pith. sign in

arxiv: 2509.20846 · v3 · pith:7GLKV6HNnew · submitted 2025-09-25 · 💻 cs.LG

Causal Time Series Generation via Diffusion Models

Yutong Xia , Chang Xu , Yuxuan Liang , Li Zhao , Qingsong Wen , Roger Zimmermann , Jiang Bian This is my paper

Pith reviewed 2026-05-18 14:40 UTC · model grok-4.3

classification 💻 cs.LG
keywords causal time series generationdiffusion modelsbackdoor adjustmentinterventional generationcounterfactualsPearl's causal laddertime series synthesisscore function adjustment
0
0 comments X

The pith

A diffusion model framework derives causal score functions to generate time series under interventions and as individual counterfactuals.

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

This paper introduces causal time series generation as a distinct task family that extends standard synthesis to include sequences produced after interventions on covariates and sequences representing specific counterfactual outcomes for individuals. It formalizes the tasks inside Pearl's three-rung causal ladder, moving from pure observation to intervention and counterfactual reasoning. The authors build CaTSG as a single diffusion model whose sampling is steered by score functions adjusted via backdoor criteria and the abduction-action-prediction steps. A reader would care because ordinary conditional generators only reproduce correlations and break down once the underlying system is changed or when asking what would have happened under alternate histories. Experiments on synthetic and real data indicate that the causal adjustments preserve fidelity to observed sequences while adding the new interventional and counterfactual capabilities.

Core claim

By applying backdoor adjustment to the diffusion score function for interventions and the full abduction-action-prediction procedure for counterfactuals, CaTSG creates a unified sampling process that supports observational, interventional, and counterfactual time series generation within one model while keeping the generated observational distribution faithful to the training data.

What carries the argument

Backdoor-adjusted guidance applied to the diffusion model's score function, which modifies the reverse denoising trajectory to account for unobserved confounding and to target specific interventions or counterfactual worlds.

If this is right

  • A single trained diffusion model can now produce sequences that reflect the effect of an intervention on one or more covariates.
  • Individual counterfactual trajectories become generatable by first inferring the noise and then applying the appropriate action and prediction steps.
  • Observational fidelity metrics remain competitive with or better than non-causal baselines even after the causal adjustments are added.
  • The same architecture supplies all three levels of Pearl's causal ladder without separate models for each rung.

Where Pith is reading between the lines

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

  • Extension: The same score-adjustment idea could be tried on other sequential generative architectures such as autoregressive transformers to add causal control without diffusion-specific machinery.
  • Extension: In domains like policy evaluation, the framework supplies a way to simulate alternate histories for individual units rather than only population averages.
  • Extension: Combining the method with data-driven causal discovery algorithms might reduce the need for a pre-specified graph when applying the backdoor adjustment.

Load-bearing premise

Backdoor adjustment and the abduction-action-prediction steps can be transplanted directly onto the diffusion score function for time series without breaking temporal dependencies or requiring extra causal graph details.

What would settle it

On a synthetic time series dataset generated from a fully known causal graph, compare the marginal statistics of sequences produced by CaTSG under a specific intervention against the true post-intervention distribution obtained by do-calculus on the graph; large mismatch falsifies the claim that the adjusted scores correctly steer the sampler.

Figures

Figures reproduced from arXiv: 2509.20846 by Chang Xu, Jiang Bian, Li Zhao, Qingsong Wen, Roger Zimmermann, Yutong Xia, Yuxuan Liang.

Figure 1
Figure 1. Figure 1: Motivation for causal time series generation. (a) We observe time series [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Structural Causal Models (SCMs) for different conditional time series generation paradigms. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of CaTSG. (a) Generative objectives for observational, interventional, and counterfactual TSG. (b) Model instantiation: CaTSG consists of EnvInfer, Environment Bank, and Denoiser. Blue variables are used only for Lsw, Env. Bank is regularized by Lorth, and Denoiser is regularized by Leps. (c) Swapped prediction loss. Training Objective. We jointly optimize three components: Denoiser εθ, EnvInfer q… view at source ↗
Figure 4
Figure 4. Figure 4: (a) Factual samples X and (b) counterfactual gen￾erations Xˆ′ on Traffic. (c) Global distribution over all sam￾ples: histogram and ECDF of X (blue) and Xˆ′ (orange). For qualitative analysis on real-world data, we provide visualization results ( [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: (a) Environment posterior w across splits. (b) t-SNE of representations on Harmonic-VM. Shaded ellipses are one standard-deviation regions. (c) Ablation of components. (d) Hyperparameter sensitivity on Harmonic-VM. 5.5 ABLATION STUDY [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Three rules of Do-calculus. The do-calculus (Pearl et al., 2016) is a formal system that enables the identification of interventional distributions, such as P(Y | do(X)), from purely observational data, given a SCM. Its core goal is to transform expressions involving interventions (i.e., do(·)) into expressions involving only observed variables. Do-calculus consists of three transformation rules that opera… view at source ↗
Figure 7
Figure 7. Figure 7: Illustration of (a) original DAG G, (b) interventional graph Gdo(X) , and (c) nullified graph Gnull(X) Before introducing the rules, we first define the relevant graph transformations used to represent interventions and their structural consequences as shown in [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Architecture of EnvInfer qϕ. The module takes input sequences (x, c) (or their augmented versions (xv, cv)) together with the environment bank E, and pro￾duces latent representations h, sample weights w, and log￾its s. Concatenating all representations, we have h ′′ = [hstat, hatt, hspec] ∈ R N×(5H+Kp) . We then project this representation of features to latent and normaliza￾tion. A LayerNorm and MLP proje… view at source ↗
Figure 9
Figure 9. Figure 9: (a) A damped harmonic oscillator governed by [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Geographic distribution of stations by split in the Air Quality dataset. We further evaluate our framework on two real-word datasets, i.e., Air Quality and Traffic. Across datasets, we construct environment splits by using selected observable variables as proxies to delineate latent contexts (i.e., treated as unobserved environments). These variables are used only for splitting Train/Val/Test and and the … view at source ↗
Figure 11
Figure 11. Figure 11: Empirical diagnostics on synthetic datasets. (a,b) Harmonic-VM and (c,d) Harmonic-VP with [PITH_FULL_IMAGE:figures/full_fig_p027_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Empirical diagnostics on real-world datasets. (a) and (b) Air Quality with [PITH_FULL_IMAGE:figures/full_fig_p028_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Environment posterior and t-SNE of latent representations on (a) Harmonic-VM and (b) Air Quality dataset [PITH_FULL_IMAGE:figures/full_fig_p030_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Speed-quality trade-off on Harmonic-VM: runtime vs. steps ( [PITH_FULL_IMAGE:figures/full_fig_p031_14.png] view at source ↗
read the original abstract

Time series generation (TSG) synthesizes realistic sequences and has achieved remarkable success. Among TSG, conditional models generate sequences given observed covariates, however, such models learn observational correlations without considering unobserved confounding. In this work, we propose a causal perspective on conditional TSG and introduce causal time series generation as a new TSG task family, formalized within Pearl's causal ladder, extending beyond observational generation to include interventional and counterfactual settings. To instantiate these tasks, we develop CaTSG, a unified diffusion-based framework with backdoor-adjusted guidance that causally steers sampling toward desired interventions and individual counterfactuals while preserving observational fidelity. Specifically, our method derives causal score functions via backdoor adjustment and the abduction-action-prediction procedure, thus enabling principled support for all three levels of TSG. Extensive experiments on both synthetic and real-world datasets show that CaTSG achieves superior fidelity and also supporting interventional and counterfactual generation that existing baselines cannot handle. Overall, we propose the causal TSG family and instantiate it with CaTSG, providing an initial proof-of-concept and opening a promising direction toward more reliable simulation under interventions and counterfactual generation.

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

3 major / 2 minor

Summary. The manuscript introduces causal time series generation (CaTSG) as a new task family extending standard conditional time series generation to Pearl's three rungs of the causal ladder (observational, interventional, and counterfactual). It proposes a unified diffusion-based framework that derives causal score functions by substituting backdoor-adjusted conditionals and applying the abduction-action-prediction procedure, enabling steered sampling for interventions and individual counterfactuals while preserving observational fidelity. Experiments on synthetic and real-world datasets are reported to show superior fidelity and the ability to handle causal queries that baselines cannot.

Significance. If the central construction is valid, the work would be significant for generative modeling in domains requiring intervention-aware simulation such as healthcare and finance. The unified treatment of all three causal levels within a single diffusion framework, together with the explicit use of backdoor adjustment on the score function, represents a concrete step beyond purely observational TSG. The provision of both synthetic and real-world empirical support strengthens the initial proof-of-concept.

major comments (3)
  1. [§3] §3 (method derivation): the substitution of the backdoor-adjusted conditional directly into the diffusion score s_θ(x_t, t) is presented without an explicit time-unrolled causal graph or a proof that the resulting score equals the gradient of the log-density of the post-intervention marginal; temporal ordering of edges and potential additional conditioning paths introduced by the diffusion process are not addressed, which is load-bearing for the claim that interventional samples are correctly generated.
  2. [§4] §4 (abduction-action-prediction): the implementation of the abduction step for individual counterfactuals in the diffusion setting is described at a high level; it is unclear how the noise variables are recovered from observed trajectories while respecting the sequential nature of the data, and no verification is given that the subsequent action and prediction steps commute with the reverse diffusion process.
  3. [Table 2, §5.2] Table 2 and §5.2: the reported improvements in interventional and counterfactual metrics are shown without error bars or statistical significance tests across multiple random seeds; given that the central claim rests on the correctness of the causal steering, the absence of these controls weakens the empirical support for the method's reliability.
minor comments (2)
  1. Notation for the time-indexed variables and the backdoor set Z should be introduced earlier and used consistently; current presentation requires the reader to infer the precise conditioning from the text.
  2. The related-work section would benefit from explicit comparison to recent causal diffusion models (e.g., those using do-calculus on score functions) to better situate the novelty of the time-series extension.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and outline the revisions planned for the next version of the manuscript.

read point-by-point responses

  1. Referee: [§3] §3 (method derivation): the substitution of the backdoor-adjusted conditional directly into the diffusion score s_θ(x_t, t) is presented without an explicit time-unrolled causal graph or a proof that the resulting score equals the gradient of the log-density of the post-intervention marginal; temporal ordering of edges and potential additional conditioning paths introduced by the diffusion process are not addressed, which is load-bearing for the claim that interventional samples are correctly generated.

    Authors: We agree that §3 would benefit from greater formal detail. In the revision we will insert an explicit time-unrolled causal graph that depicts the diffusion process together with the structural causal model at each time step. We will also add a short proof sketch establishing that the backdoor-adjusted score is the gradient of the log-density of the post-intervention marginal, and we will discuss why the diffusion’s additional conditioning paths do not violate the adjustment when the backdoor set is chosen with respect to the original SCM. revision: yes

  2. Referee: [§4] §4 (abduction-action-prediction): the implementation of the abduction step for individual counterfactuals in the diffusion setting is described at a high level; it is unclear how the noise variables are recovered from observed trajectories while respecting the sequential nature of the data, and no verification is given that the subsequent action and prediction steps commute with the reverse diffusion process.

    Authors: We acknowledge that the current description of the abduction step is high-level. In the revised manuscript we will provide a concrete algorithm that recovers the per-step noise variables from an observed trajectory while respecting the autoregressive structure of the time series. We will also include a brief argument (or empirical check) showing that the action and prediction operations can be interleaved with the reverse diffusion steps without changing the final counterfactual distribution. revision: yes

  3. Referee: [Table 2, §5.2] Table 2 and §5.2: the reported improvements in interventional and counterfactual metrics are shown without error bars or statistical significance tests across multiple random seeds; given that the central claim rests on the correctness of the causal steering, the absence of these controls weakens the empirical support for the method's reliability.

    Authors: We thank the referee for this observation. In the revision we will rerun all experiments with at least five independent random seeds, report means and standard deviations (error bars) for every metric in Table 2, and add paired statistical significance tests (e.g., t-tests) between CaTSG and the strongest baseline for the interventional and counterfactual settings. revision: yes

Circularity Check

0 steps flagged

No circularity: applies external causal adjustment to diffusion scores

full rationale

The paper's central construction derives interventional and counterfactual score functions by substituting backdoor-adjusted conditionals and the abduction-action-prediction steps into the standard diffusion score s_θ(x_t, t). These steps are imported from Pearl's causal ladder and standard causal inference literature rather than being fitted to the paper's own data or derived from internal parameters. No self-citation chain, self-definitional loop, or fitted-input-renamed-as-prediction is present in the derivation; the observational fidelity term remains the original diffusion objective while the causal steering is an additive guidance term justified externally. The time-series extension is presented as an application, not a reduction to the inputs by construction. This is the normal case of a self-contained method that builds on independent external results.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard causal inference tools applied to diffusion models; no free parameters or new entities are introduced in the abstract description.

axioms (2)
  • domain assumption Backdoor adjustment can be applied to derive causal score functions for diffusion sampling in time series data.
    Invoked when describing how the method steers sampling toward interventions.
  • domain assumption The abduction-action-prediction procedure from Pearl's framework extends directly to individual counterfactual generation in sequential data.
    Used to support counterfactual tasks in the unified framework.

pith-pipeline@v0.9.0 · 5737 in / 1329 out tokens · 46924 ms · 2026-05-18T14:40:36.191413+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

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.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Causal Diffusion Models for Counterfactual Outcome Distributions in Longitudinal Data

    stat.ML 2026-04 unverdicted novelty 7.0

    Causal Diffusion Model is the first diffusion-based method to produce full probabilistic counterfactual outcome distributions for sequential interventions in longitudinal data, showing 15-30% better distributional acc...

  2. Intervention-Based Time Series Causal Discovery via Simulator-Generated Interventional Distributions

    cs.LG 2026-05 unverdicted novelty 6.0

    SVAR-FM uses simulator clamping to produce interventional distributions and flow matching to identify time series causal structures, with an error bound that predicts sign reversal of causal effects below a simulator ...

  3. Interventional Time Series Priors for Causal Foundation Models

    cs.LG 2026-03 unverdicted novelty 6.0

    CausalTimePrior generates synthetic temporal structural causal models with paired observational and interventional time series to train prior-data fitted networks for in-context causal effect estimation on held-out data.

Reference graph

Works this paper leans on

51 extracted references · 51 canonical work pages · cited by 3 Pith papers · 4 internal anchors

  1. [1]

    Diffusion-based time series imputation and forecasting with structured state space models

    Juan Lopez Alcaraz and Nils Strodthoff. Diffusion-based time series imputation and forecasting with structured state space models. Transactions on Machine Learning Research, 2023

    work page 2023

  2. [2]

    Tsgbench: Time series generation benchmark

    Yihao Ang, Qiang Huang, Yifan Bao, Anthony KH Tung, and Zhiyong Huang. Tsgbench: Time series generation benchmark. Proceedings of the VLDB Endowment, 17 0 (3): 0 305--318, 2023

    work page 2023

  3. [3]

    An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling

    Shaojie Bai, J Zico Kolter, and Vladlen Koltun. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  4. [4]

    Towards controllable time series generation

    Yifan Bao, Yihao Ang, Qiang Huang, Anthony KH Tung, and Zhiyong Huang. Towards controllable time series generation. arXiv preprint arXiv:2403.03698, 2024

    work page arXiv 2024

  5. [5]

    Unsupervised learning of visual features by contrasting cluster assignments

    Mathilde Caron, Ishan Misra, Julien Mairal, Priya Goyal, Piotr Bojanowski, and Armand Joulin. Unsupervised learning of visual features by contrasting cluster assignments. Advances in neural information processing systems, 33: 0 9912--9924, 2020

    work page 2020

  6. [6]

    Challenges and corresponding solutions of generative adversarial networks (gans): a survey study

    Haiyang Chen. Challenges and corresponding solutions of generative adversarial networks (gans): a survey study. In Journal of Physics: Conference Series, volume 1827, pp.\ 012066. IOP Publishing, 2021

    work page 2021

  7. [7]

    Beijing multi-site air-quality data, 2019

    Song Chen. Beijing multi-site air-quality data, 2019. URL https://doi.org/10.24432/C5RK5G

    work page doi:10.24432/c5rk5g 2019

  8. [8]

    On the constrained time-series generation problem

    Andrea Coletta, Sriram Gopalakrishnan, Daniel Borrajo, and Svitlana Vyetrenko. On the constrained time-series generation problem. Advances in Neural Information Processing Systems, 36: 0 61048--61059, 2023

    work page 2023

  9. [9]

    Sinkhorn distances: Lightspeed computation of optimal transport

    Marco Cuturi. Sinkhorn distances: Lightspeed computation of optimal transport. Advances in neural information processing systems, 26, 2013

    work page 2013

  10. [10]

    Adversarial audio synthesis

    Chris Donahue, Julian McAuley, and Miller Puckette. Adversarial audio synthesis. In International Conference on Learning Representations, 2019

    work page 2019

  11. [11]

    Real-valued (Medical) Time Series Generation with Recurrent Conditional GANs

    Crist \'o bal Esteban, Stephanie L Hyland, and Gunnar R \"a tsch. Real-valued (medical) time series generation with recurrent conditional gans. arXiv preprint arXiv:1706.02633, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  12. [12]

    Classifier-Free Diffusion Guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  13. [13]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33: 0 6840--6851, 2020

    work page 2020

  14. [14]

    Metro interstate traffic volume data set, 2019

    John Hogue. Metro interstate traffic volume data set, 2019. URL https://doi.org/10.24432/C5X60B

    work page doi:10.24432/c5x60b 2019

  15. [15]

    A causal lens for controllable text generation

    Zhiting Hu and Li Erran Li. A causal lens for controllable text generation. Advances in Neural Information Processing Systems, 34: 0 24941--24955, 2021

    work page 2021

  16. [16]

    Timedp: Learning to generate multi-domain time series with domain prompts

    Yu-Hao Huang, Chang Xu, Yueying Wu, Wu-Jun Li, and Jiang Bian. Timedp: Learning to generate multi-domain time series with domain prompts. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pp.\ 17520--17527, 2025

    work page 2025

  17. [17]

    Causalgan: Learning causal implicit generative models with adversarial training

    Murat Kocaoglu, Christopher Snyder, Alexandros G Dimakis, and Sriram Vishwanath. Causalgan: Learning causal implicit generative models with adversarial training. In International Conference on Learning Representations, 2018

    work page 2018

  18. [18]

    From identifiable causal repre- sentations to controllable counterfactual generation: A survey on causal generative modeling

    Aneesh Komanduri, Xintao Wu, Yongkai Wu, and Feng Chen. From identifiable causal representations to controllable counterfactual generation: A survey on causal generative modeling. arXiv preprint arXiv:2310.11011, 2023

    work page arXiv 2023

  19. [19]

    Robert L. Lehrman. Physics the Easy Way. Barron's Educational Series, Hauppauge, NY, 3rd edition, 1998. OCLC: 1036803976

    work page 1998

  20. [20]

    Bridge: Bootstrapping text to control time-series generation via multi-agent iterative optimization and diffusion modelling

    Hao Li, Yu-Hao Huang, Chang Xu, Viktor Schlegel, Ren-He Jiang, Riza Batista-Navarro, Goran Nenadic, and Jiang Bian. Bridge: Bootstrapping text to control time-series generation via multi-agent iterative optimization and diffusion modelling. ICML, 2025

    work page 2025

  21. [21]

    Diffusion models for time-series applications: a survey

    Lequan Lin, Zhengkun Li, Ruikun Li, Xuliang Li, and Junbin Gao. Diffusion models for time-series applications: a survey. Frontiers of Information Technology & Electronic Engineering, 25 0 (1): 0 19--41, 2024

    work page 2024

  22. [22]

    A causal inference look at unsupervised video anomaly detection

    Xiangru Lin, Yuyang Chen, Guanbin Li, and Yizhou Yu. A causal inference look at unsupervised video anomaly detection. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, pp.\ 1620--1629, 2022

    work page 2022

  23. [23]

    Learning causal semantic representation for out-of-distribution prediction

    Chang Liu, Xinwei Sun, Jindong Wang, Haoyue Tang, Tao Li, Tao Qin, Wei Chen, and Tie-Yan Liu. Learning causal semantic representation for out-of-distribution prediction. Advances in Neural Information Processing Systems, 34: 0 6155--6170, 2021

    work page 2021

  24. [24]

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps

    Cheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen, Chongxuan Li, and Jun Zhu. Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps. Advances in neural information processing systems, 35: 0 5775--5787, 2022

    work page 2022

  25. [25]

    Causal inference challenges and new directions for epidemiologic research on the health effects of social policies

    Ellicott C Matthay and M Maria Glymour. Causal inference challenges and new directions for epidemiologic research on the health effects of social policies. Current Epidemiology Reports, 9 0 (1): 0 22--37, 2022

    work page 2022

  26. [26]

    Neural variational inference and learning in belief networks

    Andriy Mnih and Karol Gregor. Neural variational inference and learning in belief networks. In International Conference on Machine Learning, pp.\ 1791--1799. PMLR, 2014

    work page 2014

  27. [27]

    Time weaver: A conditional time series generation model

    Sai Shankar Narasimhan, Shubhankar Agarwal, Oguzhan Akcin, Sujay Sanghavi, and Sandeep P Chinchali. Time weaver: A conditional time series generation model. In International Conference on Machine Learning, pp.\ 37293--37320. PMLR, 2024

    work page 2024

  28. [28]

    Causality

    Judea Pearl. Causality. Causality: Models, Reasoning, and Inference. Cambridge University Press, 2009. ISBN 9780521895606. URL https://books.google.com/books?id=f4nuexsNVZIC

    work page 2009

  29. [29]

    Causal inference in statistics: A primer

    Judea Pearl, Madelyn Glymour, and Nicholas P Jewell. Causal inference in statistics: A primer. John Wiley & Sons, 2016

    work page 2016

  30. [30]

    Models, reasoning and inference

    Judea Pearl et al. Models, reasoning and inference. Cambridge, UK: CambridgeUniversityPress, 19 0 (2), 2000

    work page 2000

  31. [31]

    T-CGAN: Conditional Generative Adversarial Network for Data Augmentation in Noisy Time Series with Irregular Sampling

    Giorgia Ramponi, Pavlos Protopapas, Marco Brambilla, and Ryan Janssen. T-cgan: Conditional generative adversarial network for data augmentation in noisy time series with irregular sampling. arXiv preprint arXiv:1811.08295, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  32. [32]

    Counterfactual generative networks

    Axel Sauer and Andreas Geiger. Counterfactual generative networks. In International Conference on Learning Representations, 2021

    work page 2021

  33. [33]

    Conditional gan for timeseries generation

    Kaleb E Smith and Anthony O Smith. Conditional gan for timeseries generation. arXiv preprint arXiv:2006.16477, 2020

    work page arXiv 2006

  34. [34]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International conference on machine learning, pp.\ 2256--2265. pmlr, 2015

    work page 2015

  35. [35]

    Generative modeling by estimating gradients of the data distribution

    Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. Advances in neural information processing systems, 32, 2019

    work page 2019

  36. [36]

    Csdi: Conditional score-based diffusion models for probabilistic time series imputation

    Yusuke Tashiro, Jiaming Song, Yang Song, and Stefano Ermon. Csdi: Conditional score-based diffusion models for probabilistic time series imputation. Advances in neural information processing systems, 34: 0 24804--24816, 2021

    work page 2021

  37. [37]

    Deepfake electrocardiograms using generative adversarial networks are the beginning of the end for privacy issues in medicine

    Vajira Thambawita, Jonas L Isaksen, Steven A Hicks, Jonas Ghouse, Gustav Ahlberg, Allan Linneberg, Niels Grarup, Christina Ellervik, Morten Salling Olesen, Torben Hansen, et al. Deepfake electrocardiograms using generative adversarial networks are the beginning of the end for privacy issues in medicine. Scientific reports, 11 0 (1): 0 21896, 2021

    work page 2021

  38. [38]

    On the combination of independent test statistics

    WR Van Zwet and J Oosterhoff. On the combination of independent test statistics. The Annals of Mathematical Statistics, 38 0 (3): 0 659--680, 1967

    work page 1967

  39. [39]

    Towards precise embodied dialogue localization via causality guided diffusion

    Haoyu Wang, Le Wang, Sanping Zhou, Jingyi Tian, Zheng Qin, Yabing Wang, Gang Hua, and Wei Tang. Towards precise embodied dialogue localization via causality guided diffusion. In Proceedings of the Computer Vision and Pattern Recognition Conference, pp.\ 13350--13360, 2025

    work page 2025

  40. [40]

    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. In Proceedings of the 31st ACM International Conference on Advances in Geographic Information Systems, pp.\ 1--12, 2023

    work page 2023

  41. [41]

    Deciphering spatio-temporal graph forecasting: A causal lens and treatment

    Yutong Xia, Yuxuan Liang, Haomin Wen, Xu Liu, Kun Wang, Zhengyang Zhou, and Roger Zimmermann. Deciphering spatio-temporal graph forecasting: A causal lens and treatment. Advances in Neural Information Processing Systems, 36: 0 37068--37088, 2023

    work page 2023

  42. [42]

    Rewarddance: Scaling visual reward modeling via generative next-token prediction.arXiv preprint arXiv:2504.12345, 2025

    Yutong Xia, Ao Qu, Yunhan Zheng, Yihong Tang, Dingyi Zhuang, Yuxuan Liang, Shenhao Wang, Cathy Wu, Lijun Sun, Roger Zimmermann, et al. Reimagining urban science: Scaling causal inference with large language models. arXiv preprint arXiv:2504.12345, 2025 a

    work page arXiv 2025

  43. [43]

    Capulse: Detecting anomalies by tuning in to the causal rhythms of time series

    Yutong Xia, Yingying Zhang, Yuxuan Liang, Lunting Fan, Qingsong Wen, and Roger Zimmermann. Capulse: Detecting anomalies by tuning in to the causal rhythms of time series. arXiv preprint arXiv:2508.04630, 2025 b

    work page arXiv 2025

  44. [44]

    Causalvae: Disentangled representation learning via neural structural causal models

    Mengyue Yang, Furui Liu, Zhitang Chen, Xinwei Shen, Jianye Hao, and Jun Wang. Causalvae: Disentangled representation learning via neural structural causal models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp.\ 9593--9602, 2021

    work page 2021

  45. [45]

    A survey on diffusion models for time series and spatio-temporal data

    Yiyuan Yang, Ming Jin, Haomin Wen, Chaoli Zhang, Yuxuan Liang, Lintao Ma, Yi Wang, Chenghao Liu, Bin Yang, Zenglin Xu, et al. A survey on diffusion models for time series and spatio-temporal data. arXiv preprint arXiv:2404.18886, 2024

    work page arXiv 2024

  46. [46]

    Time-series generative adversarial networks

    Jinsung Yoon, Daniel Jarrett, and Mihaela Van der Schaar. Time-series generative adversarial networks. Advances in neural information processing systems, 32, 2019

    work page 2019

  47. [47]

    Anonymization through data synthesis using generative adversarial networks (ads-gan)

    Jinsung Yoon, Lydia N Drumright, and Mihaela Van Der Schaar. Anonymization through data synthesis using generative adversarial networks (ads-gan). IEEE journal of biomedical and health informatics, 24 0 (8): 0 2378--2388, 2020

    work page 2020

  48. [48]

    Diffusion-ts: Interpretable diffusion for general time series generation

    Xinyu Yuan and Yan Qiao. Diffusion-ts: Interpretable diffusion for general time series generation. In The Twelfth International Conference on Learning Representations, 2024

    work page 2024

  49. [49]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should not add it explicitly Type <Return> for now, but then later remove the command n...

    work page

  50. [50]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@first@sw \@firstoftwo \@ifundefined NAT@b*@#2 \@firstoftwo @num @NAT@ctr \@secondoft...

    work page

  51. [51]

    Angular and Scale Alignment losses are evaluated for long-term video generation, with MSE as a naive baseline of aligning both angular and scale information

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibsetup #1 @NAT@ctr @ @openbib .11em \@plus.33em \@minus.07em 4000 4000 `\.\@m @bibit...

    work page arXiv