arxiv: 2602.16515 · v2 · submitted 2026-02-18 · ⚛️ physics.geo-ph

Recognition: no theorem link

Generative deep learning improves reconstruction of global historical climate records

Zhen Qian , Teng Liu , Sebastian Bathiany , Shangshang Yang , Philipp Hess , Nils Bochow , Christian Burmester , Maximilian Gelbrecht

show 2 more authors

Brian Groenke Niklas Boers

Authors on Pith no claims yet

Pith reviewed 2026-05-15 21:12 UTC · model grok-4.3

classification ⚛️ physics.geo-ph

keywords generative deep learninghistorical climate reconstructionArctic warmingprobabilistic inferenceclimate variabilitysparse observationstemperature fieldsprecipitation

0 comments

The pith

A generative deep learning framework reconstructs consistent historical climate fields back to 1850 and indicates stronger early Arctic warming than prior estimates.

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

The paper establishes a probabilistic generative deep learning method that infers spatiotemporally consistent temperature and precipitation fields from sparse observations since 1850 by applying a learned prior of Earth system dynamics. Traditional statistical methods smooth out local features and extremes, but this approach preserves higher-order statistics and variability while quantifying uncertainty. If the central claim holds, it revises upward estimates of early 20th-century global warming, driven mainly by polar amplification exceeding previous benchmarks by 0.15-0.29 °C per decade in the Arctic from 1900-1980, and identifies intense localized modern hotspots. Readers should care because more accurate historical records directly improve assessments of anthropogenic climate change and natural variability.

Core claim

The authors develop a unified probabilistic generative deep learning framework that leverages a generative prior of Earth system dynamics learned from modern data to perform probabilistic inference on sparse historical observations, yielding consistent temperature and precipitation reconstructions back to 1850 that reduce smoothing artifacts present in conventional products and reveal higher early 20th-century global warming levels, particularly from pronounced polar warming with Arctic trends 0.15-0.29 °C per decade above established values for 1900-1980, while resolving previously unrecognized intense localized hotspots in the modern Barents Sea and Northeastern Greenland.

What carries the argument

The learned generative prior of Earth system dynamics, which supports probabilistic inference to reconstruct sparse data while preserving higher-order climate statistics and producing uncertainty-aware fields.

If this is right

Reconstructions indicate higher early 20th-century global warming driven by stronger polar amplification than in existing reference products.
Broad modern Arctic warming trends are likely overestimated, yet intense localized hotspots appear in the Barents Sea and Northeastern Greenland.
Smoothing effects in widely used historical datasets, including those for IPCC assessments, are mitigated, improving representation of extremes.
The method supplies uncertainty-aware fields that support more robust evaluation of climate variability and change.
Probabilistic outputs enable better assessment of intrinsic variability without the underestimation common in deterministic interpolations.

Where Pith is reading between the lines

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

The same generative prior could be retrained on other variables such as sea level pressure or ocean salinity to extend consistent reconstructions.
If validated, the revised early warming rates would tighten constraints on climate sensitivity estimates derived from historical periods.
Ensemble reconstructions from this approach might improve detection and attribution studies by providing more realistic variability baselines.
Application to even sparser pre-1850 data could test whether the prior generalizes across longer timescales.

Load-bearing premise

The generative prior learned from modern dense observations accurately represents the dynamics governing the sparse historical data back to 1850 without systematic biases or unphysical artifacts.

What would settle it

Comparison of the reconstructed 1900-1980 Arctic temperature trends against independent high-resolution proxy records or reanalyses that either confirms the elevated rates of 0.15-0.29 °C per decade or shows substantially lower values.

read the original abstract

Accurate assessment of anthropogenic climate change relies on historical instrumental data, yet observations from the early 20th century are sparse, fragmented, and uncertain. Conventional reconstructions rely on disparate statistical interpolation, which tends to smooth local features and create unphysical artifacts, often leading to an underestimation of intrinsic variability and extremes. While recent machine learning approaches have improved reconstruction accuracy, they remain confined to purely spatial inpainting of coarse-resolution fields. Here, we present a unified, probabilistic generative deep learning framework that overcomes these limitations and reveals previously unresolved historical climate variability back to 1850. Leveraging a learned generative prior of Earth system dynamics, our model performs probabilistic inference to estimate spatiotemporally consistent historical temperature and precipitation fields from sparse observations. Our approach preserves the higher-order statistics of climate dynamics, transforming reconstruction into a robust uncertainty-aware assessment. We demonstrate that our reconstruction mitigates the smoothing effects inherent in widely used historical reference products, including those underlying IPCC assessments, especially regarding extreme weather events. Notably, we uncover higher early 20th-century global warming levels compared to existing reconstructions, primarily driven by more pronounced polar warming, with mean Arctic warming trends exceeding established benchmarks by 0.15--0.29C per decade for 1900--1980. Conversely, for the modern era, our reconstruction indicates that the broad Arctic warming trend is likely overestimated in recent assessments, yet explicitly resolves previously unrecognized intense, localized hotspots in the Barents Sea and Northeastern Greenland.

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.

Desk Editor's Note private letter to a colleague

The paper's generative model for spatiotemporal climate reconstruction claims stronger early Arctic warming than benchmarks, but the non-stationarity of the modern-trained prior is a load-bearing risk that needs direct checks.

read the letter

The core advance is a probabilistic generative deep learning framework that moves beyond spatial inpainting to full spatiotemporal inference on sparse 1850–present observations while trying to keep higher-order statistics. That is a clear step past the methods cited in the abstract, and if the implementation holds it could give uncertainty-aware fields that reduce the smoothing seen in standard products like those used in IPCC reports.

Referee Report

2 major / 1 minor

Summary. The paper introduces a unified probabilistic generative deep learning framework that learns a prior on Earth system dynamics from modern dense observations and uses it to perform spatiotemporal reconstruction of global temperature and precipitation fields from sparse instrumental data back to 1850. It claims to reduce smoothing artifacts relative to conventional statistical methods and IPCC reference products, preserve higher-order statistics, and yield new quantitative findings: higher early-20th-century global warming driven by stronger polar amplification (Arctic trends 0.15–0.29 °C/decade above benchmarks for 1900–1980) together with previously unrecognized localized hotspots in the modern Arctic (Barents Sea, NE Greenland) while suggesting that broad Arctic trends are overestimated in recent assessments.

Significance. If the central claims hold after rigorous validation, the work would represent a meaningful advance in historical climate reconstruction by supplying uncertainty-aware fields that better retain extremes and variability. This could affect assessments of polar amplification, early-century warming rates, and the fidelity of reference datasets used in IPCC reports.

major comments (2)

[Abstract] Abstract: the quantitative claim that mean Arctic warming trends exceed established benchmarks by 0.15–0.29 °C per decade for 1900–1980 is presented without error bars, explicit baseline definitions, stationarity diagnostics, or out-of-distribution validation metrics. Because this difference is the primary novel finding, the absence of these supporting details makes the result impossible to assess from the given text.
[Abstract] The generative prior is learned exclusively from modern, densely observed states yet is applied to the non-stationary 1850–1950 regime (different greenhouse-gas forcing, aerosol loading, and sea-ice conditions). No explicit test of prior generalization or regime-shift robustness is described; if the learned distribution does not capture these shifts, the reported polar amplification and localized hotspots could be systematic artifacts rather than data-driven inferences.

minor comments (1)

The abstract would benefit from a concise statement of the training data period, network architecture family, and the precise probabilistic inference procedure (e.g., how the posterior is sampled) to allow readers to gauge the method’s scope immediately.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed comments. We address each major point below and indicate the revisions that will be incorporated in the next version of the manuscript.

read point-by-point responses

Referee: [Abstract] Abstract: the quantitative claim that mean Arctic warming trends exceed established benchmarks by 0.15–0.29 °C per decade for 1900–1980 is presented without error bars, explicit baseline definitions, stationarity diagnostics, or out-of-distribution validation metrics. Because this difference is the primary novel finding, the absence of these supporting details makes the result impossible to assess from the given text.

Authors: We agree that the abstract requires additional supporting details for this key quantitative result. In the revised manuscript we will add error bars derived from the probabilistic ensemble, explicitly name the benchmark products (HadCRUT5 and the other IPCC reference reconstructions), and cross-reference the stationarity diagnostics and out-of-distribution validation metrics already reported in the supplementary information. These elements exist in the full text but were omitted from the abstract for brevity; we will restore them. revision: yes
Referee: [Abstract] The generative prior is learned exclusively from modern, densely observed states yet is applied to the non-stationary 1850–1950 regime (different greenhouse-gas forcing, aerosol loading, and sea-ice conditions). No explicit test of prior generalization or regime-shift robustness is described; if the learned distribution does not capture these shifts, the reported polar amplification and localized hotspots could be systematic artifacts rather than data-driven inferences.

Authors: We acknowledge the importance of demonstrating that the learned prior generalizes across the 1850–1950 regime shift. The current manuscript contains cross-validation on held-out modern periods and synthetic experiments that emulate historical sparsity and altered forcing; however, we did not include an explicit regime-shift test in the abstract or main text. In the revision we will add a dedicated subsection describing additional robustness checks (sensitivity to prescribed forcing changes and comparison against independent early-20th-century proxy constraints) and will update the abstract to reference these tests. We therefore treat this as a partial revision. revision: partial

Circularity Check

0 steps flagged

No circularity: generative prior applied to historical data is independent of target reconstruction

full rationale

The provided abstract and description contain no equations, self-citations, or derivation steps that reduce the historical temperature/precipitation fields to a direct fit, self-definition, or renaming of the modern training data. The model learns a generative prior from modern observations and performs probabilistic inference on sparse 1850+ records; this is a standard transfer-learning setup whose outputs are not forced by construction to match any fitted parameter. The central claims (higher early-20th-century Arctic trends, localized hotspots) are presented as empirical results of the inference, not tautological re-expressions of inputs. Non-stationarity concerns affect generalization validity but do not constitute circularity under the enumerated patterns.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim depends on the domain assumption that a generative model trained on modern climate data generalizes to historical periods, plus numerous free parameters in the neural network architecture and training process.

free parameters (1)

neural network hyperparameters and training parameters
Deep generative models require extensive hyperparameter choices and fitting to modern data to learn the prior.

axioms (1)

domain assumption A generative prior learned from modern Earth system data accurately captures historical climate dynamics for probabilistic inference from sparse observations.
Invoked to justify applying the trained model to 1850 onward data without bias.

pith-pipeline@v0.9.0 · 5589 in / 1438 out tokens · 33893 ms · 2026-05-15T21:12:01.961376+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

66 extracted references · 66 canonical work pages · 2 internal anchors

[1]

Von Storch, H.et al.Reconstructing past climate from noisy data.Science306, 679–682 (2004)

work page 2004
[2]

J.et al.inLinking Global to Regional Climate Change(eds Masson-Delmotte, V.et al.)Climate Change 2021: The Physical Science Basis

Doblas-Reyes, F. J.et al.inLinking Global to Regional Climate Change(eds Masson-Delmotte, V.et al.)Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change1363–1512 (Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 2021)

work page 2021
[3]

J., Jones, P

Harris, I., Osborn, T. J., Jones, P. & Lister, D. Version 4 of the CRU TS monthly high-resolution gridded multivariate climate dataset.Sci. Data7, 109 (2020)

work page 2020
[4]

P.et al.An updated assessment of near-surface temperature change from 1850: The HadCRUT5 data set.J

Morice, C. P.et al.An updated assessment of near-surface temperature change from 1850: The HadCRUT5 data set.J. Geophys. Res.: Atmos.126, e2019JD032361 (2021)

work page 2021
[5]

Yamamoto, J. K. Correcting the smoothing effect of ordinary kriging estimates.Math. Geol.37, 69–94 (2005). 20

work page 2005
[6]

Kadow, C., Hall, D. M. & Ulbrich, U. Artificial intelligence reconstructs missing climate information.Nat. Geosci.13, 408–413 (2020)

work page 2020
[7]

& Boers, N

Bochow, N., Poltronieri, A., Rypdal, M. & Boers, N. Reconstructing historical climate fields with deep learning.Sci. Adv.11, eadp0558 (2025)

work page 2025
[8]

B.et al.Past warming trend constrains future warming in CMIP6 models.Sci

Tokarska, K. B.et al.Past warming trend constrains future warming in CMIP6 models.Sci. Adv.6, eaaz9549 (2020)

work page 2020
[9]

Gulev, S. K.et al. Changing State of the Climate System. In Climate Change 2021: The Phys- ical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change(Cambridge University Press, 2021)

work page 2021
[10]

Bojinski, S.et al.The concept of essential climate variables in support of climate research, applications, and policy.Bull. Am. Meteorol. Soc.95, 1431–1443 (2014)

work page 2014
[11]

F.et al.The version-2 global precipitation climatology project (GPCP) monthly precipitation analysis (1979–present).J

Adler, R. F.et al.The version-2 global precipitation climatology project (GPCP) monthly precipitation analysis (1979–present).J. Hydrometeorol.4, 1147–1167 (2003)

work page 1979
[12]

Kent, E. C. & Kennedy, J. J. Historical estimates of surface marine temperatures.Annu. Rev. Mar. Sci.13, 283–311 (2021)

work page 2021
[13]

Allen, M. R. & Ingram, W. J. Constraints on future changes in climate and the hydrologic cycle. Nature419, 224–232 (2002)

work page 2002
[14]

Hansen, J., Ruedy, R., Sato, M. & Lo, K. Global surface temperature change.Rev. Geophys.48 (2010)

work page 2010
[15]

& Way, R

Cowtan, K. & Way, R. G. Coverage bias in the hadcrut4 temperature series and its impact on recent temperature trends.Q. J. R. Meteorolog. Soc.140, 1935–1944 (2014)

work page 1935
[16]

B.et al.Globally resolved surface temperatures since the Last Glacial Maximum

Osman, M. B.et al.Globally resolved surface temperatures since the Last Glacial Maximum. Nature599, 239–244 (2021)

work page 2021
[17]

J., Wang, J

Neukom, R., Steiger, N., G´ omez-Navarro, J. J., Wang, J. & Werner, J. P. No evidence for globally coherent warm and cold periods over the preindustrial Common Era.Nature571, 550–554 (2019)

work page 2019
[18]

& Evensen, G

Carrassi, A., Bocquet, M., Bertino, L. & Evensen, G. Data assimilation in the geosciences: An overview of methods, issues, and perspectives.Wiley Interdiscip. Rev. Clim. Change9, e535 (2018)

work page 2018
[19]

Geostat.: An Overview1, 1–7 (2013)

Rohde, R.et al.A new estimate of the average Earth surface land temperature spanning 1753 to 2011.Geoinfor. Geostat.: An Overview1, 1–7 (2013)

work page 2011
[20]

& Briffa, K

Jones, P., Osborn, T. & Briffa, K. Estimating sampling errors in large-scale temperature averages. J. Clim.10, 2548–2568 (1997)

work page 1997
[21]

& Jones, P

Hofstra, N., Haylock, M., New, M. & Jones, P. D. Testing E-OBS European high-resolution gridded data set of daily precipitation and surface temperature.J. Geophys. Res. Atmos.114 (2009)

work page 2009
[22]

Bracco, A.et al.Machine learning for the physics of climate.Nat. Rev. Phys.7, 6–20 (2025)

work page 2025
[23]

Scheffer, M.et al.Early-warning signals for critical transitions.Nature461, 53–59 (2009)

work page 2009
[24]

& Stocker, T

Boers, N., Ghil, M. & Stocker, T. F. Theoretical and paleoclimatic evidence for abrupt transitions in the Earth system.Environ. Res. Lett.17, 093006 (2022)

work page 2022
[25]

Geosci.1–12 (2025)

Boers, N.et al.Destabilization of Earth system tipping elements.Nat. Geosci.1–12 (2025). 21

work page 2025
[26]

Geosci.14, 550–558 (2021)

Brovkin, V.et al.Past abrupt changes, tipping points and cascading impacts in the Earth system.Nat. Geosci.14, 550–558 (2021)

work page 2021
[27]

& Laepple, T

Rietkerk, M., Skiba, V., Weinans, E., H´ ebert, R. & Laepple, T. Ambiguity of early warning signals for climate tipping points.Nat. Clim. Change1–10 (2025)

work page 2025
[28]

Liu, T.et al.Data gaps and outliers distort critical slowing down-based resilience indicators. Sci. Adv. (accepted)(2026)

work page 2026
[29]

Hess, P., Dr¨ uke, M., Petri, S., Strnad, F. M. & Boers, N. Physically constrained generative adversarial networks for improving precipitation fields from Earth system models.Nat. Mach. Intell.4, 828–839 (2022)

work page 2022
[30]

(ed.IEEE)Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 11461–11471 (IEEE, 2022)

Lugmayr, A.et al.IEEE (ed.)Repaint: Inpainting using denoising diffusion probabilistic models. (ed.IEEE)Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 11461–11471 (IEEE, 2022)

work page 2022
[31]

Price, I.et al.Probabilistic weather forecasting with machine learning.Nature637, 84–90 (2025)

work page 2025
[32]

& Boers, N

Hess, P., Aich, M., Pan, B. & Boers, N. Fast, scale-adaptive and uncertainty-aware downscaling of Earth system model fields with generative machine learning.Nat. Mach. Intell.1–11 (2025)

work page 2025
[33]

& Boers, N

Aich, M., Bathiany, S., Hess, P., Huang, Y. & Boers, N. Diffusion models for probabilistic precipitation generation from atmospheric variables.arXiv preprint arXiv:2504.00307(2025)

work page arXiv 2025
[34]

Neural Inf

Ho, J.et al.Video diffusion models.Adv. Neural Inf. Process. Syst.35, 8633–8646 (2022)

work page 2022
[35]

& Wood, F

Harvey, W., Naderiparizi, S., Masrani, V., Weilbach, C. & Wood, F. Flexible diffusion modeling of long videos.Adv. Neural Inf. Process. Syst.35, 27953–27965 (2022)

work page 2022
[36]

Li, Z.et al.Learning spatiotemporal dynamics with a pretrained generative model.Nat. Mach. Intell.6, 1566–1579 (2024)

work page 2024
[37]

Neural Inf

Srivastava, P.et al.Precipitation downscaling with spatiotemporal video diffusion.Adv. Neural Inf. Process. Syst.37, 56374–56400 (2024)

work page 2024
[38]

Yang, S.et al.Generative assimilation and prediction for weather and climate.arXiv preprint arXiv:2503.03038(2025)

work page arXiv 2025
[39]

Song, Y.et al.Score-based generative modeling through stochastic differential equations.arXiv preprint arXiv:2011.13456(2020)

work page internal anchor Pith review Pith/arXiv arXiv 2011
[40]

Diffusion Posterior Sampling for General Noisy Inverse Problems

Chung, H., Kim, J., Mccann, M. T., Klasky, M. L. & Ye, J. C. Diffusion posterior sampling for general noisy inverse problems.arXiv preprint arXiv:2209.14687(2022)

work page internal anchor Pith review Pith/arXiv arXiv 2022
[41]

Bi, K.et al.Accurate medium-range global weather forecasting with 3D neural networks.Nature 619, 533–538 (2023)

work page 2023
[42]

Lam, R.et al.Learning skillful medium-range global weather forecasting.Science382, 1416–1421 (2023)

work page 2023
[43]

Kochkov, D.et al.Neural general circulation models for weather and climate.Nature632, 1060–1066 (2024)

work page 2024
[44]

Bodnar, C.et al.A foundation model for the Earth system.Nature641, 1180–1187 (2025)

work page 2025
[45]

Allen, A.et al.End-to-end data-driven weather prediction.Nature641, 1172–1179 (2025)

work page 2025
[46]

& Ziese, M

Schneider, U., H¨ ansel, S., Finger, P., Rustemeier, E. & Ziese, M. GPCC Full Data Monthly Product Version 2022 at 0.5 ◦: Monthly Land-Surface Precipitation from Rain-Gauges built on GTS-based and Historical Data (2022). URL https://doi.org/10.5676/DWD GPCC/FD M 22 V2022 050. Global Precipitation Climatology Centre (GPCC)

work page doi:10.5676/dwd 2022
[47]

Nature566, 373–377 (2019)

Boers, N.et al.Complex networks reveal global pattern of extreme-rainfall teleconnections. Nature566, 373–377 (2019)

work page 2019
[48]

Suvorov, R.et al.Resolution-robust Large Mask Inpainting with Fourier Convolutions.arXiv preprint arXiv:2109.07161(2021)

work page arXiv 2021
[49]

D.et al.Causes of higher climate sensitivity in CMIP6 models.Geophys

Zelinka, M. D.et al.Causes of higher climate sensitivity in CMIP6 models.Geophys. Res. Lett. 47, e2019GL085782 (2020)

work page 2020
[50]

& Dong, X

Tian, B. & Dong, X. The double-ITCZ bias in CMIP3, CMIP5, and CMIP6 models based on annual mean precipitation.Geophys. Res. Lett.47, e2020GL087232 (2020)

work page 2020
[51]

& Kleinen, T

Held, H. & Kleinen, T. Detection of climate system bifurcations by degenerate fingerprinting. Geophys. Res. Lett.31(2004)

work page 2004
[52]

& Jayaprakash, C

Guttal, V. & Jayaprakash, C. Changing skewness: an early warning signal of regime shifts in ecosystems.Ecol. Lett.11, 450–460 (2008)

work page 2008
[53]

Dyn.14, 173–183 (2023)

Smith, T.et al.Reliability of resilience estimation based on multi-instrument time series.Earth Syst. Dyn.14, 173–183 (2023)

work page 2023
[54]

H., Neal, V

Quinn, W. H., Neal, V. T. & Antunez de Mayolo, S. E. El Ni˜ no occurrences over the past four and a half centuries.J. Geophys. Res.: Oceans92, 14449–14461 (1987)

work page 1987
[55]

& Kim, S

Yu, J.-Y. & Kim, S. T. Identifying the types of major El Ni˜ no events since 1870.Int. J. Climatol. 33, 2105–2112 (2012)

work page 2012
[56]

Cold & Warm Episodes by Season

Climate Prediction Center. Cold & Warm Episodes by Season. National Oceanic and Atmospheric Administration (n.d.). URL https://www.cpc.ncep.noaa.gov/products/analysis monitoring/ensostuff/ONI v5.php. Accessed: 2025-12-29

work page 2025
[57]

S.et al.The 1918/19 El Ni˜ no.Bull

Giese, B. S.et al.The 1918/19 El Ni˜ no.Bull. Am. Meteorol. Soc.91, 177–183 (2010)

work page 1918
[58]

Rantanen, M.et al.The Arctic has warmed nearly four times faster than the globe since 1979. Commun. Earth Environ.3, 168 (2022)

work page 1979
[59]

Held, I. M. & Soden, B. J. Robust responses of the hydrological cycle to global warming.J. Clim.19, 5686–5699 (2006)

work page 2006
[60]

P.et al.Advances in understanding large-scale responses of the water cycle to climate change.Ann

Allan, R. P.et al.Advances in understanding large-scale responses of the water cycle to climate change.Ann. N. Y. Acad. Sci.1472, 49–75 (2020)

work page 2020
[61]

D.et al.Consistency of modelled and observed temperature trends in the tropical troposphere.Int

Santer, B. D.et al.Consistency of modelled and observed temperature trends in the tropical troposphere.Int. J. Climatol.28, 1703–1722 (2008)

work page 2008
[62]

The stippling shows statistically significant grid points

Wilks, D. “The stippling shows statistically significant grid points”: How research results are routinely overstated and overinterpreted, and what to do about it.Bull. Am. Meteorol. Soc.97, 2263–2273 (2016)

work page 2016
[63]

& Abbeel, P

Ho, J., Jain, A. & Abbeel, P. Denoising diffusion probabilistic models.Adv. Neural Inf. Process. Syst.33, 6840–6851 (2020)

work page 2020
[64]

& Dayan, P

Hyv¨ arinen, A. & Dayan, P. Estimation of non-normalized statistical models by score matching. J. Mach. Learn. Res.6(2005)

work page 2005
[65]

& Laine, S

Karras, T., Aittala, M., Aila, T. & Laine, S. Elucidating the design space of diffusion-based generative models.Adv. Neural Inf. Process. Syst.35, 26565–26577 (2022)

work page 2022
[66]

Generative deep learning improves reconstruction of global historical climate records

Efron, B. Tweedie’s formula and selection bias.J. Am. Stat. Assoc.106, 1602–1614 (2011). 23 Supplementary Information for “Generative deep learning improves reconstruction of global historical climate records” Zhen Qian 1,2, Teng Liu 1,2,3,∗, Sebastian Bathiany 1,2, Shangshang Yang 1,2, Philipp Hess1,2, Nils Bochow 2,4, Christian Burmester 1, Maximilian G...

work page arXiv 2011