Causal Discovery with Inverted Self-attention for Multivariate Time Series

Huan Huo; Tianqing Zhu; Xiufeng Liu; Yifan Yin; Yong Wang; Yusen Liu

REVIEW 5 major objections 5 minor 25 references

Inverting the tokens in a transformer and sparsifying its attention scores recovers Granger-causal graphs in multivariate time series better than existing methods.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-31 14:29 UTC pith:3ZFEFAKX

load-bearing objection Clean engineering combo of inverted tokens + Sparsemax + rank aggregation + PI that beats standard baselines on the usual nonlinear suites, still resting on the usual unproven attention-to-Granger leap. the 5 major comments →

arxiv 2607.28212 v1 pith:3ZFEFAKX submitted 2026-07-30 cs.CL cs.LG

Causal Discovery with Inverted Self-attention for Multivariate Time Series

Yusen Liu , Yong Wang , Yifan Yin , Tianqing Zhu , Xiufeng Liu , Huan Huo This is my paper

classification cs.CL cs.LG

keywords causal discoverymultivariate time seriesself-attentiontransformerGranger causalitySparsemaxpermutation importance

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Causal discovery among many interacting time series is hard because ordinary correlations mix with true directed influences, especially when the links are nonlinear or high-dimensional. This paper claims that a simple change to the transformer’s self-attention block solves much of the problem: treat each variable’s entire history as one token, compute attention across variables rather than across time, and force the resulting scores to be sparse. The sparse scores are then summed globally and checked by a permutation test, yielding a directed graph of Granger causes. On both linear chaotic maps and nonlinear systems such as Lorenz-96 and simulated fMRI, the resulting graphs match ground truth more accurately than classic statistical tests and recent neural baselines. If the method generalizes, analysts in climate, finance and neuroscience could extract usable causal structure directly from the same architectures already used for forecasting.

Core claim

By inverting the usual tokenization of a multivariate series so that each token is one variable’s full trajectory, applying Sparsemax attention, aggregating the sparse scores into a global causal matrix, and verifying each retained edge with permutation importance, one obtains Granger-causal graphs that outperform BGGranger, KGC, tsFCI, TCDF, PCMCI and DYNOTEARS on precision, recall and F1 across four standard benchmarks, with the largest gains on the nonlinear sets.

What carries the argument

The inverted causal self-attention mechanism (CSAM): each token is a complete univariate series; attention is computed across series, sparsified by Sparsemax, then globally thresholded and permutation-verified to produce the edge set.

Load-bearing premise

Non-zero sparse attention weights obtained from the inverted tokens, once summed and rank-thresholded, correspond to true Granger-causal edges rather than leftover predictive associations or training artifacts.

What would settle it

On a controlled synthetic system whose true Granger graph is known, replace Sparsemax with ordinary Softmax or disable the permutation check; if precision and F1 collapse to the level of the weaker baselines, the claimed advantage of the inverted sparse pipeline is falsified.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

Forecasting models that already use transformers can extract an explicit causal graph as a free by-product of training.
High-dimensional nonlinear systems (climate, BOLD fMRI, multi-asset finance) become amenable to automated causal screening without hand-crafted kernels or conditional-independence tests.
Sparsity induced by Sparsemax systematically reduces false-positive edges relative to dense attention baselines.
The same global aggregation-plus-verification step can be attached to other attention architectures to convert predictive attention into claimed causality.

Where Pith is reading between the lines

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

Because the method never assumes linearity, it may serve as a drop-in replacement for kernel Granger causality in domains where the appropriate kernel is unknown.
The rank-k threshold choice is still a free hyper-parameter; an information-theoretic or stability-based rule for choosing k would make the pipeline fully automatic.
If contemporaneous causation is later allowed, the same inverted-token construction could be paired with a structural-equation layer to recover instantaneous edges.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

Clean engineering combo of inverted tokens + Sparsemax + rank aggregation + PI that beats standard baselines on the usual nonlinear suites, still resting on the usual unproven attention-to-Granger leap.

read the letter

The one thing worth knowing is that this is a solid, fully-specified neural Granger pipeline: invert the tokens so each series is a token, run Sparsemax on the cross-series scores, sum them into a global strength matrix, hard-threshold at rank-k, then filter survivors with permutation importance. That combination is not in TCDF or the other cited baselines, and Table 1 plus the fMRI matrix show clear F1 gains on the two nonlinear sets (0.769 Lorenz, 0.837 fMRI). Ablations in Fig. 4 and the small head/block table back the two design choices they claim matter.

What is actually new is the packaging, not a new causal principle. Inverted tokens are already in the forecasting literature, Sparsemax is off-the-shelf, global rank thresholding follows TCDF’s style, and PI is standard post-hoc. They put the pieces together cleanly, write the algorithm out, and beat BGGranger/KGC/tsFCI/TCDF/PCMCI/DYNOTEARS on the four public benchmarks they chose. That is real incremental value for anyone already running attention-based Granger tools.

Soft spots are the usual ones for this genre and should be weighed as such. Nothing in Assumptions 1–4 or Eqs. 5–10 proves that a non-zero Sparsemax weight after summation equals a Granger edge rather than residual predictive association or optimizer artifact; PI only checks the candidates that already survived the free rank-k cut. k itself, heads, and blocks are free parameters with only light sensitivity. No error bars, multi-seed numbers, or code. Circularity is mild—they train for next-step prediction and post-process—so the recovered graph is Granger by construction. None of this sinks the empirical claim; it just keeps the paper inside the normal neural-discovery envelope.

This is for people who need a stronger exploratory tool on high-dimensional nonlinear series (climate, finance, fMRI-style). It will not change theory or give interventional guarantees. I would send it to referees; the method is reproducible on paper, the gains are consistent, and the gaps are fixable with code and stats. Worth a look if you already cite TCDF or DYNOTEARS; skip if you only care about identification guarantees.

Referee Report

5 major / 5 minor

Summary. The paper proposes a transformer-based framework for Granger-style causal discovery in multivariate time series. It introduces an inverted causal self-attention mechanism (CSAM) that treats each full series as a token, computes attention scores, applies Sparsemax for sparsity, aggregates scores via a global causal algorithm (Algorithm 1) that thresholds at the k-th largest strength, and verifies candidates with permutation importance. Experiments on Hénon, FINANCE, Lorenz-96 and fMRI report higher Precision/Recall/F1 than BGGranger, KGC, tsFCI, TCDF, PCMCI and DYNOTEARS, with ablations on n2n prediction and Softmax vs Sparsemax plus a small sensitivity table on heads/blocks.

Significance. If the empirical gains hold under tighter controls, the work is a useful incremental contribution to attention-based causal discovery for nonlinear multivariate time series, extending the line of TCDF-style methods with token inversion, Sparsemax, global aggregation and PI verification. Strengths include a fully specified pipeline (CSAM + Algorithm 1 + PI), consistent outperformance on the two nonlinear benchmarks (F1 0.769 / 0.837), and ablations (Fig. 4) plus sensitivity (Table 2) that support the two main design choices. The result is of practical interest for domains that already rely on Granger graphs, but the theoretical link from sparse inverted attention to true Granger edges remains empirical rather than proven.

major comments (5)

[§4.1] §4.1 (Eqs. 2–3 and surrounding text) contains a load-bearing dimensional inconsistency: after inversion the authors state ˆK, ˆQ, ˆV ∈ R^{T×d}, yet immediately extract the i-th row of ˆQ for each time-series variable i and refer to N as the number of variables. Inverted tokens should yield matrices whose first dimension equals the number of series M, not T. This makes the construction of ei, αi and the causal tensor Cijt difficult to reproduce or verify.
[Table 1] Table 1 reports single-point Precision/Recall/F1 with no error bars, multiple random seeds, or statistical tests. Several margins are small (e.g., Finance F1 0.586 vs PCMCI 0.556; Lorenz recall lower than PCMCI). Without variability estimates it is impossible to judge whether the claimed superiority is stable or an artifact of a single training run.
[§4.2, Algorithm 1] The global causal algorithm (Algorithm 1, Eqs. 8–10) selects edges by a free rank parameter k that sets τ = sorted(Hij)[k]. The paper states that k is chosen “following [10]” but never ablates or reports sensitivity to k, unlike the heads/blocks analysis in Table 2. Because the entire candidate set P (and therefore the final graph after PI) is determined by this discrete threshold, the central empirical claim rests on an untuned hyper-parameter whose effect size is unknown.
[§4.1–4.3] Assumptions 1–4 and the construction in §4.1–4.2 equate non-zero Sparsemax weights on inverted tokens (after summation and rank-k thresholding) with Granger-causal edges. Nothing in the derivation shows that surviving (i→j) pairs satisfy the predictive-improvement definition of Granger causality rather than residual associations or optimization artifacts. PI verification (§4.3) is applied only after selection and cannot recover edges that were never candidates; the attention-to-causality step therefore remains an unproven modeling assumption on which the reported gains depend.
[§3, Assumption 3] Assumption 3 (“if Xi causes Xj then i < j”) imposes a strict total order on variable indices. Variable labels are arbitrary; enforcing i < j either silently renumbers the series or incorrectly forbids legitimate edges. The algorithm itself does not appear to enforce this constraint, creating an internal inconsistency between the stated assumptions and the implemented procedure.

minor comments (5)

[Fig. 3] Fig. 3 caption and color scheme (dark blue = true, red = error) are helpful, but the matrices themselves are small and hard to compare quantitatively; a numeric Hamming or SHD overlay would improve readability.
[Table 2] Table 2 sensitivity is reported only for one dataset (apparently fMRI) and only for heads/blocks; window size, learning-rate schedule and dropout are fixed without comment.
[§1] Several typos and spacing artifacts appear throughout (e.g., “acausalverificationmoduletoensurerobustness”, “intheidentifiedrelationships”, missing spaces after periods in the contributions list).
[§2] Related-work discussion of neural Granger methods is thin; recent continuous-optimization and attention-based baselines beyond TCDF/DYNOTEARS could be acknowledged for completeness.
[§5.1] Input/output window sizes are fixed at 10/1 with no justification relative to the intrinsic time scales of Hénon, Lorenz-96 or fMRI.

Circularity Check

0 steps flagged

No circularity: attention-to-causality pipeline is a heuristic extraction evaluated against external ground truth, not a result forced by its own inputs

full rationale

The paper trains a transformer (with inverted-token CSAM and Sparsemax) for next-step prediction under MSE, then post-processes attention scores via summation, a free rank-k threshold (Alg. 1, Eqs. 8–10), and optional PI verification to produce a candidate Granger graph. That graph is scored by Precision/Recall/F1 against independently known ground-truth structures on Hénon, FINANCE, Lorenz-96 and fMRI. Nothing in the construction equates the output edges to the training loss or to a fitted parameter by definition; the attention-to-edge step is an unproven methodological proxy (shared in spirit with TCDF), not a self-definitional or fitted-input tautology. Citations are to standard external literature (Granger, Sparsemax, PCMCI, NOTEARS, etc.); there is no load-bearing self-citation, uniqueness theorem imported from the authors, or renamed known identity. Free knobs (k, heads, blocks) and the validity of the attention≈causality leap are correctness/sensitivity issues, not circularity. Derivation chain is therefore self-contained and non-circular.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 2 invented entities

The claim rests on four explicit causal assumptions, on the unproved identification of Sparsemax attention with Granger edges, and on several hand-chosen architectural and thresholding parameters. No new physical entities are postulated; the invented objects are algorithmic modules.

free parameters (4)

order parameter k (threshold rank τ) = dataset-dependent; not numerically fixed in the text
Selects how many of the strongest aggregated attention scores become candidate edges; taken from prior work without a data-driven rule.
number of attention heads = 8
Architectural hyper-parameter shown to affect F1 in sensitivity table; default 8 chosen by preliminary tuning.
number of attention blocks/layers = 3
Depth hyper-parameter; default 3 chosen by preliminary tuning.
learning rate / scheduler / dropout / window sizes = lr=0.001, step=10, gamma=0.98, windows 10/1
Training hyper-parameters set by preliminary tuning; affect the attention maps that become the causal scores.

axioms (5)

domain assumption Causal Markov condition: each variable is independent of its non-effects given its direct causes.
Stated as Assumption 1; required for the recovered DAG to be interpretable as a causal graph.
domain assumption Faithfulness: observed conditional independencies reflect the true causal structure.
Assumption 2; standard but untested on the finite noisy series used.
domain assumption Strict causal ordering: if Xi causes Xj then i < j; no contemporaneous causation.
Assumption 3; rules out same-time edges that many real systems possess.
domain assumption Bounded dependencies: true causal edges produce a prediction-error gap at least ε for large T.
Assumption 4; guarantees a minimum signal that the attention/PI pipeline can detect.
ad hoc to paper Non-zero Sparsemax attention after inversion and global sum identifies Granger-causal influence.
Core methodological identification in §4.1–4.2; not derived from the four assumptions, only motivated.

invented entities (2)

Inverted Causal Self-Attention Mechanism (CSAM) no independent evidence
purpose: Compute sparse cross-variable attention scores treated as candidate causal strengths.
New module that inverts tokenization and replaces Softmax with Sparsemax inside a transformer block.
Global causal algorithm (Algorithm 1) no independent evidence
purpose: Aggregate per-token attention into a single ranked influence matrix and threshold it.
Post-processing procedure introduced to turn local attention into a global graph.

pith-pipeline@v1.2.0-daily-grok45 · 14208 in / 3190 out tokens · 65236 ms · 2026-07-31T14:29:49.199270+00:00 · methodology

0 comments

read the original abstract

Causal discovery in multivariate time series data is challenging due to complex interactions, high dimensionality, and nonlinear dependencies among variables. Existing methods often struggle to capture these complexities, resulting in inaccurate causal structures. To address this issue, we propose a novel framework that leverages self-attention mechanisms within the transformer architecture for causal discovery. Our approach introduces a novel inverted causal self-attention mechanism (CSAM) that emphasizes latent and indirect causal relationships by inverting tokens and inducing sparsity in attention scores, focusing on significant causal interactions and reducing spurious correlations. Additionally, we develop a global causal algorithm to identify global causal links, providing a holistic metric for causal influence, along with a causal verification module to ensure robustness in the identified causal relationships, enhancing the reliability of our framework. Experiments on both linear and nonlinear datasets, along with ablation studies and sensitivity analyses, show that our framework outperforms existing methods, demonstrating its potential for causal discovery in complex multivariate time series.

Figures

Figures reproduced from arXiv: 2607.28212 by Huan Huo, Tianqing Zhu, Xiufeng Liu, Yifan Yin, Yong Wang, Yusen Liu.

**Figure 2.** Figure 2: Illustration of the process of our causal self-attention mechanism (CSAM). [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

**Figure 3.** Figure 3: Comparison of causality matrices from various methods on the fMRI [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

**Figure 4.** Figure 4: Ablation study results across datasets. 5.2 Overall Performance The quantitative results in [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗

discussion (0)

Reference graph

Works this paper leans on

25 extracted references

[1]

ICLR (2015)

Bahdanau, D., Cho, K., Bengio, Y.: Neural machine translation by jointly learning to align and translate. ICLR (2015)

2015
[2]

Journal of Climate25(17), 5648–5665 (2012)

Ebert-Uphoff, I., Deng, Y.: Causal discovery for climate research using graphical models. Journal of Climate25(17), 5648–5665 (2012)

2012
[3]

Prob- abilistic graphical models pp

Entner, D., Hoyer, P.O.: On causal discovery from time series data using fci. Prob- abilistic graphical models pp. 121–128 (2010)

2010
[4]

Econometrica: journal of the Econometric Society pp

Granger, C.W.: Investigating causal relations by econometric models and cross- spectral methods. Econometrica: journal of the Econometric Society pp. 424–438 (1969)

1969
[5]

Cambridge University Press (2013)

Kleinberg, S.: Causality, probability, and time. Cambridge University Press (2013)

2013
[6]

Physical Review E87(6), 062918 (2013)

Kugiumtzis, D.: Direct-coupling information measure from nonuniform embedding. Physical Review E87(6), 062918 (2013)

2013
[7]

In: Proc

Lorenz, E.N.: Predictability: A problem partly solved. In: Proc. of Seminar on predictability. vol. 1. Reading (1996)

1996
[8]

Physical review letters100(14), 144103 (2008)

Marinazzo, D., Pellicoro, M., Stramaglia, S.: Kernel method for nonlinear granger causality. Physical review letters100(14), 144103 (2008)

2008
[9]

In: ICML

Martins, A., Astudillo, R.: From softmax to sparsemax: A sparse model of attention and multi-label classification. In: ICML. pp. 1614–1623. PMLR (2016)

2016
[10]

Machine Learning and Knowledge Extraction1(1) (2019)

Nauta, M., Bucur, D., Seifert, C.: Causal discovery with attention-based convolu- tional neural networks. Machine Learning and Knowledge Extraction1(1) (2019)

2019
[11]

In: AISTATS

Pamfil, R., Sriwattanaworachai, N., Desai, S., Pilgerstorfer, P., Georgatzis, K., Beaumont, P., Aragam, B.: Dynotears: Structure learning from time-series data. In: AISTATS. pp. 1595–1605. Pmlr (2020)

2020
[12]

In: Proc

Pereira, J.P., Stroes, E.S., Zwinderman, A.H., Levin, E.: Covered information dis- entanglement: model transparency via unbiased permutation importance. In: Proc. of the AAAI Conference on Artificial Intelligence. vol. 36, pp. 7984–7992 (2022)

2022
[13]

Chaos: An Interdisciplinary Journal of Nonlin- ear Science28(7) (2018)

Runge, J.: Causal network reconstruction from time series: From theoretical as- sumptions to practical estimation. Chaos: An Interdisciplinary Journal of Nonlin- ear Science28(7) (2018)

2018
[14]

Nature communications10(1), 2553 (2019)

Runge, J., Bathiany, S., Bollt, E., Camps-Valls, G., Coumou, D., Deyle, E., Gly- mour, C., Kretschmer, M., Mahecha, M.D., Muñoz-Marí, J., et al.: Inferring cau- sation from time series in earth system sciences. Nature communications10(1), 2553 (2019)

2019
[15]

Science advances5(11), eaau4996 (2019)

Runge, J., Nowack, P., Kretschmer, M., Flaxman, S., Sejdinovic, D.: Detecting and quantifying causal associations in large nonlinear time series datasets. Science advances5(11), eaau4996 (2019)

2019
[16]

Physical review letters85(2) (2000)

Schreiber, T.: Measuring information transfer. Physical review letters85(2) (2000)

2000
[17]

Machine Learning108, 1421–1441 (2019)

Shih, S.Y., Sun, F.K., Lee, H.y.: Temporal pattern attention for multivariate time series forecasting. Machine Learning108, 1421–1441 (2019)

2019
[18]

Social science computer review9(1), 62–72 (1991)

Spirtes, P., Glymour, C.: An algorithm for fast recovery of sparse causal graphs. Social science computer review9(1), 62–72 (1991)

1991
[19]

IEEE TPAMI44(8), 4267–4279 (2021)

Tank, A., Covert, I., Foti, N., Shojaie, A., Fox, E.B.: Neural granger causality. IEEE TPAMI44(8), 4267–4279 (2021)

2021
[20]

NeurIPS30(2017)

Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez,A.N.,Kaiser, Ł., Polosukhin, I.: Attention is all you need. NeurIPS30(2017)

2017
[21]

Nature communications11(1), 5046 (2020) Causal Discovery with Inverted Self-attention for Multivariate Time Series 13

Yan, Y., Dahmani, L., Ren, J., Shen, L., Peng, X., Wang, R., He, C., Jiang, C., Gong, C., Tian, Y., et al.: Reconstructing lost bold signal in individual participants using deep machine learning. Nature communications11(1), 5046 (2020) Causal Discovery with Inverted Self-attention for Multivariate Time Series 13

2020
[22]

Journal of Economic Dynamics and Control139, 104434 (2022)

Zema, S.M.: Directed acyclic graph based information shares for price discovery. Journal of Economic Dynamics and Control139, 104434 (2022)

2022
[23]

In: ICML (2022)

Zhang, Y., Yan, J.: Crossformer: Transformer utilizing cross-dimension dependency for multivariate time series forecasting. In: ICML (2022)

2022
[24]

NeurIPS31(2018)

Zheng, X., Aragam, B., Ravikumar, P.K., Xing, E.P.: Dags with no tears: Contin- uous optimization for structure learning. NeurIPS31(2018)

2018
[25]

In: Proc

Zhou, H., Zhang, S., Peng, J., Zhang, S., Li, J., Xiong, H., Zhang, W.: Informer: Beyond efficient transformer for long sequence time-series forecasting. In: Proc. of the AAAI conference on artificial intelligence. vol. 35, pp. 11106–11115 (2021)

2021

[1] [1]

ICLR (2015)

Bahdanau, D., Cho, K., Bengio, Y.: Neural machine translation by jointly learning to align and translate. ICLR (2015)

2015

[2] [2]

Journal of Climate25(17), 5648–5665 (2012)

Ebert-Uphoff, I., Deng, Y.: Causal discovery for climate research using graphical models. Journal of Climate25(17), 5648–5665 (2012)

2012

[3] [3]

Prob- abilistic graphical models pp

Entner, D., Hoyer, P.O.: On causal discovery from time series data using fci. Prob- abilistic graphical models pp. 121–128 (2010)

2010

[4] [4]

Econometrica: journal of the Econometric Society pp

Granger, C.W.: Investigating causal relations by econometric models and cross- spectral methods. Econometrica: journal of the Econometric Society pp. 424–438 (1969)

1969

[5] [5]

Cambridge University Press (2013)

Kleinberg, S.: Causality, probability, and time. Cambridge University Press (2013)

2013

[6] [6]

Physical Review E87(6), 062918 (2013)

Kugiumtzis, D.: Direct-coupling information measure from nonuniform embedding. Physical Review E87(6), 062918 (2013)

2013

[7] [7]

In: Proc

Lorenz, E.N.: Predictability: A problem partly solved. In: Proc. of Seminar on predictability. vol. 1. Reading (1996)

1996

[8] [8]

Physical review letters100(14), 144103 (2008)

Marinazzo, D., Pellicoro, M., Stramaglia, S.: Kernel method for nonlinear granger causality. Physical review letters100(14), 144103 (2008)

2008

[9] [9]

In: ICML

Martins, A., Astudillo, R.: From softmax to sparsemax: A sparse model of attention and multi-label classification. In: ICML. pp. 1614–1623. PMLR (2016)

2016

[10] [10]

Machine Learning and Knowledge Extraction1(1) (2019)

Nauta, M., Bucur, D., Seifert, C.: Causal discovery with attention-based convolu- tional neural networks. Machine Learning and Knowledge Extraction1(1) (2019)

2019

[11] [11]

In: AISTATS

Pamfil, R., Sriwattanaworachai, N., Desai, S., Pilgerstorfer, P., Georgatzis, K., Beaumont, P., Aragam, B.: Dynotears: Structure learning from time-series data. In: AISTATS. pp. 1595–1605. Pmlr (2020)

2020

[12] [12]

In: Proc

Pereira, J.P., Stroes, E.S., Zwinderman, A.H., Levin, E.: Covered information dis- entanglement: model transparency via unbiased permutation importance. In: Proc. of the AAAI Conference on Artificial Intelligence. vol. 36, pp. 7984–7992 (2022)

2022

[13] [13]

Chaos: An Interdisciplinary Journal of Nonlin- ear Science28(7) (2018)

Runge, J.: Causal network reconstruction from time series: From theoretical as- sumptions to practical estimation. Chaos: An Interdisciplinary Journal of Nonlin- ear Science28(7) (2018)

2018

[14] [14]

Nature communications10(1), 2553 (2019)

Runge, J., Bathiany, S., Bollt, E., Camps-Valls, G., Coumou, D., Deyle, E., Gly- mour, C., Kretschmer, M., Mahecha, M.D., Muñoz-Marí, J., et al.: Inferring cau- sation from time series in earth system sciences. Nature communications10(1), 2553 (2019)

2019

[15] [15]

Science advances5(11), eaau4996 (2019)

Runge, J., Nowack, P., Kretschmer, M., Flaxman, S., Sejdinovic, D.: Detecting and quantifying causal associations in large nonlinear time series datasets. Science advances5(11), eaau4996 (2019)

2019

[16] [16]

Physical review letters85(2) (2000)

Schreiber, T.: Measuring information transfer. Physical review letters85(2) (2000)

2000

[17] [17]

Machine Learning108, 1421–1441 (2019)

Shih, S.Y., Sun, F.K., Lee, H.y.: Temporal pattern attention for multivariate time series forecasting. Machine Learning108, 1421–1441 (2019)

2019

[18] [18]

Social science computer review9(1), 62–72 (1991)

Spirtes, P., Glymour, C.: An algorithm for fast recovery of sparse causal graphs. Social science computer review9(1), 62–72 (1991)

1991

[19] [19]

IEEE TPAMI44(8), 4267–4279 (2021)

Tank, A., Covert, I., Foti, N., Shojaie, A., Fox, E.B.: Neural granger causality. IEEE TPAMI44(8), 4267–4279 (2021)

2021

[20] [20]

NeurIPS30(2017)

Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez,A.N.,Kaiser, Ł., Polosukhin, I.: Attention is all you need. NeurIPS30(2017)

2017

[21] [21]

Nature communications11(1), 5046 (2020) Causal Discovery with Inverted Self-attention for Multivariate Time Series 13

Yan, Y., Dahmani, L., Ren, J., Shen, L., Peng, X., Wang, R., He, C., Jiang, C., Gong, C., Tian, Y., et al.: Reconstructing lost bold signal in individual participants using deep machine learning. Nature communications11(1), 5046 (2020) Causal Discovery with Inverted Self-attention for Multivariate Time Series 13

2020

[22] [22]

Journal of Economic Dynamics and Control139, 104434 (2022)

Zema, S.M.: Directed acyclic graph based information shares for price discovery. Journal of Economic Dynamics and Control139, 104434 (2022)

2022

[23] [23]

In: ICML (2022)

Zhang, Y., Yan, J.: Crossformer: Transformer utilizing cross-dimension dependency for multivariate time series forecasting. In: ICML (2022)

2022

[24] [24]

NeurIPS31(2018)

Zheng, X., Aragam, B., Ravikumar, P.K., Xing, E.P.: Dags with no tears: Contin- uous optimization for structure learning. NeurIPS31(2018)

2018

[25] [25]

In: Proc

Zhou, H., Zhang, S., Peng, J., Zhang, S., Li, J., Xiong, H., Zhang, W.: Informer: Beyond efficient transformer for long sequence time-series forecasting. In: Proc. of the AAAI conference on artificial intelligence. vol. 35, pp. 11106–11115 (2021)

2021