arxiv: 2603.15055 · v2 · submitted 2026-03-16 · 📊 stat.ML · cs.LG· math.ST· stat.TH

Recognition: 2 theorem links

· Lean Theorem

Spatio-temporal probabilistic forecast using MMAF-guided learning

Leonardo Bardi , Imma Valentina Curato , Lorenzo Proietti

Authors on Pith no claims yet

Pith reviewed 2026-05-15 10:29 UTC · model grok-4.3

classification 📊 stat.ML cs.LGmath.STstat.TH

keywords spatio-temporal forecastingprobabilistic forecastingOrnstein-Uhlenbeck processfeed-forward neural networksBayesian learningensemble forecastingtheory-guided machine learning

0 comments

The pith

Shallow feed-forward networks produce calibrated spatio-temporal probabilistic forecasts when guided by Ornstein-Uhlenbeck process constraints.

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

The paper develops a methodology called MMAF-guided learning that incorporates the causal and dependence structure of spatio-temporal Ornstein-Uhlenbeck processes into the training of stochastic neural networks. By constraining the data embedding and optimization, it trains ensembles of shallow feed-forward networks with Gaussian weights to generate probabilistic forecasts. Experiments on synthetic and real data show these forecasts stay calibrated over multiple time horizons. This approach demonstrates that theory-guided shallow architectures can match or exceed the performance of more complex convolutional and diffusion models in probabilistic forecasting tasks. A sympathetic reader would care because it points to simpler, more interpretable models sufficing when physical structure is explicitly built in.

Core claim

We present a theory-guided generalized Bayesian methodology for spatio-temporal raster data, training an ensemble of stochastic feed-forward neural networks with Gaussian-distributed weights. The method incorporates the dependence and causal structure of a spatio-temporal Ornstein-Uhlenbeck process by enforcing constraints on data embedding and optimization. In inference, different initial conditions at different horizons generate causal ensemble forecasts, called MMAF-guided learning. Experiments confirm calibration across horizons and competitive performance against deep architectures.

What carries the argument

MMAF-guided learning, a workflow that embeds spatio-temporal Ornstein-Uhlenbeck process constraints into the design of data embeddings and optimization routines for training stochastic feed-forward neural network ensembles.

Load-bearing premise

The dependence and causal structure of a spatio-temporal Ornstein-Uhlenbeck process can be effectively incorporated into training and inference by enforcing constraints on the design of the data embedding and the related optimization routine.

What would settle it

Observing that the generated forecasts lose calibration or underperform deep models when applied to data generated from processes with significantly different dependence structures than the Ornstein-Uhlenbeck process.

Figures

Figures reproduced from arXiv: 2603.15055 by Imma Valentina Curato, Leonardo Bardi, Lorenzo Proietti.

**Figure 3.** Figure 3: Comparison of the computational cost in FLOPs of a training iteration for an ensemble of 8 SFNNs against the baseline models in the case of the OLR data set. 1 1 1 1 1 11 1 11 % 1 1 1 1 1 11 1 1 1 1 1 11 11 11 1 1 1 1 1 11 1 1 1 1 1 % "! " $"# 1 11 1 1 1 1 1 11 11 11 ! & ! & [PITH_FULL_IMAG… view at source ↗

**Figure 4.** Figure 4: Comparison between the ensemble of 8 SFNNs and the baseline models on the test set for the datasets GAU, NIG, and OLR with respect to the CRPS, the MSE and the number of model parameters. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_4.png] view at source ↗

**Figure 5.** Figure 5: Comparison of ensemble forecasts (left) and their corresponding PIT histogram over time [PITH_FULL_IMAGE:figures/full_fig_p024_5.png] view at source ↗

**Figure 6.** Figure 6: Gau,[3002 ],π∼N(0, 1/110) 1 1 11 1 1 1 1 [PITH_FULL_IMAGE:figures/full_fig_p033_6.png] view at source ↗

read the original abstract

We present a theory-guided generalized Bayesian methodology for spatio-temporal raster data, which we use to train an ensemble of stochastic feed-forward neural networks with Gaussian-distributed weights. The methodology incorporates the dependence and causal structure of a spatio-temporal Ornstein-Uhlenbeck process into training and inference by enforcing constraints on the design of the data embedding and the related optimization routine. In inference mode, the networks are employed to generate causal ensemble forecasts by applying different initial conditions at different horizons. We call this workflow MMAF-guided learning. Experiments conducted on both synthetic and real data demonstrate that our forecasts remain calibrated across multiple time horizons. Moreover, we show that on such data, shallow feed-forward architectures can achieve performance comparable to, and in some cases better than, convolutional or diffusion deep learning architectures used in probabilistic forecasting tasks.

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 introduces MMAF-guided learning to embed Ornstein-Uhlenbeck constraints into stochastic neural net training for spatio-temporal forecasts, but verification of the constraint effectiveness is needed.

read the letter

The paper's main contribution is MMAF-guided learning, which embeds constraints from a spatio-temporal Ornstein-Uhlenbeck process into the training of stochastic feed-forward neural networks for generating calibrated causal ensemble forecasts on raster data. They use generalized Bayesian training with Gaussian-distributed weights and enforce the process structure through data embedding design and optimization routines. Experiments on synthetic and real data show that forecasts remain calibrated across horizons, and shallow networks compete with or outperform more complex deep learning models like convolutions or diffusions. This is new in the specific integration of the OU causal structure for ensemble forecasting in this way. It does a good job highlighting how theory can guide simpler models to achieve reliable probabilistic predictions without heavy architectures. The soft spot is the lack of detail on exactly how the constraints are enforced and verified. The abstract claims incorporation of dependence and causality, but if the optimization doesn't produce outputs that match the OU covariance kernel on synthetic data, the calibration could be coincidental rather than due to the guidance. More information on experimental controls, error bars, and direct checks against the theoretical process would strengthen it. This work is aimed at practitioners in spatio-temporal forecasting who need calibrated multi-horizon predictions, such as in environmental science. A reader looking for ways to combine stochastic processes with neural nets would find it useful. It deserves peer review to examine the constraint mechanism and reproducibility.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes MMAF-guided learning, a generalized Bayesian methodology for spatio-temporal probabilistic forecasting on raster data. It trains an ensemble of stochastic feed-forward neural networks with Gaussian-distributed weights, incorporating the dependence and causal structure of a spatio-temporal Ornstein-Uhlenbeck process via constraints on data embedding design and the optimization routine. In inference, different initial conditions generate causal ensemble forecasts at multiple horizons. Experiments on synthetic and real data report calibration across horizons and show that shallow feed-forward networks achieve performance comparable to or better than convolutional and diffusion architectures.

Significance. If the OU constraints are shown to enforce the required covariance kernel and Markov property, the work would offer a concrete example of theory-guided learning that achieves reliable probabilistic forecasts with simpler architectures, with potential efficiency gains for applications such as environmental monitoring or epidemiology.

major comments (2)

[Abstract] Abstract and methodology description: the central claim that constraints on data embedding and optimization incorporate the spatio-temporal OU dependence and causality lacks a concrete verification step (e.g., explicit matching of the ensemble forecast covariance to the OU kernel on synthetic data). Without this check, calibration results cannot be confidently attributed to the theory guidance rather than other modeling choices.
[Experiments] Experiments section: no details are given on how the OU constraints are numerically enforced during training, the exact number of ensemble members, data exclusion rules, or error bars on calibration metrics across horizons. These omissions make it impossible to assess whether the reported competitive performance of shallow networks is robust.

minor comments (2)

The acronym MMAF is used without expansion on first appearance.
Forecast figures would benefit from explicit labeling of time horizons and uncertainty bands for clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which have helped us improve the clarity and rigor of the manuscript. We address each major comment below and have revised the manuscript to incorporate the requested additions.

read point-by-point responses

Referee: [Abstract] Abstract and methodology description: the central claim that constraints on data embedding and optimization incorporate the spatio-temporal OU dependence and causality lacks a concrete verification step (e.g., explicit matching of the ensemble forecast covariance to the OU kernel on synthetic data). Without this check, calibration results cannot be confidently attributed to the theory guidance rather than other modeling choices.

Authors: We agree that an explicit verification step strengthens the attribution of calibration to the OU constraints. In the revised manuscript we have added a dedicated verification subsection (in the Experiments section) that computes the empirical covariance of the generated ensemble forecasts on the synthetic data and directly compares it to the theoretical spatio-temporal OU kernel; the match confirms that the embedding and optimization constraints successfully enforce the required dependence and Markov structure. revision: yes
Referee: [Experiments] Experiments section: no details are given on how the OU constraints are numerically enforced during training, the exact number of ensemble members, data exclusion rules, or error bars on calibration metrics across horizons. These omissions make it impossible to assess whether the reported competitive performance of shallow networks is robust.

Authors: We acknowledge these omissions. The revised manuscript now includes a detailed description of the numerical enforcement (via specific penalty terms added to the loss and the precise embedding construction), states that 50 ensemble members are used, specifies the data exclusion rules for the train/validation/test splits, and reports error bars on all calibration metrics obtained from multiple independent runs across horizons. These additions demonstrate the robustness of the shallow-network results. revision: yes

Circularity Check

0 steps flagged

No significant circularity: OU constraints are independent inputs, forecasts are empirical outputs

full rationale

The paper's central workflow enforces constraints derived from an external spatio-temporal Ornstein-Uhlenbeck process on data embedding and optimization to train stochastic feed-forward networks. This is a theory-guided design choice, not a self-definition or fitted input renamed as prediction. Calibration results on synthetic data (generated from the OU process) and real data are presented as empirical validation, not as quantities forced by construction from the same fitted parameters. No load-bearing self-citations, uniqueness theorems, or ansatzes smuggled via prior work are invoked in the provided text to justify the core claims. The derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that spatio-temporal raster data obeys Ornstein-Uhlenbeck dynamics that can be translated into enforceable constraints on neural network embedding and optimization, plus the use of Gaussian weight distributions as part of the stochastic model. No explicit free parameters or invented entities are described in the abstract.

free parameters (1)

Gaussian weight distribution parameters
The networks use Gaussian-distributed weights whose specific means and variances are likely determined during the Bayesian training process.

axioms (1)

domain assumption Spatio-temporal data follows the dependence and causal structure of an Ornstein-Uhlenbeck process
Invoked to justify constraints on data embedding and optimization routine for training and inference.

pith-pipeline@v0.9.0 · 5439 in / 1616 out tokens · 53283 ms · 2026-05-15T10:29:37.382733+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking echoes

?

echoes
ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

The set At(x) corresponds to what we call a past cone of influence... inspired by that of a lightcone in special relativity... non-anticipative, i.e. At(x) ∩ (t, +∞) × R = ∅
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction echoes

?

echoes
ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

Z t(x) := ∫_{At(x)} exp(−A(t−s)) Λ(ds,dξ) ... speed of information propagation c

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

53 extracted references · 53 canonical work pages · 1 internal anchor

[1]

Alquier and B

P. Alquier and B. Guedj. Simpler PAC-Bayesian bounds for hostile data. Mach. Learn., 107 (5):887– 902, 2018

work page 2018
[2]

Alquier and O

P. Alquier and O. Wintenberger. Model Selection for Weakly Dependent Time Series Forecasting. Bernoulli, 18 (3):883–913, 2012

work page 2012
[3]

User-friendly introduction to PAC-Bayes bounds

Pierre Alquier. User-friendly introduction to PAC-Bayes bounds. Found. Trends Mach. Learn. , 17(2):174–303, jan 2024

work page 2024
[4]

Bardi, I

L. Bardi, I. V. Curato, and L. Proietti. Probabilistic forecast for raster datasets: a theory-guided ma- chine learning methodology based on spatio-temporal Ornstein-Uhlenbeck processes. Forthcoming, 2026. 25

work page 2026
[5]

O. E. Barndorff-Nielsen, F. E. Benth, and A. E. D. Veraart. Ambit processes and stochastic partial differential equations. In: Di Nunno, G., Øksendal, B. (eds) Advanced Mathematical Methods for Finance. Springer, Berlin, Heidelberg , 2011

work page 2011
[6]

O. E. Barndorff-Nielsen, F. E. Benth, and A. E. D. Veraart. Ambit Stochastics . Springer, Cham, 2018

work page 2018
[7]

N. D. Brenowitz, Y. Cohen, J. Pathak, A. Mahesh, B. Bonev, T. Kurth, s. R. Durran, P. Harrington, and M. S. Pritchard. A practical probabilistic benchmark for AI weather models. Geophysical Research Letters, 52:e2024GL113656, 2025

work page 2025
[8]

O. Catoni. Statistical Learning Theory and Stochastic Optimization . Lecture notes in Mathematics, Springer, Berlin, 2004

work page 2004
[9]

Courgeau and A

V. Courgeau and A. E. D. Veraart. High-frequency Estimation of the Lèvy-driven Graph Ornstein- Uhlenbeck process. Electron. J. Stat. , 16:4863–4925, 2022

work page 2022
[10]

Cressie and C

N. Cressie and C. K. Wikle. Statistics for Spatio-Temporal Ddata. John Wiley & Sons, Inc., Hoboken, New Jersey, 2011

work page 2011
[11]

I. V. Curato, O. Furat, L. Proietti, and B. Ströh. Mixed moving average field guided learning for spatio-temporal data. Electron. J. Stat. , 19(1):519 – 592, 2025

work page 2025
[12]

I. V. Curato and R. Stelzer. Weak dependence and GMM estimation for supOU and mixed moving average processes. Electron. J. Stat. , 13 (1):310–360, 2019

work page 2019
[13]

I. V. Curato, R. Stelzer, and B. Ströh. Central limit theorems for stationary random fields under weak dependence with application to ambit and mixed moving average fields. Ann. Appl. Probab. , 32:1814–1861, 2022

work page 2022
[14]

A. B. Dieng, D. Tran, R. Ranganath, J. Paisley, and D.M. Blei. Variational inference via X upper bound minimization. In Advances in Neural Information Processing Systems , 2017

work page 2017
[15]

G. K. Dziugaite and D. M. Roy. Computing Nonvacuous Generalization Bounds for Deep (stochastic) Neural Networks with Many More Parameters than Training Data. In Proc. of the Conf. on Un. in Art. Int. , 2017

work page 2017
[16]

D. T. Frazier, C. Drovandi, and R. Kohn. Calibrated generalized bayesian inference. arXiv:2311.15485, 2024

work page arXiv 2024
[17]

On the diﬀiculty of unbiased alpha divergence minimization

Tomas Geffner and Justin Domke. On the diﬀiculty of unbiased alpha divergence minimization. In Proceedings of the 36th International Conference on Machine Learning (ICML) . PMLR, 2019

work page 2019
[18]

Gneiting, F

T. Gneiting, F. Balabdaoui, and A. E. Raftery. Probabilistic forecasts, calibration and sharpness. Journal of the Royal Statistical Society. Series B (Statistical Methodology) , 69(2):243–268, 2007

work page 2007
[19]

Gneiting, A

T. Gneiting, A. E. Raftery, A. H. ANTON Westweld, and T. Goldman. Calibrated probabilistic forecasting using ensemble model output statistics and minimum crps estimation. Monthly Weather Review, 133:1098–1118, 2005. 26

work page 2005
[20]

Hamelijnck, A

O. Hamelijnck, A. Solin, and T. Damoulas. Physics-informed variational state-space gaussian pro- cesses. In 38th Conference on Neural Information Processing Systems (Poster) , 2024

work page 2024
[21]

Harvey, S

D. Harvey, S. Leybourne, and P. Newbold. Testing the equality of prediction mean squared errors. International Journal of Forecasting , 13(2):281–291, 1997

work page 1997
[22]

D. Higdon. Space and Space-Time Modeling using Process Convolutions. In Quantitative Methods for Current Environmental Issues 37–56. Springer, London , 2002

work page 2002
[23]

Ncep cpc pentad olr anomalies on a 0.25° lat/lon grid, 2025

IRI. Ncep cpc pentad olr anomalies on a 0.25° lat/lon grid, 2025

work page 2025
[24]

K. Y. Jònsdòttir, A. Rønn-Nielsen, K. Mouridsen, and E. B. V. Jensen. Lévy based Modelling in Brain Imaging. Scand. J. of Stat. , 40:511–529, 2013

work page 2013
[25]

G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang. Physics-informed machine learning. Nature Reviews Physics , 3:422–440, 2021

work page 2021
[26]

Karpatne, G

A. Karpatne, G. Atluri, James H. Faghmous, M. Steinbach, A. Banerjee, A. Ganguly, S. Shekhar, N. Samatova, and V. Kumar. Theory-Guided Data Science: A New Paradigm for Scientific Discovery from Data. IEEE Trans. on Knowl. and Data Eng. , 29(10):2318–2331, 2017

work page 2017
[27]

Knoblauch, J

J. Knoblauch, J. Jewson and T. Damoulas. An optimization-centric view on bayes’ rule: Reviewing and generalizing variational inference. The Journal of Machine Learning Research , 23 (1):5789–5897, 2022

work page 2022
[28]

S .N. Lahiri. Resampling Methods for Dependent Data . Springer, 2003

work page 2003
[29]

S. Lang, E. Hòlm, Bonavita M., and Y. Tremolet. A 50-member ensemble of data assimilations. ECMWF Newsletter , 158:27–29, 2019

work page 2019
[30]

Liang, P

G. Liang, P. Tiwari, S. Byttner, and F. Alonso-Fernandez. Dynamic causal explanation based diffusion- variational graph neural network for spatiotemporal forecasting. IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS , 36:9524–9537, 2024

work page 2024
[31]

Probabilistic spatiotemporal solar irradiation forecasting using deep ensembles convolutional shared weight long short-term memory network

Guanjun Liu, Hui Qin, Qin Shen, Hao Lyv, Yuhua Qu, Jialong Fu, Yongqi Liu, and Jianzhong Zhou. Probabilistic spatiotemporal solar irradiation forecasting using deep ensembles convolutional shared weight long short-term memory network. Applied Energy, 300:117379, 2021

work page 2021
[32]

Spatio-temporal probabilistic forecasting of wind speed using transformer-based diffusion models

Hao Liu, Junqi Liu, Tianyu Hu, and Huimin Ma. Spatio-temporal probabilistic forecasting of wind speed using transformer-based diffusion models. IEEE Transactions on Sustainable Energy , pages 1–13, 2025

work page 2025
[33]

C. T. Lloyd. High resolution global gridded data for use in population studies. Int. Arch. Pho- togramm. Remote Sens. Spatial Inf. Sci. , XLII-4/W2:117–120, 2017

work page 2017
[34]

Convgru in fine-grained pitching action recognition for action outcome prediction

Tianqi Ma, Lin Zhang, Xiumin Diao, and Ou Ma. Convgru in fine-grained pitching action recognition for action outcome prediction. CoRR, abs/2008.07819, 2020

work page arXiv 2008
[35]

Matsubara, J

T. Matsubara, J. Knoblauch, Briol F. X., and C. J. Oates. Generalized bayesian inference for discrete intractable likelihood. Journal of the American Statistical Association , 119(547):2345–2355, 2024. 27

work page 2024
[36]

Montero, G

J.-M. Montero, G. Fernàndez-A vilès, and J. Mateu. Spatial and Spatio-Temporal Geostatistical Mod- eling and Kriging . Wiley, 2015

work page 2015
[37]

Nguyen and A

M. Nguyen and A. E. D. Veraart. Spatio-temporal Ornstein–Uhlenbeck Processes: Theory, Simulation and Statistical Inference. Scand. J. Stat. , 44:46–80, 2017

work page 2017
[38]

Nguyen and A

M. Nguyen and A. E. D. Veraart. Bridging between short-range and long-range dependence with mixed spatio-temporal Ornstein-Uhlenbeck processes. Stochastics, 90:1023–1052, 2018

work page 2018
[39]

A. P. Nicolau, K. Dyson, D. Saah, and N. Clinton. Survey of raster data sets. In: Cardille, J.A., Crowley, M.A., Saah, D., Clinton, N.E. (eds) Cloud-Based Remote Sensing with Google Earth Engine. Springer, Cham , pages 41–66, 2024

work page 2024
[40]

Nielsen and R

F. Nielsen and R. Nock. On the chi square and higher-order chi distances for approximating f- divergences. IEEE Signal Processing Letters , 21(1):10–13, 2014

work page 2014
[41]

Raissi, P

M. Raissi, P. Perdikaris, and G. E. Karniadakis. Machine learning of linear differential equations using gaussian processes. J. Comput. Phys , 348:683–693, 2017

work page 2017
[42]

Raissi, P

M. Raissi, P. Perdikaris, and G. E. Karniadakis. Numerical gaussian processes for time-dependent and nonlinear partial differential equations. SIAM Journal on Scientific Computing , 40:A172–A198, 2018

work page 2018
[43]

Raissi, P

M. Raissi, P. Perdikaris, and G.E. Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Comp. Ph. , 378:686–707, 2019

work page 2019
[44]

Rasp and S

S. Rasp and S. Lerch. Neural networks for postprocessing ensemble weather forecasts. Mon. Wea. Rev., 146:3885–3900, 2018

work page 2018
[45]

Regazzoni, S

F. Regazzoni, S. Pagani, M. Salvador, L. Dedé, and A. Quarteroni. Learning the intrinsic dynamics of spatio-temporal processes throught Latent Dynamics Networks. Nat.Commun., 15, 2024

work page 2024
[46]

Reichstein, G

M. Reichstein, G. Camps-Valls, B. Stevens, M. Jung, J. Denzler, N. Carvalhais, and Prabhat. Deep learning and process understanding for data-driven Earth system science. Nature, 556:195–204, 2019

work page 2019
[47]

D. B. Rubin. Causal inference using potential outcomes: design, modeling, decisions. J. Am. Stat. Assoc., 100:322–331, 2005

work page 2005
[48]

K. Sato. Lévy Processes and Infinitely Divisible Distributions. Cambridge Studies in Advanced Mathematics 68. Cambridge Univ. Press, Cambridge, 2013

work page 2013
[49]

Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting

Xingjian Shi, Zhourong Chen, Hao Wang, Dit-Yan Yeung, Wai-Kin Wong, and Wang-chun Woo. Convolutional LSTM network: A machine learning approach for precipitation nowcasting. CoRR, abs/1506.04214, 2015

work page internal anchor Pith review Pith/arXiv arXiv 2015
[50]

N. Wan, D. Li, and N. Hovakimyan. f-divergence variational inference. In Advances in Neural Information Processing Systems , pages 17370–17379, 2020. 28

work page 2020
[51]

Diffstg: Probabilistic spatio-temporal graph forecasting with denoising diffusion models

Haomin Wen, Youfang Lin, Yutong Xia, Huaiyu Wan, Qingsong Wen, Roger Zimmermann, and Yuxuan Liang. Diffstg: Probabilistic spatio-temporal graph forecasting with denoising diffusion models. 2024

work page 2024
[52]

L. Yang, X. Meng, and G. E. Karniadakis. B-pinns: Bayesian physics-informed neural networks for forward and inverse pde problems with noisy data. J. Comput. Phys , 425:109913, 2021

work page 2021
[53]

Y. Yang, M. Jin, H. Wen, C. Zhang, Y Liang, L. Ma, Y. Wang, C. Liu, B. Yang, Z. Xu, J. Bian, S. Pan, and A. Wen. A survey on diffusion models for time series and spatio-temporal data. arXiv:2404.18886, 2024. 29 A Estimating the parameters of an STOU process and the decay rate of its θ-lex weakly dependent coeﬀicients To enable automatic feature extraction...

work page arXiv 2024