Pith.

Machine Learning · arXiv:2611.10668 v1 · submitted 7 Nov 2026 · 49 references

Causal Heteroscedastic Structure Learning From Incomplete Temporal data

Abdellah Rahmani, Pascal Frossard

The paper says

“Discovering causal structures in multivariate time series (MTS) is critical in domains such as finance and neuroscience, yet remains challenging in practice due to three distinct obstacles.”

more of the abstract

First, MTS are routinely corrupted by missing values, where naive impute-then-discover pipelines distort the underlying data distribution and yield spurious causal graphs. Second, identifying instantaneous effects is difficult even on complete data, and missingness further entangles imputation with graph recovery. Third, real-world data frequently exhibits heteroscedastic noise with variance driven by both instantaneous and lagged causes, violating the core assumptions of standard discovery methods. Existing approaches fail to address these challenges concurrently, typically presuming complete data, homoscedastic noise, or strictly time-lagged relationships. Crucially, we show through ablation studies that resolving these issues sequentially, rather than jointly, leads to a substantial degradation in performance. To overcome this, we introduce Cheesefill, an optimal transport framework for causal structure learning under missingness and heteroscedastic noise. Cheesefill parameterizes causal mechanisms using conditional normalizing flows and jointly learns a stochastic correction map that refines naive imputations into trajectories consistent with the inferred causal mechanisms. Theoretically, we prove that optimizing over stochastic correction maps is equivalent to a conditional kernel reformulation of the Kantorovich problem, from which our neural parameterization yields a tractable objective. Extensive experiments on synthetic and real-world benchmarks demonstrate that Cheesefill consistently outperforms existing baselines across diverse missingness regimes.

from the abstract

Pith says

Machine review · deepseek-v4-flash · 10 Nov 2026 · what was read was not recorded · not checked by a person

It can be wrong. Challenge this review or reply to it.

The passes disagreed. Reader: holds up. Skeptic: agrees with the reader. Referee: major revision.

Solid method for a hard problem, but the theoretical framing overpromises relative to the practical relaxation. This line is wrong

2 major objections · 6 minor

How often objections from reviews like this one are right: not yet measured

Are you an author? What you can do.

Not checked: figures, code, what the 49 references say.

The paper says, at §3.2 (Eq. (6) and §4, Eq. (10) and §6 (Limitations)

The review did not quote this passage.

Pith says: Objection 1 of 2 · major

The central theoretical justification (Proposition 3.1) for the stochastic correction map relies on the exact admissibility condition (Definition 2.1). The practical algorithm, however, replaces this condition with an MMD penalty (Section 4, last paragraph and Eq. (10)). The authors acknowledge this relaxation as a limitation (Section 6, Appendix B), but they do not analyze how the MMD penalty relates to the theoretical optimal transport bound.

more of objection 1

Concretely, the bound in Eq. (5) holds only for admissible kernels; when the MMD penalty is minimized but not zero, the optimized ψ may not lie in Ψ_adm, and Eq. (5) no longer bounds the Kantorovich cost. This gap is load-bearing because the paper motivates the correction map by Proposition 3.1, yet the algorithm's objective is a relaxation whose proximity to the Kantorovich problem is unquantified. I request that the authors either (a) provide an analysis (e.g., a bound on the gap in terms of the MMD error) or (b) soften the claim that the theoretical result directly justifies the practical objective, and discuss under what conditions (e.g., when λ_map is large enough) the relaxation is expected to be approximately valid.

The paper says, at §4 (Correction map parametrization)

The review did not quote this passage.

Pith says: Objection 2 of 2 · major

The correction map T_ψ is parameterized as a diagonal Gaussian over missing entries only, with fixed identity covariance. The authors justify this choice by noting that the CNF already models parent-dependent residual scale. However, the expressiveness of this family is crucial for the ε-optimal admissibility condition in Proposition 3.1.

more of objection 2

The paper does not discuss whether this restricted family can approximate an admissible kernel to within a small ε for the problems considered. I recommend adding a remark about the expressiveness limitation and, if possible, a sensitivity analysis (e.g., varying the covariance or allowing non-diagonal structure) to assess the practical impact.

6 minor objections

The paper says, at Abstract and §6

Pith says: minor objection 1 of 6

The conclusion contains a typographical artifact: 'noise regimes.]' has a stray closing bracket after the period. Please correct.

The paper says, at §1 (Introduction, first paragraph)

Pith says: minor objection 2 of 6

The related-work paragraph lists 'neural approaches' for Granger causality; consider citing a more recent survey or standard reference for completeness.

show 4 more

The paper says, at Figure 1

Pith says: minor objection 3 of 6

The notation in Figure 1 uses 'G→' twice and the arrow label 'G→' could be clearer; consider using distinct labels for the graph (G*) and the TGNN/CNF parameters (Θ).

The paper says, at §3.2 (Eq. (5))

Pith says: minor objection 4 of 6

After Eq. (5), the text says 'The inequality is tight whenever the parametric family contains an optimal admissible kernel.' This is correct, but for precision, consider noting that the family must also be able to represent that kernel exactly (not just in the limit).

The paper says, at §5.1 (Table 1)

Pith says: minor objection 5 of 6

The ablation table (Table 1) reports F1, precision, recall, and ratio. What does 'Ratio' refer to? Please define in the caption or text.

The paper says, at Appendix G (MNAR Experiments)

Pith says: minor objection 6 of 6

The MNAR experiments use 'a small observed subset to model the missingness process'. It would be helpful to specify exactly which 10% of variables are fully observed and how this choice was made (e.g., fixed random seed).

What would settle it

On the synthetic heteroscedastic 20-node MCAR dataset with 30% missingness used in Figure 3, run a sensitivity analysis: train Cheesefill with λ_map ∈ {0, 0.1, 1, 10}. For each setting, compute the final validation MMD between corrected trajectories and SEM-generated trajectories, as well as instantaneous and lagged F1. If increasing λ_map from 0.1 to 10 changes F1 by more than 5 percentage points, then the MMD relaxation critically influences recovery and the theoretical connection is not operational. If F1 is stable and MMD is below 0.05 (on normalized data), the relaxation is harmless and the concern is resolved.

from the skeptic's reading

How the review went

4 readings by deepseek-v4-flash · 10 Nov 2026

  1. Reader holds up confidence high
  2. Skeptic agrees 1 min 0 s · 35,072 tokens read, 6,153 written
  3. Editor wrote the verdict above 13 s · 35,860 tokens read, 1,178 written
  4. Referee major revision: 2 major objections, 6 minor 23 s · 35,462 tokens read, 2,331 written

Not run on this paper: the simulated authors' defence.

Pith did not check this review's objections against the paper.

The review in full

The review in brief, machine-written

The paper tackles a genuinely hard open problem: causal structure learning from incomplete multivariate time series when you have both instantaneous and lagged links plus heteroscedastic noise. The combination of optimal transport, a stochastic correction map, and conditional normalizing flows is novel and well motivated. The experiments are extensive—over 100 synthetic settings plus two real benchmarks—and the results are consistently strong across MCAR, MAR, and even MNAR. The ablation studies confirm that both the correction map and the normalizing flow matter.

The main soft spot is a gap between theory and practice. Proposition 3.1 gives a clean equivalence: optimizing over admissible stochastic kernels is exactly the Kantorovich problem. But the practical algorithm replaces the exact admissibility constraint with an MMD penalty. That is a relaxation, and the theoretical upper bound (Eq. 5) no longer applies directly to the optimized objective. The authors acknowledge this in the Limitations section, but they do not quantify the impact—no sensitivity analysis on the MMD weight, no check of how well the penalty actually enforces the constraint. For a paper that leans heavily on the OT formalism, this is a real and load-bearing gap.

That said, the empirical evidence is strong enough that the method works despite the relaxation. The paper is a genuine step forward for practitioners. What it delivers is a well-engineered neural pipeline that outperforms existing methods; the theory is a useful framing but not a provable guarantee in practice.

Who is it for: researchers working on causal discovery from time series, especially those dealing with real-world missingness. The paper deserves a serious peer review. I would recommend accept with revision, asking the authors to either (a) provide an experiment that varies the MMD weight and shows how the recovered graph degrades, or (b) add a formal argument that the MMD relaxation still leads to a consistent estimator under some conditions. Without that, readers may overinterpret the theoretical result.

The strongest objection

The paper's Proposition 3.1 establishes that optimizing over admissible stochastic kernels in an unrestricted family is equivalent to the Kantorovich problem, and that a parametric family yields an upper bound that is tight if it contains an ε-optimal admissible kernel. However, the practical algorithm replaces the exact admissibility constraint (Definition 2.1) with an MMD penalty between corrected trajectories Y and SEM-generated samples Y_SEM. This penalty does not guarantee that T_ψ ∈ F(P_\tilde{X}, Q_Θ); it only encourages distributional matching in the MMD sense, which is not equivalent to the exact marginal constraint. Consequently, the theoretical guarantee in Eq. (5) does not directly apply to the optimized objective. The paper acknowledges this relaxation as a limitation but does not quantify its impact. If the MMD penalty is weak or the optimization fails to reduce it sufficiently, the corrected trajectories may remain inconsistent with the SEM, and the graph recovery could be driven by the imputer rather than the causal mechanisms. This concern is load-bearing because the central claim that Cheesefill recovers causal structure relies on the correction map being compatible with the SEM; the theoretical framing claims such compatibility; yet the practical enforcement is indirect and unvalidated.

The reader correctly identifies the expressivity of the neural family and the MMD relaxation as the weakest assumptions. Our stress-test refines this into a specific gap: the practical objective relaxes the admissibility condition, so the theoretical Proposition 3.1 does not directly justify the algorithm. The proposed concrete test would settle whether this gap matters empirically. Given the paper's strong and transparent empirical results (over 108 synthetic datasets, real-world benchmarks, ablation studies), this concern does not invalidate the central empirical claim; it merely identifies an area where the theory-practice connection could be tightened. Therefore the original ACCEPT verdict remains appropriate, with the suggestion that future work could provide a tighter analysis or a diagnostic for admissibility.

The referee's report

If the reported results hold, Cheesefill represents a meaningful advance because it is, to my knowledge, the first method to simultaneously address three realistic challenges—missing observations, instantaneous effects, and heteroscedastic noise—in a single framework. The optimal transport perspective is novel in this application and provides a clean conceptual link between imputation and causal discovery. The ablation study convincingly shows that both the stochastic correction map and the conditional normalizing flow are necessary for the strong performance. The experimental evaluation is extensive (>108 synthetic datasets, two real benchmarks) and the code is provided, supporting reproducibility.

Pith's reading of the paper (a machine's summary, not the authors')

Cheesefill is a framework for causal structure learning from multivariate time series that simultaneously handles three real-world challenges: missing values, instantaneous (contemporaneous) and time-lagged causal links, and heteroscedastic noise (noise variance that depends on the cause values). The key idea is to frame causal discovery under missingness as an optimal transport problem between the distribution of naively imputed trajectories and the distribution induced by a structural equation model. A learned stochastic correction map refines imputations to make them consistent with the causal mechanisms, and a conditional normalizing flow models the parent-dependent noise. Theoretically, optimizing over such correction maps is equivalent to a conditional kernel formulation of the Kantorovich problem. Experiments show that Cheesefill consistently outperforms impute-then-discover pipelines and existing methods designed for incomplete MTS across various missingness regimes, noise types, and graph sizes.

What carries the argument. The stochastic correction map $T_\psi(dY|\tilde{X})$ refines pseudo-complete trajectories $\tilde{X}$ from a parametric imputer into refined trajectories $Y$ that are consistent with the SEM. Proposition 3.1 proves that optimizing over admissible correction maps is equivalent to the Kantorovich optimal transport problem between the pseudo-complete and SEM-induced distributions, with the mechanism cost (negative log-likelihood under the SEM) as transport cost. The correction map is parameterized as a diagonal Gaussian over missing entries only, conditioned on the pseudo-complete trajectory. The SEM uses a temporal graph neural network for mean functions $f_i$ and conditional normalizing flows for residual densities $\rho_i$, capturing heteroscedasticity. Bayesian structure learning averages over graph samples from a variational posterior, with entropy regularization.

Core claim. This paper claims that it is possible to recover both instantaneous and time-lagged causal links from incomplete multivariate time series in the presence of heteroscedastic noise, by jointly learning an imputation strategy and a causal model in an optimal-transport framework. The central contribution is showing that a stochastic correction map that refines naive imputations can be interpreted as a conditional kernel reformulation of the Kantorovich problem, making the joint optimization theoretically principled. The method, Cheesefill, parameterizes the correction map and the causal mechanisms (mean functions and heteroscedastic noise distributions) with neural networks, and uses Bayesian structure learning over graphs to mitigate brittleness under missingness. Extensive empirical results demonstrate robust performance across MCAR, MAR, and MNAR missingness, with up to 50% missing entries.

The premise it rests on. The neural network family used for the stochastic correction map must be expressive enough to contain a kernel that perfectly matches the pseudo-complete distribution to the SEM-induced distribution; otherwise the optimal transport bound is not tight and the method may not achieve the theoretical optimum.

If the paper is right (Pith's inferences, not the authors')

Cheesefill enables reliable causal discovery from real-world time series where missing data and heteroscedastic noise are common, such as fMRI, EEG, and financial data.

The optimal transport perspective suggests that other imputation-sensitive tasks (e.g., counterfactual inference, treatment effect estimation) could benefit from a similar joint correction-and-discovery approach.

The method's robustness to MCAR, MAR, and MNAR missingness (up to 50%) opens the door for causal analysis in datasets previously considered too incomplete.

By recovering both instantaneous and lagged links, Cheesefill provides a more complete picture of causal dynamics than methods restricted to lagged relationships.

The ablation study confirms that both the stochastic correction map and the heteroscedastic noise model are essential, implying that simpler pipelines will likely fail in the combined challenging setting.

If the optimal transport framing is as fundamental as claimed, then any future method that replaces the neural correction map with a different coupling (e.g., a learned deterministic transform or a GAN-style discriminator) should be measurable against the same Kantorovich equivalence.

The paper's reliance on an MMD penalty to enforce admissibility of the correction map is a practical relaxation that may not guarantee distributional matching in high dimensions; this suggests a need for tighter surrogates, possibly based on Sinkhorn divergences or adversarial objectives.

The Bayesian structure learning component could be extended to non-stationary settings by allowing the variational posterior to vary over time or regimes, which the authors mention as a limitation.

Check this review

Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.

How often reviews like this one are right

Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.

How often objections from reviews like this one are right: not yet measured

Every value above, in the review's public record (JSON)

Reviews by people

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

Discovering causal structures in multivariate time series (MTS) is critical in domains such as finance and neuroscience, yet remains challenging in practice due to three distinct obstacles. First, MTS are routinely corrupted by missing values, where naive impute-then-discover pipelines distort the underlying data distribution and yield spurious causal graphs. Second, identifying instantaneous effects is difficult even on complete data, and missingness further entangles imputation with graph recovery. Third, real-world data frequently exhibits heteroscedastic noise with variance driven by both instantaneous and lagged causes, violating the core assumptions of standard discovery methods. Existing approaches fail to address these challenges concurrently, typically presuming complete data, homoscedastic noise, or strictly time-lagged relationships. Crucially, we show through ablation studies that resolving these issues sequentially, rather than jointly, leads to a substantial degradation in performance. To overcome this, we introduce Cheesefill, an optimal transport framework for causal structure learning under missingness and heteroscedastic noise. Cheesefill parameterizes causal mechanisms using conditional normalizing flows and jointly learns a stochastic correction map that refines naive imputations into trajectories consistent with the inferred causal mechanisms. Theoretically, we prove that optimizing over stochastic correction maps is equivalent to a conditional kernel reformulation of the Kantorovich problem, from which our neural parameterization yields a tractable objective. Extensive experiments on synthetic and real-world benchmarks demonstrate that Cheesefill consistently outperforms existing baselines across diverse missingness regimes.

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-11-10

The authors' main claim, as the classifier picked it out 0.97

“Extensive experiments on synthetic and real-world benchmarks demonstrate that Cheesefill consistently outperforms existing baselines across diverse missingness regimes.”

How the main claim is stated: asserted 1.00

What the abstract itself states, as probabilities: claims first 0.12 · states limitation 0.09 · human participants 0.07 · controlled comparison 0.81 · randomized assignment 0.02 · states quantitative result 0.06 · compares with prior results 0.96 · mentions code or data availability 0.03

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
Discovering causal structures in multivariate time series (MTS) is critical in domains such as finance and neuroscience, yet remains challenging in practice due to three distinct obstacles. background 1.00 0.06 0.03 asserted 1.00
First, MTS are routinely corrupted by missing values, where naive impute-then-discover pipelines distort the underlying data distribution and yield spurious causal graphs. background 0.88 0.10 0.02 asserted 1.00
Second, identifying instantaneous effects is difficult even on complete data, and missingness further entangles imputation with graph recovery. background 0.77 0.10 0.03 asserted 1.00
Third, real-world data frequently exhibits heteroscedastic noise with variance driven by both instantaneous and lagged causes, violating the core assumptions of standard discovery methods. background 0.89 0.11 0.03 asserted 0.98
Existing approaches fail to address these challenges concurrently, typically presuming complete data, homoscedastic noise, or strictly time-lagged relationships. gap 0.93 0.05 0.05 asserted 1.00
Crucially, we show through ablation studies that resolving these issues sequentially, rather than jointly, leads to a substantial degradation in performance. result 0.79 0.95 0.24 asserted 1.00
To overcome this, we introduce Cheesefill, an optimal transport framework for causal structure learning under missingness and heteroscedastic noise. contribution 0.92 0.84 0.21 asserted 1.00
Cheesefill parameterizes causal mechanisms using conditional normalizing flows and jointly learns a stochastic correction map that refines naive imputations into trajectories consistent with the inferred causal mechanisms. method 0.98 0.92 0.11 asserted 1.00
Theoretically, we prove that optimizing over stochastic correction maps is equivalent to a conditional kernel reformulation of the Kantorovich problem, from which our neural parameterization yields a tractable objective. method 0.61 0.95 0.06 asserted 1.00
Extensive experiments on synthetic and real-world benchmarks demonstrate that Cheesefill consistently outperforms existing baselines across diverse missingness regimes. result 1.00 0.95 0.54 asserted 1.00

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.68

What was examined, and what was not

Sources read

  • arXiv listing abstract v1

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-10: 1 request, 7,673 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

Figures (22)
Figure 1 of the paper
Figure 1. Cheesefill overview. The incomplete MTS dataset
Figure 2 of the paper
Figure 2. Temporal graph neural network (TGNN) used by Cheesefill. where χ and ϑ are neural network blocks illustrated in
Figure 3 of the paper
Figure 3. MCAR MTS with 20 nodes. F1 score (top, higher is better) and NHD ratio (bottom, lower is better) for instantaneous and time lagged edges, evaluated under 2 and 3 maximum lags as the missing rate increases from 10% to 50%. reproduces the recovery performance obtained when the stochastic correction map and CNF are used together. Benchmarks. We benchmark our model against several baselines, including causal discovery methods designed for incomplete MTS, such as CUTS [8]. Cheesefill is also compared with models that assume complete MTS, including Fantom [32], PCMCI+ [36], Dynotears…
Figure 4 of the paper
Figure 4. F1 score on NetSim across missingness levels
Figure 5 of the paper
Figure 5. Figure taken from Gamella et al. [12]. Diagrams of the wind tunel causal chamber and its main components, including the amplification circuit that drives the speaker of the wind tunnel. The variables measured by the chamber are displayed in black math print. Sensor measurements are denoted by a tilde. Manipulable variables, that is, actuators and sensor parameters, are shown in bold symbols. 4https://openreview.net/forum?id=BgwtOwFSvY 17
Figure 7 of the paper
Figure 7. F1 score on NetSim across missing￾ness levels. The dataset comprises 16 variables: controllable load of the two fans Lin, Lout, their measurable speed (˜ωin , ω¯out ), the current draw by the fans  C˜ in , C˜ out  ( C˜ in , C˜ out ), the resulting air pressure inside the chamber ( P˜ dw , P˜ up ) or at its intake ( P˜ int ), and the hatch H. In the circuit that drives the speaker, we can manipulate the potentiometers ( A1, A2 ) that control the amplification, monitoring the resulting signal at different points of the circuit  S˜ 1, S˜ 2  and through the microphone output (M˜ ).…
Figure 8 of the paper
Figure 8. MCAR MTS with 20 nodes. F1 score (top, higher is better) and NHD ratio (bottom, lower is better) for instantaneous and time lagged edges, evaluated under 2 and 3 maximum lags as the missing rate increases from 10% to 50%
Figure 9 of the paper
Figure 9. MCAR MTS with 40 nodes. F1 score (top, higher is better) and NHD ratio (bottom, lower is better) for instantaneous and time lagged edges, evaluated under 2 and 3 maximum lags as the missing rate increases from 10% to 50%. MCAR, 40 nodes
Figure 10 of the paper
Figure 10. MAR MTS with 20 nodes. F1 score (top, higher is better) and NHD ratio (bottom, lower is better) for instantaneous and time lagged edges, evaluated under 2 and 3 maximum lags as the missing rate increases from 10% to 50%
Figure 11 of the paper
Figure 11. MAR MTS with 40 nodes. F1 score (top, higher is better) and NHD ratio (bottom, lower is better) for instantaneous and time lagged edges, evaluated under 2 and 3 maximum lags as the missing rate increases from 10% to 50%. introduce systematic bias. Cheesefill remains close to the full-data reference across both 2 lag and 3 lag graphs, for both instantaneous and lagged edges. Its NHD ratio stays near zero in almost all settings, indicating that the recovered graph differs only marginally from the full-data reference even as the missing rate increases. In contrast, Fantom+TimeMixer…
Figure 12 of the paper
Figure 12. MNAR MTS with 20 nodes. F1 score (top, higher is better) and NHD ratio (bottom, lower is better) for instantaneous and time lagged edges, evaluated under 2 and 3 maximum lags as the missing rate increases from 10% to 50%. across both 2 lag and 3 lag graphs. Its F1 score stays high for instantaneous and lagged edges even as the missing rate increases, and its NHD ratio remains near zero in almost all settings. This is notable because MNAR corruption can systematically bias the pseudo complete distribution, making fixed imputations especially unreliable. Fantom+TimeMixer is…
Figure 13 of the paper
Figure 13. MCAR and MAR MTS with 20 nodes. F1 score (top, higher is better) and NHD ratio (bottom, lower is better) for instantaneous and time lagged edges, evaluated under 3 maximum lags as the missing rate increases from 10% to 50%. 23
Figure 14 of the paper
Figure 14. MCAR MTS with 20 nodes. SHD (top, lower is better), Precision (middle, higher is better) and Recall (bottom, higher is better) for instantaneous and time lagged edges, evaluated under 3 maximum lags as the missing rate increases from 10% to 50%. 0.1 0.2 0.3 0.4 0.5 0 20 40 60 80 SHD Instantaneous 0.1 0.2 0.3 0.4 0.5 0 200 400 600 800 Lag 0.1 0.2 0.3 0.4 0.5 0 20 40 60 80 100 Precision (%) 0.1 0.2 0.3 0.4 0.5 0 20 40 60 80 100 0.1 0.2 0.3 0.4 0.5 Missing data % 0 20 40 60 80 100 Recall (%) 0.1 0.2 0.3 0.4 0.5 Missing data % 0 20 40 60 80 100 MAR — 20 nodes Cheesefill PCMCI+ CUTS…
Figure 15 of the paper
Figure 15. MAR MTS with 20 nodes. SHD (top, lower is better), Precision (middle, higher is better) and Recall (bottom, higher is better) for instantaneous and time lagged edges, evaluated under 3 maximum lags as the missing rate increases from 10% to 50%. 24
Figure 16 of the paper
Figure 16. MCAR and MAR MTS with 40 nodes. F1 score (top, higher is better) and NHD ratio (bottom, lower is better) for instantaneous and time lagged edges, evaluated under 3 maximum lags as the missing rate increases from 10% to 50%. 0.1 0.2 0.3 0.4 0.5 0 50 100 150 200 SHD Instantaneous 0.1 0.2 0.3 0.4 0.5 0 500 1000 1500 2000 2500 Lag 0.1 0.2 0.3 0.4 0.5 0 20 40 60 80 100 Precision (%) 0.1 0.2 0.3 0.4 0.5 0 20 40 60 80 100 0.1 0.2 0.3 0.4 0.5 Missing data % 0 20 40 60 80 100 Recall (%) 0.1 0.2 0.3 0.4 0.5 Missing data % 0 20 40 60 80 100 MCAR — 40 nodes Cheesefill PCMCI+ CUTS TimeMixer…
Figure 17 of the paper
Figure 17. MCAR MTS with 40 nodes. SHD (top, lower is better), Precision (middle, higher is better) and Recall (bottom, higher is better) for instantaneous and time lagged edges, evaluated under 3 maximum lags as the missing rate increases from 10% to 50%. 25
Figure 18 of the paper
Figure 18. MAR MTS with 40 nodes. SHD (top, lower is better), Precision (middle, higher is better) and Recall (bottom, higher is better) for instantaneous and time lagged edges, evaluated under 3 maximum lags as the missing rate increases from 10% to 50%. H.2 MNAR results 0.1 0.2 0.3 0.4 0.5 0 20 40 60 80 100 Instantaneous 0.1 0.2 0.3 0.4 0.5 0 20 40 60 80 100 Lag 0.00 0.08 0.16 0.22 0.33 0.28 0.28 0.32 0.30 0.31 0.28 0.28 0.32 0.30 0.31 0.56 0.58 0.59 0.62 0.58 1.00 1.00 1.00 1.00 1.00 NA NA NA NA NA Cheesefill CUTS 0.1 0.2 0.3 0.4 0.5 0.00 0.04 0.06 0.12 0.20 0.71 0.72 0.72 0.72 0.72 0.71…
Figure 19 of the paper
Figure 19. MNAR MTS with 20 nodes. F1 score (top, higher is better) and NHD ratio (bottom, lower is better) for instantaneous and time lagged edges, evaluated under 3 maximum lags as the missing rate increases from 10% to 50%. 26
Figure 20 of the paper
Figure 20. MNAR MTS with 20 nodes. SHD (top, lower is better), Precision (middle, higher is better) and Recall (bottom, higher is better) for instantaneous and time lagged edges, evaluated under 3 maximum lags as the missing rate increases from 10% to 50%. 0.1 0.2 0.3 0.4 0.5 0 20 40 60 80 100 Instantaneous 0.1 0.2 0.3 0.4 0.5 0 20 40 60 80 100 Lag 0.07 0.22 0.37 0.44 0.49 0.51 0.54 0.53 0.54 0.51 0.51 0.54 0.53 0.54 0.51 0.58 0.55 0.55 0.52 0.56 0.97 1.00 1.00 1.00 1.00 NA NA NA NA NA Cheesefill CUTS 0.1 0.2 0.3 0.4 0.5 0.01 0.05 0.08 0.08 0.07 0.83 0.84 0.84 0.84 0.84 0.83 0.84 0.84 0.84…
Figure 21 of the paper
Figure 21. MNAR MTS with 40 nodes. F1 score (top, higher is better) and NHD ratio (bottom, lower is better) for instantaneous and time lagged edges, evaluated under 3 maximum lags as the missing rate increases from 10% to 50%. 27
Figure 22 of the paper
Figure 22. MNAR MTS with 40 nodes. SHD (top, lower is better), Precision (middle, higher is better) and Recall (bottom, higher is better) for instantaneous and time lagged edges, evaluated under 3 maximum lags as the missing rate increases from 10% to 50%. I Graph average cost details In this section, we announce the proposition for the graph averaged mechanism cost and we show that the equivalence and parametric upper bound results of Proposition 3.1 remain valid when cG is replaced by c¯ϕ. We present also the full proof for this proposition. Proposition I.1 (Graph averaged correction map…

Showing 21 of the paper's 22 figures. Figures are the authors', reproduced from the paper.

References (49)
  1. [1] Charles K Assaad, Emilie Devijver, and Eric Gaussier. Survey and evaluation of causal discovery methods for time series.Journal of Artificial Intelligence Research, 2022 on Pith
  2. [2] Karim Assaad, Emilie Devijver, Eric Gaussier, and Ali Ait-Bachir. A mixed noise and constraint- based approach to causal inference in time series. InECML PKDD, 2021 on Pith
  3. [3] Rohit Bhattacharya, Razieh Nabi, Ilya Shpitser, and James M. Robins. Identification in missing data models represented by directed acyclic graphs. InProceedings of the 35th Conference on Uncertainty in Artificial Intelligence, volume 115 ofProceedings of Machine Learning Resea... on Pith
  4. [4] Bart Bussmann, Jannes Nys, and Steven Latré. Neural additive vector autoregression models for causal discovery in time series. InDiscovery Science, 2021 on Pith
  5. [5] Wei Cao, Dong Wang, Jian Li, Hao Zhou, Lei Li, and Yitan Li. Brits: Bidirectional recurrent imputation for time series.Advances in neural information processing systems, 31, 2018 on Pith
  6. [6] Yuxiao Cheng, Lianglong Li, Tingxiong Xiao, Zongren Li, Jinli Suo, Kunlun He, and Qionghai Dai. Cuts+: High-dimensional causal discovery from irregular time-series. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 11525–11533, 2024 on Pith
  7. [7] Yuxiao Cheng, Ziqian Wang, Tingxiong Xiao, Qin Zhong, Jinli Suo, and Kunlun He. Causaltime: Realistically generated time-series for benchmarking of causal discovery.arXiv preprint arXiv:2310.01753, 2023 on PithPith's reviewarXiv
  8. [8] Yuxiao Cheng, Runzhao Yang, Tingxiong Xiao, Zongren Li, Jinli Suo, Kunlun He, and Qionghai Dai. Cuts: Neural causal discovery from irregular time-series data.arXiv preprint arXiv:2302.07458, 2023 on PithPith's reviewarXiv
  9. [9] Andrea Cini, Ivan Marisca, and Cesare Alippi. Filling the g_ap_s: Multivariate time series imputation by graph neural networks.arXiv preprint arXiv:2108.00298, 2021 on PithPith's reviewarXiv
  10. [10] Conor Durkan, Artur Bekasov, Iain Murray, and George Papamakarios. Neural spline flows. Advances in neural information processing systems, 2019 on Pith
  11. [11] Doris Entner and Patrik O Hoyer. On causal discovery from time series data using fci.Proba- bilistic graphical models, 2010 on Pith
  12. [12] Juan L Gamella, Jonas Peters, and Peter Bühlmann. Causal chambers as a real-world physical testbed for ai methodology.Nature Machine Intelligence, 7(1):107–118, 2025 on Pith
  13. [13] Tomas Geffner, Javier Antoran, Adam Foster, Wenbo Gong, Chao Ma, Emre Kiciman, Amit Sharma, Angus Lamb, Martin Kukla, Nick Pawlowski, et al. Deep end-to-end causal inference. arXiv preprint arXiv:2202.02195, 2022 on PithPith's reviewarXiv
  14. [14] Wenbo Gong, Joel Jennings, Cheng Zhang, and Nick Pawlowski. Rhino: Deep causal temporal relationship learning with history-dependent noise.Preprint arXiv:2210.14706, 2022 on PithPith's reviewarXiv
  15. [15] Wiebke Günther, Urmi Ninad, Jonas Wahl, and Jakob Runge. Conditional independence testing with heteroskedastic data and applications to causal discovery.Advances in Neural Information Processing Systems, 2022 on Pith
  16. [16] James D Hamilton and Raul Susmel. Autoregressive conditional heteroskedasticity and changes in regime.Journal of econometrics, 1994 on Pith
  17. [17] Stefan Haufe, Klaus-Robert Müller, Guido Nolte, and Nicole Krämer. Sparse causal discovery in multivariate time series. Incausality: objectives and assessment. PMLR, 2010. 11 on Pith
  18. [18] Hilde M Huizenga and Peter CM Molenaar. Equivalent source estimation of scalp potential fields contaminated by heteroscedastic and correlated noise.Brain topography, 1995 on Pith
  19. [19] Aapo Hyvärinen, Kun Zhang, Shohei Shimizu, and Patrik O Hoyer. Estimation of a structural vector autoregression model using non-gaussianity.Journal of Machine Learning Research, 2010 on Pith
  20. [20] Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel-softmax. arXiv preprint arXiv:1611.01144, 2016 on PithPith's reviewarXiv
show 29 more
  1. [21] LK Kantorovich. On a problem of monge.Journal of Mathematical Sciences, 133(4), 2006 on Pith
  2. [22] Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014 on PithPith's reviewarXiv
  3. [23] Roderick JA Little and Donald B Rubin.Statistical analysis with missing data. John Wiley & Sons, 2019 on Pith
  4. [24] Sindy Löwe, David Madras, Richard Zemel, and Max Welling. Amortized causal discovery: Learning to infer causal graphs from time-series data. InConference on Causal Learning and Reasoning, 2022 on Pith
  5. [25] Chao Ma and Cheng Zhang. Identifiable generative models for missing not at random data imputation. InAdvances in Neural Information Processing Systems, volume 34, pages 27645– 27658, 2021 on Pith
  6. [26] Boris Muzellec, Julie Josse, Claire Boyer, and Marco Cuturi. Missing data imputation using optimal transport. InInternational Conference on Machine Learning, pages 7130–7140. PMLR, 2020 on Pith
  7. [27] Razieh Nabi, Rohit Bhattacharya, and Ilya Shpitser. Full law identification in graphical models of missing data: Completeness results. InProceedings of the 37th International Conference on Machine Learning, volume 119 ofProceedings of Machine Learning Research, pages 7153–7163... on Pith
  8. [28] Mark Newman.Networks. Oxford university press, 2018 on Pith
  9. [29] Roxana Pamfil, Nisara Sriwattanaworachai, Shaan Desai, Philip Pilgerstorfer, Konstantinos Georgatzis, Paul Beaumont, and Bryon Aragam. Dynotears: Structure learning from time-series data. InInternational conference on artificial intelligence and statistics, pages 1595–1605. Pmlr, 2020 on Pith
  10. [30] Jonas Peters, Dominik Janzing, and Bernhard Schölkopf. Causal inference on time series using restricted structural equation models.Advances in neural information processing systems, 2013 on Pith
  11. [31] Jonas Peters, Dominik Janzing, and Bernhard Schölkopf.Elements of causal inference: founda- tions and learning algorithms. The MIT Press, 2017 on Pith
  12. [32] Abdellah Rahmani and Pascal Frossard. Flow based approach for dynamic temporal causal models with non-gaussian or heteroscedastic noises. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems on Pith
  13. [33] Abdellah Rahmani and Pascal Frossard. Causal temporal regime structure learning. InThe 28th International Conference on Artificial Intelligence and Statistics, 2025 on Pith
  14. [34] Abdellah Rahmani, Arun Venkitaraman, and Pascal Frossard. A meta-gnn approach to per- sonalized seizure detection and classification. InIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023 on Pith
  15. [35] Jakob Runge. Causal network reconstruction from time series: From theoretical assumptions to practical estimation.Chaos: An Interdisciplinary Journal of Nonlinear Science, 2018 on Pith
  16. [36] Jakob Runge. Discovering contemporaneous and lagged causal relations in autocorrelated nonlinear time series datasets. InConference on Uncertainty in Artificial Intelligence. PMLR, 2020. 12 on Pith
  17. [37] Jakob Runge, Peer Nowack, Marlene Kretschmer, Seth Flaxman, and Dino Sejdinovic. Detecting and quantifying causal associations in large nonlinear time series datasets.Science advances, 2019 on Pith
  18. [38] Karen Sachs, Omar Perez, Dana Pe’er, Douglas A Lauffenburger, and Garry P Nolan. Causal protein-signaling networks derived from multiparameter single-cell data.Science, 2005 on Pith
  19. [39] Christof Seiler and Susan Holmes. Multivariate heteroscedasticity models for functional brain connectivity.Frontiers in neuroscience, 2017 on Pith
  20. [40] Xinpeng Shen, Sisi Ma, Prashanthi Vemuri, and Gyorgy Simon. Challenges and opportunities with causal discovery algorithms: application to alzheimer’s pathophysiology.Scientific reports, 2020 on Pith
  21. [41] Le Song, Mladen Kolar, and Eric Xing. Time-varying dynamic bayesian networks.In Advances in Neural Information Processing Systems, 2009 on Pith
  22. [42] Peter Spirtes, Clark N Glymour, Richard Scheines, and David Heckerman.Causation, prediction, and search. MIT press, 2000 on Pith
  23. [43] Vy V o, He Zhao, Trung Le, Edwin V Bonilla, and Dinh Phung. Optimal transport for structure learning under missing data.arXiv preprint arXiv:2402.15255, 2024 on PithPith's reviewarXiv
  24. [44] Shiyu Wang, Haixu Wu, Xiaoming Shi, Tengge Hu, Huakun Luo, Lintao Ma, James Y Zhang, and Jun Zhou. Timemixer: Decomposable multiscale mixing for time series forecasting. In International Conference on Learning Representations (ICLR), 2024 on Pith
  25. [45] Ian R White, Patrick Royston, and Angela M Wood. Multiple imputation using chained equations: issues and guidance for practice.Statistics in medicine, 30(4):377–399, 2011 on Pith
  26. [46] Xun Zheng, Bryon Aragam, Pradeep K Ravikumar, and Eric P Xing. Dags with no tears: Continuous optimization for structure learning.Advances in neural information processing systems, 31, 2018. 13 Table of content A Detailed related work 15 B Limitations 15 C Data generation and ... on Pith
  27. [47] to handle latent confounders. PCMCI [ 37] introduced a scalable two-stage algorithm for high-dimensional time series, initially restricted to lagged links and later extended to PCMCI+ on Pith
  28. [48] for contemporaneous edges. NBCB [ 2] combines noise-based and constraint-based ideas to recover summary causal graphs beyond Markov equivalence, even in the presence of instantaneous relationships. Functional-model approachesexploit identifiability under restricted noise distr... on Pith
  29. [49] Institutional review board (IRB) approvals or equivalent for research with human subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals... on Pith

Challenges and discussion

Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.

No comments yet.

Sign in to comment