pith. sign in

arxiv: 2606.31340 · v1 · pith:O6WK5DRHnew · submitted 2026-06-30 · ⚛️ physics.geo-ph

Scenario-conditioned flow matching for probabilistic generation of three-component ground-motion waveforms

Yi Ding , Jinjun Hu , Su Chen , Xianwei Liu , Zhongxiang Zhang , Zongchao Li , Xiaojun Li , Lili Xie This is my paper

Pith reviewed 2026-07-01 02:32 UTC · model grok-4.3

classification ⚛️ physics.geo-ph
keywords ground motion generationthree-component waveformsflow matchingwavelet-packet representationPGA interfacescenario conditioningNGA-West2probabilistic seismic model
0
0 comments X

The pith

WaveFlowGMM generates three-component ground-motion waveforms by predicting PGA separately from shape generation in wavelet-packet space.

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

The paper introduces a two-stage probabilistic model that first estimates component-wise peak ground acceleration and its cross-component covariance from earthquake scenarios using symbolic learning, then generates normalized waveform shapes via flow matching in an invertible wavelet-packet representation before rescaling. This separation is tested on an event-level holdout set from NGA-West2 data, where the outputs recover magnitude, distance, and site scaling relations, produce near-zero residuals in peaks and spectra, maintain three-component amplitude correlations, and integrate to velocity and displacement time histories without systematic drift. The approach addresses the need in performance-based seismic risk assessment for full acceleration histories that are compatible with specified source, path, and site conditions rather than relying solely on scalar intensity measures.

Core claim

WaveFlowGMM is a two-stage model in which an amplitude stage uses physics-informed symbolic learning to estimate PGA medians and covariance while a waveform stage applies few-step AlphaFlow in invertible wavelet-packet coefficient space to produce normalized three-component histories that are then rescaled by sampled PGA values, and validation on NGA-West2 holdout events confirms that the generated motions recover main scaling trends, keep peak and spectral residuals near zero, preserve three-component dependence, and integrate without drift.

What carries the argument

Two-stage structure that treats PGA as an amplitude interface between scenario conditioning and AlphaFlow-based waveform generation in wavelet-packet space.

If this is right

  • Generated motions recover the main magnitude, distance, and site scaling present in the training data.
  • Peak and spectral residuals remain close to zero on holdout events.
  • Three-component amplitude dependence is preserved across the generated histories.
  • Integration of the generated accelerations produces velocity and displacement histories without systematic drift.

Where Pith is reading between the lines

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

  • The separation of amplitude and shape stages may allow independent updating of the PGA prediction module when new ground-motion data or physics constraints become available.
  • Because the waveform stage operates on normalized histories, the same flow-matching machinery could be retrained for other target intensity measures if a suitable interface variable is identified.
  • The absence of integration drift suggests the outputs are directly usable as input to nonlinear structural response analyses without additional baseline correction.

Load-bearing premise

That PGA supplies a sufficient amplitude interface which fully decouples scenario conditioning from waveform shape without losing information critical to the final motions, and that the wavelet-packet representation preserves everything needed for drift-free integration to velocity and displacement.

What would settle it

A new event-level holdout set in which the generated three-component accelerations, after integration, exhibit systematic drift in velocity or displacement time histories or produce spectral residuals that deviate substantially from zero across a range of periods.

Figures

Figures reproduced from arXiv: 2606.31340 by Jinjun Hu, Lili Xie, Su Chen, Xianwei Liu, Xiaojun Li, Yi Ding, Zhongxiang Zhang, Zongchao Li.

Figure 1
Figure 1. Figure 1: Two-stage WaveFlowGMM framework. (a) Wavelet-packet coefficient representation; red arrows denote the inverse￾transform reconstruction path. (b) PISL amplitude stage, giving the three-component PGA medians and sampling amplitude anchors from the inter-event covariance Σ𝜏 and intra-event covariance Σ𝜙 . (c) Conditional U-Net waveform generator, in which scenario and PGA tokens modulate the velocity field th… view at source ↗
Figure 2
Figure 2. Figure 2: Condition distribution of the NGA-West2 training and test sets [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: PGA residual covariance check. H1 IMs overlap over the main amplitude range, and the peak and spectral measures show no tail truncation or mode split￾ting. Integrated and energy measures shift slightly towards higher generated values, consistent with the negative biases in [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: H1 marginal distributions of intensity measures. source scaling beyond the PGA amplitude anchor. The gen￾erated medians remain continuous at small magnitudes and fall within the main observed scatter. Although records are sparse at the large-magnitude, long-period end, the flattening of the 5 s curve remains consistent with the available obser￾vations and may indicate long-period magnitude saturation. Unde… view at source ↗
Figure 5
Figure 5. Figure 5: RotD50 path-attenuation curves. PGA, PGV, and PSA at 0.2, 1, 3, and 5 s are shown as functions of 𝑅𝐽𝐵 for strike-slip conditions at 𝑉𝑆30 = 760 m/s. 4.3. Event residuals and period-dependent dispersion This section uses inter-event and intra-event residual di￾agnostics to check whether systematic bias remains beyond the median-scaling curves. For record 𝑗 from event 𝑖, the residuals of any scalar intensity … view at source ↗
Figure 6
Figure 6. Figure 6: RotD50 source-scaling curves. PGA and multi-period PSA are shown as functions of 𝑀𝑤 for strike-slip conditions at 𝑅𝐽𝐵 = 30 km and 𝑉𝑆30 = 760 m/s. For the vertical component, WaveFlowGMM is also close to SBSA16, with the main difference concentrated around the short-period peak. In both components, the dispersion is dominated by the within-event term 𝜙, while the period dependence of 𝜏 follows a shape simil… view at source ↗
Figure 7
Figure 7. Figure 7: RotD50 site-response curves. PGA, PGV, and PSA are shown as functions of 𝑉𝑆30 under three magnitude-distance scenarios [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Observed-generated waveforms and condition-perturbation samples. The upper panels show three-component acceleration, velocity, and displacement, and the lower panels show generated acceleration under distance, site, and magnitude perturbations. samples the three components as independent marginal am￾plitudes loses this dependence even if each marginal distri￾bution is reasonable. The full covariance theref… view at source ↗
Figure 9
Figure 9. Figure 9: Three-component multi-measure summary of binned intra-event residuals [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Three-component multi-measure summary of binned inter-event residuals. ground-motion databases for the next generation of GMMs [65]. Such larger and more diverse datasets should allow future versions of waveform-level GMMs to learn sharper conditional distributions, reduce smoothing in data-poor regions, and test regional transferability more directly. 6. Conclusions This study introduced WaveFlowGMM, a t… view at source ↗
Figure 11
Figure 11. Figure 11: H1 residual trends along source, path, and site axes [PITH_FULL_IMAGE:figures/full_fig_p016_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Sample-median PSA residual dispersion versus spectral period. Yi Ding et al.: Preprint submitted to Elsevier Page 15 of 20 [PITH_FULL_IMAGE:figures/full_fig_p016_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Sampling cost and waveform accuracy. Data Availability Statement The NGA-West2 database used in this study is available from the Pacific Earthquake Engineering Research Center (https://ngawest2.berkeley.edu/, last accessed May 2024). A. Residual diagnostics for the H2 and V components [PITH_FULL_IMAGE:figures/full_fig_p017_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: H2 residual diagnostics along the three conditioning axes. Yi Ding et al.: Preprint submitted to Elsevier Page 17 of 20 [PITH_FULL_IMAGE:figures/full_fig_p018_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: V residual diagnostics along the three conditioning axes. Yi Ding et al.: Preprint submitted to Elsevier Page 18 of 20 [PITH_FULL_IMAGE:figures/full_fig_p019_15.png] view at source ↗
read the original abstract

Performance-based seismic risk assessment requires three-component acceleration histories compatible with specified source, path, and site conditions. Conventional ground-motion prediction equations provide scalar intensity measures, while many generative waveform models learn amplitude and waveform shape within a single high-dimensional target. We present WaveFlowGMM, a two-stage probabilistic ground-motion model that uses peak ground acceleration (PGA) as an amplitude interface between scenario conditioning and waveform generation. The amplitude stage uses physics-informed symbolic learning to estimate component-wise PGA medians and a full cross-component covariance. The waveform stage uses few-step AlphaFlow in an invertible wavelet-packet coefficient space to generate normalised three-component histories that are rescaled by sampled PGA. Tests on an event-level NGA-West2 holdout set show that the generated motions recover the main magnitude, distance, and site scaling, keep peak and spectral residuals close to zero, preserve three-component amplitude dependence, and yield velocity and displacement histories without systematic drift after integration of the generated three-component acceleration histories. The framework provides an interpretable and computationally efficient candidate component for waveform-level seismic hazard and risk analysis.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The manuscript presents WaveFlowGMM, a two-stage probabilistic model for generating three-component ground-motion acceleration waveforms conditioned on seismic scenarios. Stage one uses physics-informed symbolic learning to predict component-wise PGA medians and full cross-component covariance. Stage two applies few-step AlphaFlow (flow matching) in an invertible wavelet-packet coefficient space to produce normalized waveform shapes that are subsequently rescaled by sampled PGA values. On an event-level NGA-West2 holdout set, the generated motions are reported to recover magnitude-distance-site scaling, maintain near-zero peak and spectral residuals, preserve three-component amplitude dependence, and integrate to drift-free velocity and displacement time histories.

Significance. If the quantitative validation holds, the work supplies an interpretable, computationally efficient component for waveform-level seismic hazard and risk analysis by cleanly separating amplitude (via PGA) from shape generation. The invertible wavelet-packet representation and physics-informed symbolic regression for the amplitude stage are concrete strengths that could improve physical consistency over single-stage generative approaches.

major comments (2)
  1. [Abstract] Abstract: the central claim that 'peak and spectral residuals close to zero' and that 'main magnitude, distance, and site scaling' are recovered is presented without numerical values, error bars, sample counts, or any description of whether residuals were examined at fixed PGA. This information is load-bearing for assessing whether the PGA interface discards frequency-dependent correlations (as flagged in the skeptic note) and therefore whether the two-stage decoupling is valid.
  2. [Abstract] Abstract (holdout test description): no test is reported that checks for residual dependence of spectral shape, duration, or phase on magnitude or distance once PGA is fixed. If such dependence exists, the generated spectra will be biased even when PGA and peak residuals appear small; the manuscript must supply this diagnostic to support the sufficiency of PGA as the amplitude interface.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and the specific concerns raised about the abstract and the validation of the PGA interface. We address both major comments below and will revise the manuscript accordingly to improve clarity and provide the requested diagnostics.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that 'peak and spectral residuals close to zero' and that 'main magnitude, distance, and site scaling' are recovered is presented without numerical values, error bars, sample counts, or any description of whether residuals were examined at fixed PGA. This information is load-bearing for assessing whether the PGA interface discards frequency-dependent correlations (as flagged in the skeptic note) and therefore whether the two-stage decoupling is valid.

    Authors: We agree that the abstract would be strengthened by quantitative detail. In the revision we will add the specific values: mean log-residuals for PGA and for spectral accelerations at periods 0.1–10 s (with standard deviations), the number of holdout events (N=XXX), and explicit reference to the figures that demonstrate magnitude-distance-site scaling recovery. The residuals reported are for the full two-stage model (sampled PGA followed by shape generation) rather than conditioned on fixed PGA; we will clarify this distinction in the abstract and add a sentence noting that the shape stage is trained to be independent of scenario given the PGA amplitude. This directly addresses the concern about frequency-dependent correlations being discarded by the interface. revision: yes

  2. Referee: [Abstract] Abstract (holdout test description): no test is reported that checks for residual dependence of spectral shape, duration, or phase on magnitude or distance once PGA is fixed. If such dependence exists, the generated spectra will be biased even when PGA and peak residuals appear small; the manuscript must supply this diagnostic to support the sufficiency of PGA as the amplitude interface.

    Authors: We accept that an explicit conditional-independence diagnostic is needed to substantiate the PGA interface. Although the overall validation (Section 4) shows small residuals and correct scaling, we did not report a test that bins spectral-shape, duration, or phase residuals by magnitude or distance at fixed PGA. In the revised manuscript we will add this analysis (e.g., residual trends versus magnitude within narrow PGA bins on the holdout set) and include the corresponding figure or table. If the test reveals residual dependence we will discuss its magnitude and implications for the two-stage approach. revision: yes

Circularity Check

0 steps flagged

No circularity; two-stage PGA interface is an explicit modeling choice with independent evaluation

full rationale

The abstract and described framework present the PGA amplitude interface, symbolic learning for medians/covariance, and AlphaFlow in wavelet-packet space as deliberate architectural decisions rather than quantities that reduce to the target waveforms by construction. Evaluation on an event-level NGA-West2 holdout set tests recovery of scaling and residuals without any indication that predictions are statistically forced by the inputs or by self-citation chains. No load-bearing self-definitional steps, fitted-input predictions, or uniqueness theorems from prior author work are visible in the provided text. The derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities can be extracted beyond the modeling choice of using PGA as the amplitude interface.

pith-pipeline@v0.9.1-grok · 5739 in / 1184 out tokens · 38714 ms · 2026-07-01T02:32:27.989749+00:00 · methodology

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

63 extracted references · 41 canonical work pages · 12 internal anchors

  1. [1]

    Moehle, G

    J. Moehle, G. G. Deierlein, A framework methodology for performance-based earthquake engineering, in: 13th World Confer- ence on Earthquake Engineering, Vancouver, Canada, 2004, place- holder entry; verify before submission

    2004

  2. [2]

    rep., Applied Technology Council, Washington, DC, placeholder entry; verify before submis- sion (2018)

    Federal Emergency Management Agency, Seismic performance as- sessment of buildings (fema p-58), Tech. rep., Applied Technology Council, Washington, DC, placeholder entry; verify before submis- sion (2018)

    2018

  3. [3]

    T. D. Ancheta, R. B. Darragh, J. P. Stewart, E. Seyhan, W. J. Silva, B. S.-J. Chiou, K. E. Wooddell, R. W. Graves, A. R. Kottke, D. M. Boore, T. Kishida, J. L. Donahue, NGA-West2 database, Earthquake Spectra 30 (3) (2014) 989–1005.doi:10.1193/070913EQS197M

    work page doi:10.1193/070913eqs197m 2014

  4. [4]

    J. J. Bommer, A. B. Acevedo, Hazard-consistent earthquake scenar- ios, Journal of Earthquake Engineering 8 (sup001) (2004) 43–91, placeholder entry; verify before submission

    2004

  5. [5]

    N. Luco, P. Bazzurro, Structure-specific scalar intensity measures, EarthquakeEngineering&StructuralDynamics36(13)(2007)1813– 1835, placeholder entry; verify before submission

    2007

  6. [6]

    N. A. Abrahamson, W. J. Silva, R. Kamai, Summary of the ASK14 groundmotionrelationforactivecrustalregions,EarthquakeSpectra 30 (3) (2014) 1025–1055.doi:10.1193/070913EQS198M

    work page doi:10.1193/070913eqs198m 2014

  7. [7]

    D. M. Boore, J. P. Stewart, E. Seyhan, G. M. Atkinson, NGA-West2 equationsforpredictingPGA,PGV,and5%dampedPSAforshallow crustal earthquakes, Earthquake Spectra 30 (3) (2014) 1057–1085. doi:10.1193/070113EQS184M

    work page doi:10.1193/070113eqs184m 2014

  8. [8]

    K. W. Campbell, Y. Bozorgnia, NGA-West2 ground motion model fortheaveragehorizontalcomponentsofPGA,PGV,and5%damped linearaccelerationresponsespectra,EarthquakeSpectra30(3)(2014) 1087–1115.doi:10.1193/062913EQS175M

    work page doi:10.1193/062913eqs175m 2014

  9. [9]

    B. S.-J. Chiou, R. R. Youngs, Update of the chiou and youngs NGA model for the average horizontal component of peak ground motion and response spectra, Earthquake Spectra 30 (3) (2014) 1117–1153. doi:10.1193/072813EQS219M

    work page doi:10.1193/072813eqs219m 2014

  10. [10]

    R. H. Scanlan, K. Sachs, Earthquake time histories and response spectra, Journal of the Engineering Mechanics Division 100 (4) (1974) 635–655.doi:10.1061/JMCEA3.0001911

    work page doi:10.1061/jmcea3.0001911 1974

  11. [11]

    Al Atik, N

    L. Al Atik, N. Abrahamson, An improved method for nonstationary spectral matching, Earthquake Spectra 26 (3) (2010) 601–617.doi: 10.1193/1.3459159

    work page doi:10.1193/1.3459159 2010

  12. [12]

    Ni, W.-C

    S.-H. Ni, W.-C. Xie, M. Pandey, Application of hilbert-huang trans- form in generating spectrum-compatible earthquake time histories, International Scholarly Research Notices 2011 (1) (2011) 563678. doi:10.5402/2011/563678

    work page doi:10.5402/2011/563678 2011

  13. [13]

    Huang, G

    D. Huang, G. Wang, Energy-compatible and spectrum-compatible (ECSC)groundmotionsimulationusingwaveletpackets,Earthquake Engineering and Structural Dynamics 46 (11) (2017) 1855–1873. doi:10.1002/eqe.2887

    work page doi:10.1002/eqe.2887 2017

  14. [14]

    P.M.Mai,G.C.Beroza,Ahybridmethodforcalculatingnear-source broadbandseismograms,PhysicsoftheEarthandPlanetaryInteriors 137 (2002) 183–199, placeholder entry; verify before submission

    2002

  15. [15]

    R.W.Graves,A.Pitarka,Broadbandground-motionsimulationusing a hybrid approach, Bulletin of the Seismological Society of America 100 (5A) (2010) 2095–2123.doi:10.1785/0120100057

    work page doi:10.1785/0120100057 2010

  16. [16]

    A. J. Rodgers, A. Pitarka, R. Pankajakshan, B. Sjögreen, N. A. Petersson, Regional-scale 3D ground-motion simulations of mw 7 earthquakes on the hayward fault, northern california resolving fre- quencies 0–10 hz and including site-response corrections, Bulletin of the Seismological Society of America 110 (6) (2020) 2862–2881. doi:10.1785/0120200147

    work page doi:10.1785/0120200147 2020

  17. [17]

    D. M. Boore, Simulation of ground motion using the stochastic method, Pure and Applied Geophysics 160 (3) (2003) 635–676.doi: 10.1007/PL00012553

    work page doi:10.1007/pl00012553 2003

  18. [18]

    K. B. Olsen, R. Takedatsu, Hybrid broadband ground-motion simu- lation, Bulletin of the Seismological Society of America 105 (2015) 951–968, placeholder entry; verify before submission

    2015

  19. [19]

    Bayless, N

    J. Bayless, N. A. Abrahamson, Summary of the ba18 ground-motion model, Bulletin of the Seismological Society of America 109 (2019) 2088–2105, placeholder entry; verify before submission

    2019

  20. [20]

    M. A. Florez, M. Caporale, P. Buabthong, Z. E. Ross, D. Asimaki, M.-A. Meier, Data-driven synthesis of broadband earthquake ground motionsusingartificialintelligence,BulletinoftheSeismologicalSo- cietyofAmerica112(4)(2022)1979–1996.doi:10.1785/0120210264

    work page doi:10.1785/0120210264 2022

  21. [21]

    R. D. D. Esfahani, F. Cotton, M. Ohrnberger, F. Scherbaum, TFCGAN: Nonstationary ground-motion simulation in the time– frequency domain using conditional generative adversarial network (CGAN) and phase retrieval methods, Bulletin of the Seismological SocietyofAmerica113(1)(2023)453–467.doi:10.1785/0120220068

    work page doi:10.1785/0120220068 2023

  22. [22]

    Y. Shi, G. Lavrentiadis, D. Asimaki, Z. E. Ross, K. Azizzadenesheli, Broadbandground-motionsynthesisviagenerativeadversarialneural operators: Development and validation, Bulletin of the Seismolog- ical Society of America 114 (4) (2024) 2151–2171.doi:10.1785/ 0120230207

    2024

  23. [24]

    Yamaguchi, Y

    J. Yamaguchi, Y. Tomozawa, T. Saka, Site-specific ground-motion waveform generation using a conditional generative adversarial net- work and generalized inversion technique, Bulletin of the Seismo- logical Society of America 114 (4) (2024) 2118–2137.doi:10.1785/ 0120230209

    2024

  24. [25]

    G. Lin, X. Hu, Near-fault ground motion synthesis based on condi- tional generation adversarial network, Computers and Structures 313 (2025) 107740.doi:10.1016/j.compstruc.2025.107740

    work page doi:10.1016/j.compstruc.2025.107740 2025

  25. [26]

    K. Chen, H. Pan, M. Zhang, Z.-H. Li, Ground motion simulation via generative adversarial network, Applied Geophysics 22 (3) (2025) 684–697.doi:10.1007/s11770-025-1154-7

    work page doi:10.1007/s11770-025-1154-7 2025

  26. [28]

    P. Ren, R. Nakata, M. Lacour, I. Naiman, N. Nakata, J. Song, Z. Bi, O. A. Malik, D. Morozov, O. Azencot, N. B. Erichson, M. W. Mahoney, Learning earthquake ground motions via conditional gen- erativemodeling,NatureCommunications(Mar.2026).doi:10.1038/ s41467-026-70719-2

    2026

  27. [29]

    Bergmeister, K

    A. Bergmeister, K. H. Palgunadi, A. Bosisio, L. Ermert, M. Ko- roni, N. Perraudin, S. Dirmeier, M.-A. Meier, High resolution seis- mic waveform generation using denoising diffusion, arXiv preprint arXiv:2410.19343 (Oct. 2024).arXiv:2410.19343,doi:10.48550/ arXiv.2410.19343

    work page arXiv 2024

  28. [30]

    V., Elvira, T

    Y. Huang, X. Sun, J. You, C. Yang, D. Lu, J. Zhang, Ground-motion generationsusingmulti-labelconditionalembedding–conditionalde- noising diffusion probabilistic model (ML–cDDPM), Soil Dynamics and Earthquake Engineering 191 (2025) 109274.doi:10.1016/j. soildyn.2025.109274

    work page doi:10.1016/j 2025

  29. [31]

    Z. Bi, N. Nakata, R. Nakata, P. Ren, X. Wu, M. W. Mahoney, Advancing data-driven broadband seismic wavefield simulation with multiconditional diffusion model, IEEE Transactions on Geoscience and Remote Sensing 63 (2025) 1–9.doi:10.1109/TGRS.2025.3591881

    work page doi:10.1109/tgrs.2025.3591881 2025

  30. [32]

    J. Jung, J. Lee, C. Jung, H. Kim, B. Jung, D. Lee, Broadband ground motion synthesis by diffusion model with minimal condition, arXiv preprint arXiv:2412.17333 (May 2025).arXiv:2412.17333,doi:10. 48550/arXiv.2412.17333

    work page arXiv 2025

  31. [33]

    Waveform-Based Probabilistic Seismic Hazard Analysis Using Ground-Motion Generative Models

    Y. Matsumoto, T. Yaoyama, S. Lee, A. Iwaki, T. Itoi, Waveform- Based Probabilistic Seismic Hazard Analysis Using Ground-Motion Generative Models, arXiv preprint arXiv:2511.22106 (Nov. 2025). arXiv:2511.22106,doi:10.48550/arXiv.2511.22106

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2511.22106 2025

  32. [34]

    Sohl-Dickstein, E

    J. Sohl-Dickstein, E. Weiss, N. Maheswaranathan, S. Ganguli, Deep Unsupervised Learning using Nonequilibrium Thermodynamics, in: Proceedingsofthe32ndInternationalConferenceonMachineLearn- ing, International Conference on Machine Learning (ICML), PMLR, Yi Ding et al.:Preprint submitted to ElsevierPage 19 of 20 Scenario-conditioned flow matching for three-c...

    2015

  33. [35]

    6840–6851

    J.Ho,A.Jain,P.Abbeel,Denoisingdiffusionprobabilisticmodels,in: AdvancesinNeuralInformationProcessingSystems,Vol.33,Neural Information Processing Systems (NeurIPS), Curran Associates, Inc., 2020, pp. 6840–6851

    2020

  34. [36]

    Elucidating the Design Space of Diffusion-Based Generative Models

    T. Karras, M. Aittala, T. Aila, S. Laine, Elucidating the de- sign space of diffusion-based generative models, arXiv preprint arXiv:2206.00364 (2022).arXiv:2206.00364,doi:10.48550/arXiv. 2206.00364

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv 2022

  35. [37]

    High-Resolution Image Synthesis with Latent Diffusion Models

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, B. Ommer, High- resolution image synthesis with latent diffusion models, arXiv preprint arXiv:2112.10752 (2021).arXiv:2112.10752,doi:10.48550/ arXiv.2112.10752

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  36. [38]

    Scalable Diffusion Models with Transformers

    W. Peebles, S. Xie, Scalable diffusion models with transformers, arXiv preprint arXiv:2212.09748 (2022).arXiv:2212.09748,doi:10. 48550/arXiv.2212.09748

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  37. [39]

    Progressive Distillation for Fast Sampling of Diffusion Models

    T. Salimans, J. Ho, Progressive Distillation for Fast Sampling of Dif- fusion Models, arXiv preprint arXiv:2202.00512 (Jun. 2022).arXiv: 2202.00512,doi:10.48550/arXiv.2202.00512

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2202.00512 2022

  38. [40]

    arXiv preprint arXiv:2311.17042 , year=

    A. Sauer, D. Lorenz, A. Blattmann, R. Rombach, Adversarial diffu- siondistillation,arXivpreprintarXiv:2311.17042(Nov.2023).arXiv: 2311.17042,doi:10.48550/arXiv.2311.17042

    work page doi:10.48550/arxiv.2311.17042 2023

  39. [41]

    Y. Song, P. Dhariwal, M. Chen, I. Sutskever, Consistency Models, arXiv preprint arXiv:2303.01469 (May 2023).arXiv:2303.01469, doi:10.48550/arXiv.2303.01469

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2303.01469 2023

  40. [42]

    Y. Song, P. Dhariwal, Improved techniques for training consistency models, arXiv preprint arXiv:2310.14189 (Oct. 2023).arXiv:2310. 14189,doi:10.48550/arXiv.2310.14189

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2310.14189 2023

  41. [43]

    C. Lu, Y. Song, Simplifying, stabilizing and scaling continuous-time consistency models, arXiv preprint arXiv:2410.11081 (Mar. 2025). arXiv:2410.11081,doi:10.48550/arXiv.2410.11081

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2410.11081 2025

  42. [44]

    Flow Matching for Generative Modeling

    Y. Lipman, R. T. Q. Chen, H. Ben-Hamu, M. Nickel, M. Le, Flow matching for generative modeling, arXiv preprint arXiv:2210.02747 (2022).doi:10.48550/ARXIV.2210.02747

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2210.02747 2022

  43. [45]

    X. Liu, C. Gong, Q. Liu, Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow, arXiv preprint arXiv:2209.03003 (Sep. 2022).arXiv:2209.03003,doi:10.48550/ arXiv.2209.03003

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  44. [46]

    Mean Flows for One-step Generative Modeling

    Z.Geng,M.Deng,X.Bai,J.Z.Kolter,K.He,Meanflowsforone-step generative modeling, arXiv preprint arXiv:2505.13447 (May 2025). arXiv:2505.13447,doi:10.48550/arXiv.2505.13447

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2505.13447 2025

  45. [47]

    Zhang, A

    H. Zhang, A. Siarohin, W. Menapace, M. Vasilkovsky, S. Tulyakov, Q. Qu, I. Skorokhodov, AlphaFlow: Understanding and improving MeanFlow models, in: The Fourteenth International Conference on Learning Representations, OpenReview, 2025

    2025

  46. [48]

    2026).arXiv:2602.07405,doi:10.48550/arXiv.2602.07405

    Y.Ding,S.Chen,J.Hu,X.Hu,Q.Zhao,X.Li,Waveletpacket-based diffusion model for ground motion generation with multi-conditional energyandspectralmatching,arXivpreprintarXiv:2602.07405(Feb. 2026).arXiv:2602.07405,doi:10.48550/arXiv.2602.07405

    work page doi:10.48550/arxiv.2602.07405 2026

  47. [49]

    S. Chen, X. Liu, L. Fu, S. Wang, B. Zhang, X. Li, Physics sym- bolic learner for discovering ground-motion models via NGA-West2 database, Earthquake Engineering and Structural Dynamics 53 (1) (2024) 138–151.doi:10.1002/eqe.4013

    work page doi:10.1002/eqe.4013 2024

  48. [50]

    X. Liu, C. Su, X. Li, L. Fu, F. Cotton, A hybrid symbolic learning approach for Ground-Motion model Development, Journal of Asian Earth Sciences 281 (2025) 106498.doi:10.1016/j.jseaes.2025. 106498

    work page doi:10.1016/j.jseaes.2025 2025

  49. [51]

    N. A. Abrahamson, R. R. Youngs, A stable algorithm for regression analyses using the random effects model, Bulletin of the Seismo- logical Society of America 82 (1) (1992) 505–510.doi:10.1785/ BSSA0820010505

    1992

  50. [52]

    Daubechies, Orthonormal bases of compactly supported wavelets, Communications on Pure and Applied Mathematics 41 (7) (1988) 909–996.doi:10.1002/cpa.3160410705

    I. Daubechies, Orthonormal bases of compactly supported wavelets, Communications on Pure and Applied Mathematics 41 (7) (1988) 909–996.doi:10.1002/cpa.3160410705

    work page doi:10.1002/cpa.3160410705 1988

  51. [53]

    Frans, D

    K. Frans, D. Hafner, S. Levine, P. Abbeel, One step diffusion via shortcut models, in: International Conference on Learning Represen- tations, OpenReview, 2025, placeholder entry; verify before submis- sion

    2025

  52. [54]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, I. Polosukhin, Attention is all you need, in: Pro- ceedings of the 31st International Conference on Neural Information Processing Systems, NIPS’17, Neural Information Processing Sys- tems (NeurIPS), Curran Associates Inc., Red Hook, NY, USA, 2017, pp. 6000–6010

    2017

  53. [55]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, U-net: Convolutional networks for biomedical image segmentation, in: N. Navab, J. Hornegger, W. M. Wells, A. F. Frangi (Eds.), Medical Image Computing and Computer-assisted Intervention – MICCAI 2015, Medical Image ComputingandComputer-AssistedIntervention(MICCAI),Springer International Publishing, Cham, 2015, pp. 234–...

    2015

  54. [56]

    D.M.Boore,Orientation-independent,nongeometric-meanmeasures of seismic intensity from two horizontal components of motion, Bulletin of the Seismological Society of America 100 (4) (2010) 1830–1835.doi:10.1785/0120090400

    work page doi:10.1785/0120090400 2010

  55. [57]

    G. W. Housner, Spectrum intensities of strong-motion earthquakes, Bulletin of the Seismological Society of America 42 (2) (1952) 73– 80

    1952

  56. [58]

    Arias, A measure of earthquake intensity, in: R

    A. Arias, A measure of earthquake intensity, in: R. J. Hansen (Ed.), SeismicDesignforNuclearPowerPlants,MITPress,1970,pp.438– 483

    1970

  57. [59]

    EPRI, A criterion for determining exceedance of the operating basis earthquake, Tech. Rep. NP-5930, Electric Power Research Institute, Palo Alto, CA (1988)

    1988

  58. [60]

    Y.-J. Park, A. H.-S. Ang, Y.-K. Wen, Mechanistic seismic damage model for reinforced concrete, Journal of Structural Engineering 111 (4) (1985) 722–739.doi:10.1061/(ASCE)0733-9445(1985)111: 4(722)

    work page doi:10.1061/(asce)0733-9445(1985)111: 1985

  59. [61]

    M. D. Trifunac, A. G. Brady, A study on the duration of strong earthquake ground motion, Bulletin of the Seismological Society of America 65 (3) (1975) 581–626

    1975

  60. [62]

    Afshari, J

    K. Afshari, J. P. Stewart, Physically parameterized prediction equa- tions for significant duration in active crustal regions, Earthquake Spectra 32 (4) (2016) 2057–2081.doi:10.1193/063015EQS106M

    work page doi:10.1193/063015eqs106m 2016

  61. [63]

    J. P. Stewart, D. M. Boore, E. Seyhan, G. M. Atkinson, NGA-West2 equations for predicting vertical-component PGA, PGV, and 5%- damped PSA from shallow crustal earthquakes, Earthquake Spectra 32 (2) (2016) 1005–1031.doi:10.1193/072114eqs116m

    work page doi:10.1193/072114eqs116m 2016

  62. [64]

    J. Song, C. Meng, S. Ermon, Denoising diffusion implicit models, arXivpreprintarXiv:2010.02502(Oct.2022).arXiv:2010.02502,doi: 10.48550/arXiv.2010.02502

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2010.02502 2010

  63. [65]

    Kishida, G

    J.P.Stewart,T.E.Buckreis,R.R.Shams,S.Mohammed,L.Meibai, T. Kishida, G. Lanzano, L. Luzi, P. Zimmaro, B. A. Bradley, J.- Y. Huang, C. C. Nweke, S. J. Brandenberg, J. L. Donahue, Y. Bo- zorgnia, Overview of the nga-west3 ground motion database, Poster Presentation at the 2024 SCEC Annual Meeting, Poster #171, SCEC Contribution #13977 (Sep. 2024). URLhttps:/...

    2024