for large load interconnections,” Elevate Energy Consulting, Tech. Rep., 2025

    work page 2025

  3. [3]

    2024 United States data center energy usage report,

    A. Shehabiet al., “2024 United States data center energy usage report,” Lawrence Berkeley National Laboratory, Berkeley, CA, Tech. Rep., 2024

    work page 2024

  4. [4]

    Powering intelligence: Analyzing artificial intelligence and data center energy consumption,

    J. Aljbour, T. Wilson, and P. Patel, “Powering intelligence: Analyzing artificial intelligence and data center energy consumption,” EPRI White Paper no. 3002028905, Tech. Rep., 2024

    2024

  5. [5]

    AI load dynamics–a power electronics perspective,

    Y . Li and Y . Li, “AI load dynamics–a power electronics perspective,” arXiv preprint arXiv:2502.01647, 2025

    arXiv 2025

  6. [6]

    Data center model for transient stability analysis of power systems,

    A. Jimenez-Ruiz and F. Milano, “Data center model for transient stability analysis of power systems,”arXiv preprint arXiv:2505.16575, 2025

    arXiv 2025

  7. [7]

    Event records showing data center response to faults,

    R. O’Keefe, “Event records showing data center response to faults,”

  8. [8]

    Unplanned data center load transfer update,

    M. Parker and B. Sterling, “Unplanned data center load transfer update,” 2025. [Online]. Available: https://www.nerc.com/comm/RSTC/ LLTF/LLTF June Workshop Presentations.pdf

    2025

Show all 43 references
  1. [9]

    Integration and interaction of next-generation AI-focused data centers with smart grids and district energy systems: The state-of-the-art, opportunities and challenges,

    Y . Zhang, H. Tang, H. Li, and S. Wang, “Integration and interaction of next-generation AI-focused data centers with smart grids and district energy systems: The state-of-the-art, opportunities and challenges,” Renewable and Sustainable Energy Reviews, vol. 224, p. 116097, 2025

    2025

  2. [10]

    Balance of power: A full-stack approach to power and thermal fluctuations in ml infrastructure,

    H. Gan and P. Ranganathan, “Balance of power: A full-stack approach to power and thermal fluctuations in ml infrastructure,”

  3. [11]

    Characteristics and risks of emerging large loads,

    NERC Large Load Task Force, “Characteristics and risks of emerging large loads,” North American Electric Reliability Corporation (NERC), Tech. Rep., 2025

    2025

  4. [12]

    Available: https://cloud.google.com/blog/topics/systems/ mitigating-power-and-thermal-fluctuations-in-ml-infrastructure?hl=en

    [Online]. Available: https://cloud.google.com/blog/topics/systems/ mitigating-power-and-thermal-fluctuations-in-ml-infrastructure?hl=en

  5. [13]

    Energy and AI,

    D. D’Ambrosioet al., “Energy and AI,” International Energy Agency (IEA), Tech. Rep., 2025

    2025

  6. [14]

    Characterizing power management opportunities for LLMs in the cloud,

    P. Patelet al., “Characterizing power management opportunities for LLMs in the cloud,” inProceedings of the 29th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 3, 2024, pp. 207–222

    2024

  7. [15]

    Artificial intelligence: Supply chain constraints and energy implications,

    A. de Vries-Gao, “Artificial intelligence: Supply chain constraints and energy implications,”Joule, vol. 9, no. 6, 2025

    2025

  8. [16]

    AI, data centers and energy demand: Reassessing and exploring the trends,

    L. de Roucy-Rochegonde and A. Buffard, “AI, data centers and energy demand: Reassessing and exploring the trends,”Ifri Papers, 2025

    2025

  9. [17]

    Characterization and prediction of deep learning workloads in large-scale GPU datacenters,

    Q. Hu, P. Sun, S. Yan, Y . Wen, and T. Zhang, “Characterization and prediction of deep learning workloads in large-scale GPU datacenters,” inProceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, 2021, pp. 1–15

    2021

  10. [18]

    En- ergy and carbon considerations of fine-tuning BERT,

    X. Wang, C. Na, E. Strubell, S. Friedler, and S. Luccioni, “En- ergy and carbon considerations of fine-tuning BERT,”arXiv preprint arXiv:2311.10267, 2023

    2023 arXiv

  11. [19]

    The AI disruption: Challenges and guidance for data center design,

    V . Avelar, P. Donovan, P. Lin, W. Torell, and M. A. T. Arango, “The AI disruption: Challenges and guidance for data center design,”Artificial Intelligence in Medicine, vol. 138, 2023

    2023

  12. [20]

    Scaling intelligence: The exponential growth of AI’s power needs,

    J. Youet al., “Scaling intelligence: The exponential growth of AI’s power needs,” EPRI White Paper no. 3002033669, Tech. Rep., 2025

    2025

  13. [21]

    Characterization of large language model development in the datacenter,

    Q. Huet al., “Characterization of large language model development in the datacenter,” in21st USENIX Symposium on Networked Systems Design and Implementation (NSDI 24), 2024, pp. 709–729

    2024

  14. [22]

    Inference zones: How data centers support real- time AI,

    B. Eichman, “Inference zones: How data centers support real- time AI,” 2024. [Online]. Available: https://www.coresite.com/blog/ inference-zones-how-data-centers-support-real-time-ai

    2024

  15. [23]

    Methodology for fine- grain GPU power visibility and insights,

    V . Singhania, S. Aga, and M. Assem Ibrahim, “Methodology for fine- grain GPU power visibility and insights,”arXiv e-prints, pp. arXiv–2412, 2024

    2024

  16. [24]

    Empirical measurements of AI training power demand on a GPU-accelerated node,

    I. Latif, A. C. Newkirk, M. R. Carbone, A. Munir, Y . Lin, J. Koomey, X. Yu, and Z. Dong, “Empirical measurements of AI training power demand on a GPU-accelerated node,”arXiv preprint arXiv:2412.08602, 2024

    2024 arXiv

  17. [25]

    M100 exadata: a data collection campaign on the cineca’s marconi100 tier-0 supercomputer,

    A. Borghesiet al., “M100 exadata: a data collection campaign on the cineca’s marconi100 tier-0 supercomputer,”Scientific Data, vol. 10, no. 1, p. 288, 2023

    2023

  18. [26]

    The MIT su- percloud dataset,

    S. Samsi, M. L. Weiss, D. Bestor, B. Li, M. Jones, A. Reuther, D. Edelman, W. Arcand, C. Byun, J. Holodnacket al., “The MIT su- percloud dataset,” in2021 IEEE High Performance Extreme Computing Conference (HPEC). IEEE, 2021, pp. 1–8

    2021

  19. [27]

    Not all GPUs are created equal: characterizing variability in large-scale, accelerator-rich systems,

    P. Sinha, A. Guliani, R. Jain, B. Tran, M. D. Sinclair, and S. Venkatara- man, “Not all GPUs are created equal: characterizing variability in large-scale, accelerator-rich systems,” inSC22: International Conference for High Performance Computing, Networking, Storage and Analys...

    2022

  20. [28]

    Understanding GPU power: A survey of profiling, modeling, and simulation methods,

    R. A. Bridges, N. Imam, and T. M. Mintz, “Understanding GPU power: A survey of profiling, modeling, and simulation methods,”ACM Computing Surveys (CSUR), vol. 49, no. 3, pp. 1–27, 2016

    2016

  21. [29]

    PAL: A variability-aware policy for scheduling ML workloads in GPU clusters,

    R. Jain, B. Tran, K. Chen, M. D. Sinclair, and S. Venkataraman, “PAL: A variability-aware policy for scheduling ML workloads in GPU clusters,” inSC24: International Conference for High Performance Computing, Networking, Storage and Analysis. IEEE, 2024, pp. 1–18

    2024

  22. [30]

    Problems and opportunities in training deep learning software systems: An analysis of variance,

    H. V . Phamet al., “Problems and opportunities in training deep learning software systems: An analysis of variance,” inProceedings of the 35th IEEE/ACM international conference on automated software engineering, 2020, pp. 771–783

    2020

  23. [31]

    Battery storage applications at data centers,

    S. G. Vennelaganti and S. Jones, “Battery storage applications at data centers,” 2025. [Online]. Avail- able: https://www.nerc.com/comm/RSTC/LLTF/LLTF April Meeting & Technical Workshop Presentations .pdf

    2025

  24. [32]

    FinGraV: Methodology for fine-grain gpu power visibility and insights,

    V . Singhania, S. Aga, and M. A. Ibrahim, “FinGraV: Methodology for fine-grain gpu power visibility and insights,” in2025 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS). IEEE, 2025, pp. 96–107

    2025

  25. [33]

    Power modeling for effective datacenter planning and compute management,

    A. Radovanovic, B. Chen, S. Talukdar, B. Roy, A. Duarte, and M. Shah- bazi, “Power modeling for effective datacenter planning and compute management,”IEEE Transactions on Smart Grid, vol. 13, no. 2, pp. 1611–1621, 2021. IEEE TRANSACTIONS ON POWER SYSTEMS 14

    2021

  26. [34]

    AI-enabling workloads on large-scale GPU-accelerated system: characterization, opportunities, and implications,

    B. Li, R. Arora, S. Samsi, T. Patel, W. Arcand, D. Bestor, C. Byun, R. B. Roy, B. Bergeron, J. Holodnaket al., “AI-enabling workloads on large-scale GPU-accelerated system: characterization, opportunities, and implications,” in2022 IEEE International Symposium on High- Perform...

    2022

  27. [35]

    Power stabilization for AI training datacenters,

    E. Choukseet al., “Power stabilization for AI training datacenters,”arXiv preprint arXiv:2508.14318, 2025

    2025 arXiv

  28. [36]

    AI datacenter energy dilemma - race for AI datacenter space,

    D. N. D. Patel and J. E. Ontiveros, “AI datacenter energy dilemma - race for AI datacenter space,” 2024. [Online]. Available: https: //semianalysis.com/2024/03/13/ai-datacenter-energy-dilemma-race

    2024

  29. [37]

    AI’s power requirements under exponential growth,

    L. H. Konstantin F. Pilz, Yusuf Mahmood, “AI’s power requirements under exponential growth,” RAND, Tech. Rep., 2025. [Online]. Available: https://www.rand.org/pubs/research reports/RRA3572-1.html

    2025

  30. [38]

    Hybrid symbolic-numeric framework for power system modeling and analysis,

    H. Cui, F. Li, and K. Tomsovic, “Hybrid symbolic-numeric framework for power system modeling and analysis,”IEEE Transactions on Power Systems, vol. 36, no. 2, pp. 1373–1384, 2020

    2020

  31. [39]

    Performance of three mode-meter block-processing algorithms for automated dynamic stability assessment,

    D. J. Trudnowski, J. W. Pierre, N. Zhou, J. F. Hauer, and M. Parashar, “Performance of three mode-meter block-processing algorithms for automated dynamic stability assessment,”IEEE Transactions on Power Systems, vol. 23, no. 2, pp. 680–690, 2008

    2008

  32. [40]

    A tutorial on data-driven eigenvalue identification: Prony analysis, matrix pencil, and eigensystem realization algorithm,

    A. Almunif, L. Fan, and Z. Miao, “A tutorial on data-driven eigenvalue identification: Prony analysis, matrix pencil, and eigensystem realization algorithm,”International Transactions on Electrical Energy Systems, vol. 30, no. 4, p. e12283, 2020

    2020

  33. [41]

    Central station PV plant model validation guideline,

    WECC Renewable Energy Modeling Task Force, “Central station PV plant model validation guideline,” Western Electricity Coordinating Council (WECC), Salt Lake City, UT (United States), Tech. Rep., 2017

    2017

  34. [42]

    Effects of decreasing syn- chronous inertia on power system dynamics—overview of recent ex- periences and marketisation of services,

    B. Hartmann, I. V okony, and I. T ´aczi, “Effects of decreasing syn- chronous inertia on power system dynamics—overview of recent ex- periences and marketisation of services,”International Transactions on Electrical Energy Systems, vol. 29, no. 12, p. e12128, 2019

    2019

  35. [2025]

    Available: https://www.nerc.com/comm/RSTC/LLTF/ LLTF April Meeting & Technical Workshop Presentations .pdf

    [Online]. Available: https://www.nerc.com/comm/RSTC/LLTF/ LLTF April Meeting & Technical Workshop Presentations .pdf

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.