Pith. sign in

REVIEW 4 major objections 4 minor 34 references

Causal Representation Learning from Network Data

T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read A graph-aware VAE inherits causal identifiability and improves CRISPR intervention prediction.

desk verdict GRACE-VAE is a useful empirical architecture for incorporating network context into causal representation learning, but its identifiability proof rests on a false algebraic claim, and the headline empirical claim overstates a baseline comparison. read the letter →

arxiv 2509.01916 v2 pith:QG257I7O submitted 2025-09-02 cs.LG

classification cs.LG
keywords causalrepresentationlearninglatentgraphneuralnetworkvariationalautoencodersoftinterventionsidentifiabilityCRISPRperturbationpredictionnetwork-structureddata
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

GRACE-VAE is a graph-aware causal discrepancy variational autoencoder for learning latent causal structure from data that come with an observed interaction network, such as pathway and protein-protein associations around gene-expression measurements. The paper's central claim is that inserting a graph neural network into the variational encoder—while leaving the generative decoder as a latent structural causal model—preserves the existing identifiability guarantee: from observational and single-node soft-interventional data, the latent DAG and intervention targets are recoverable up to permutation, scaling, and shift. The practical motivation is CRISPR perturbation screens, where structured biological context improves prediction of gene-expression outcomes after interventions, including double perturbations never seen in training. A sympathetic reader would care because most causal representation learning assumes flat i.i.d. data, whereas real scientific datasets are relational; this work argues the relational structure can be exploited without sacrificing theoretical guarantees.

What carries the argument

The load-bearing design choice is the placement of the graph neural network: GRACE-VAE inserts the GNN in the variational encoder only, so the generative likelihood p(X|U) and the latent SCM over U are identical to the model in Zhang et al. (2023). Identifiability then follows from the existing three-step proof: Lemma 1 identifies the latent vector up to an affine map from the observational distribution; Assumption 2 (linear interventional faithfulness) uses the changed marginal distributions under each intervention to pin down the coordinate permutation; Assumption 3 (total separation) orients every edge of the DAG. The remaining freedom is the CD-equivalence class. The intervention encoder

What would settle it

Run GRACE-VAE on synthetic data generated from a known latent SCM using an MLP decoder with leaky ReLU activations and the same training loss; if the recovered DAG and intervention targets are not equivalent up to permutation, scaling, and shift, the implemented model does not inherit the identifiability guarantee.

Watch

Extended reading notes

Core claim

The paper's central claim is that causal representation learning from soft interventions remains identifiable when the observed data carry network structure, provided the graph information is injected only into the variational encoder. GRACE-VAE instantiates this idea: a GraphSAGE (or other GNN) encoder processes the observed entities and their auxiliary group nodes; the decoder is an unchanged latent structural causal model with a soft intervention encoder; training couples an ELBO reconstruction term with an MMD alignment term that matches generated counterfactuals to observed interventional distributions. Because the GNN never enters the likelihood, the identifiability theorem from Zhang

Load-bearing premise

The identifiability guarantee holds only if the decoder is a polynomial map with full rank and non-empty latent support, and the intervention data satisfy the paper's faithfulness and separation conditions; the implemented MLP decoder and the real biological datasets are not verified against these conditions.

Editorial extensions

If this is right

  • If the claim holds, any observed network with observational plus soft-interventional data can be used to recover latent causal variables and intervention targets without knowing those targets, up to relabeling and scaling.
  • Structured context yields measurable gains: on NORMAN, single-intervention R² rises from 0.902 with the flat baseline to 0.945 with GRACE-VAE, and double-intervention R² rises from 0.738 to 0.785.
  • Ablations show that both components matter: removing the causal decoder or removing the graph encoder degrades performance, so the benefit comes from combining network-aware inference with an explicit causal generative model.
  • The identifiability guarantee is insensitive to the choice of GNN encoder, so future encoder architectures can be swapped in without re-proving identifiability.
  • Richer pathway and gene topology helps single interventions but shows diminishing or slightly negative returns for unseen double interventions, suggesting context level should be chosen per task.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An implication left implicit is that, since the guarantee concerns only the generative side, the GNN could be replaced by any sufficiently expressive inference network over the network and the identifiability result would still hold.
  • The paper does not verify that real CRISPR data satisfy the faithfulness and separation assumptions or that the implemented MLP decoder is a polynomial map; a synthetic benchmark with known ground-truth SCM and the exact architecture would reveal whether the implemented model's recoveries are actually CD-equivalent.
  • A testable extension suggested by the framing: adding graph context that is causally irrelevant (for example, permuted pathway assignments) should not improve intervention prediction; if it does, the gain is likely regularization rather than recovered causal structure.
  • The same architecture should transfer to other domains with latent interventions and observed relational structure, such as social influence, ecology, or drug combination screens, where network context could act as weak supervision for latent mechanisms.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes GRACE-VAE, a variational autoencoder that injects structured biological network context (pathway-gene, pathway-pathway, and protein-protein interactions) through a GNN encoder, while keeping a causal SCM-based decoder. It claims that, because the GNN is placed only in the variational encoder, the model inherits the identifiability guarantees of causal discrepancy VAEs and can recover the latent causal graph and intervention targets up to CD-equivalence. The authors evaluate GRACE-VAE on three CRISPR perturbation datasets, including unseen double interventions, and compare against CMVAE, CMVAE-multihot, and VGAE, with ablations over GNN architecture and context levels. The central claims are: (1) theoretical identifiability of the latent causal structure, and (2) empirical improvement in interventional outcome prediction from using structured context.

Significance. If the theoretical claim were correct and the empirical improvements robust, this would be a useful contribution to causal representation learning in structured biological settings. The empirical study is fairly comprehensive: three datasets, 10 seeds, multiple baselines, ablations, and hyperparameter reporting. The idea of using a GNN encoder to provide structural context while leaving the causal decoder unchanged is principled and could be practically valuable. However, the theoretical guarantee as presented is not established: the appendix contains a false lemma, and the implemented model does not satisfy the polynomial-decoder assumption of the theorem. The empirical claim that GRACE-VAE outperforms all baselines on R2 is also contradicted by the paper's own Table 1. The core idea has potential, but the paper needs substantial correction before the claims are supportable.

major comments (4)
  1. [Appendix, Lemma 1 / Theorem 1] Lemma 1 is false as stated, and its proof collapses. The proof's final step invokes a 'fundamental theorem of algebra for polynomial automorphisms' to conclude that two mutually inverse polynomial maps must have degree 1. That theorem is false: F(x,y)=(x+y^2,y) is a polynomial automorphism of R^2 with polynomial inverse (x-y^2,y), is full-rank everywhere, and is not affine. Concretely, take f=identity and \f=F^{-1}, and \U=F(U). Then \f(\U)=U, so the observational distribution is identical, but \U is not ΛU+b. Since Theorem 1's proof begins with Lemma 1 to establish an affine reparameterization, the CD-equivalence theorem is not established as written. The authors must either prove a corrected lemma under additional hypotheses or explicitly defer to Zhang et al. (2023) and verify that all of that paper's assumptions hold for the present setting.
  2. [Identifiability guarantee / Implementation details] Assumption 1 assumes a full-row-rank polynomial decoder, but the implemented decoder is a DAG layer plus two fully connected layers with leaky ReLU activations, which is not a polynomial map. Assumptions 2 and 3 (linear interventional faithfulness and total separation) are also never verified on the CRISPR datasets. Therefore the statement that 'GRACE-VAE inherits the identifiability guarantee and identifies the latent causal graph and intervention targets up to CD-equivalence' does not apply to the architecture and data actually evaluated. This is a load-bearing gap between theory and experiment.
  3. [Results, R1 / Table 1] The text states: 'Across the three datasets, GRACE-VAE outperforms all baselines on R2 (Table 1).' This is contradicted by Table 1: on NORMAN single-intervention R2, VGAE achieves 0.9493±0.0123 versus GRACE-VAE's 0.9452±0.0108. The claim needs to be qualified: GRACE-VAE outperforms CMVAE and CMVAE-multihot on this metric, and it has the best or comparable RMSE/MMD, but it is not strictly the best on single R2. This overstatement affects the paper's primary empirical contribution.
  4. [Conclusion and Experimental Setup] The conclusion claims that 'GRACE-VAE yields faithful latent DAGs' and that the experiments demonstrate recovery of the true latent causal graph. However, no structure-learning metric (e.g., SHD, edge AUROC, target recovery accuracy) is reported against a ground-truth latent DAG. The three datasets are real perturbation screens without known latent SCMs, and the reported metrics evaluate prediction of interventional outcomes only. To support the 'recovery' claim, the authors should include synthetic experiments with known latent graphs and report structural accuracy.
minor comments (4)
  1. [Identifiability guarantee] There is a typo in the heading ('guanrantee') and in the text: 'leaves the observational distribution PX changed' should read 'leaves the observational distribution PX unchanged' (or 'does not change').
  2. [References] The citation Fine and Rosenberger (1997) is a textbook on the fundamental theorem of algebra; it does not support the invoked statement about polynomial automorphisms. If the authors believe such a theorem exists, they should cite a proper source and state the exact conditions.
  3. [Implementation details] The hyperparameter selection is described as based on 'R2, MMD, and RMSE' collectively, which is ambiguous. Please specify the selection rule (e.g., a primary metric or a rank-based aggregate).
  4. [Figure 2] The learned DAG in Figure 2 would benefit from a description of how edges are thresholded (if at all) and how edge weights are normalized, so that the visualization is reproducible.

Circularity Check

2 steps flagged · score 6.0 of 10

Identifiability guarantee is imported from the authors' own prior work via a self-citation chain, and the appendix's re-derivation relies on a false theorem about polynomial automorphisms; the central theoretical claim is not independently established.

  1. uniqueness imported from authors [Identifiability guarantee (main text) and Appendix, Theorem 1 proof]
    "In our setting, the identifiability guarantee still holds under assumptions 1, 2, and 3 in Zhang et al. (2023). G RACE-VAE enriches inference but leaves both the likelihood and the latent causal structure unchanged, so the original identifiability guarantee remains valid."

    The paper's central theoretical claim is not derived in this paper; it is stated to 'inherit' the identifiability theorem of Zhang et al. (2023), whose first author is also the present paper's first author. The appendix's proof of Theorem 1 then defers the permutation-pinning and edge-orientation steps to 'Lemmas 5, 6, 7, and 8 in Zhang et al. 2023' and 'Theorem 2 (Zhang et al. 2023)', so the uniqueness conclusion is imported wholesale from the same research group. This is load-bearing: if the cited theorem or its assumptions fail, the paper's guarantee has no independent support. The implementation's decoder (DAG layer + MLP with leaky ReLU) is never checked against Assumption 1's polynomial full-row-rank requirement, so the import cannot be validated independently here.

  2. other [Appendix: Identifiability guarantee, Lemma 1 proof]
    "By the fundamental theorem of algebra for polynomial automorphisms (Fine and Rosenberger 1997), poly1 and poly2 must each be of degree 1."

    This step is the hinge of Lemma 1: it turns 'bU is a polynomial in U and vice versa' into the affine conclusion bU = Lambda U + b. The cited 'fundamental theorem of algebra for polynomial automorphisms' is false; F(x,y)=(x+y^2, y) is a polynomial automorphism with polynomial inverse (x-y^2, y) and is not degree 1. Therefore the lemma's conclusion is not a consequence of Assumption 1. The proof effectively assumes the degree-1 outcome it needs to establish, so the identifiability guarantee reduces to an unstated and false premise rather than to a proved theorem.

full rationale

The paper's central theoretical contribution is the claim that GraCE-VAE inherits CD-identifiability for latent causal graphs and intervention targets. The appendix proof is not self-contained: it explicitly defers to Lemmas 5-8 and Theorem 2 of Zhang et al. (2023), whose first author overlaps with the present paper. The main text states the guarantee holds 'under assumptions 1, 2, and 3 in Zhang et al. (2023)' and that 'the original identifiability guarantee remains valid,' making the theoretical core a load-bearing self-citation. Moreover, the appendix's Lemma 1 proof invokes a false 'fundamental theorem of algebra for polynomial automorphisms' to conclude that polynomial inverses must be degree 1; the counterexample F(x,y)=(x+y^2,y) shows the step is invalid. Consequently, the affine identifiability lemma, and hence Theorem 1, is not established by the paper's own argument. The paper also does not verify that the implemented decoder (DAG layer plus MLP with leaky ReLU) satisfies Assumption 1's polynomial full-row-rank requirement, so even the imported theorem may not apply to the architecture actually evaluated. Separately, Table 1 shows VGAE's NORMAN single R2 (0.9493) exceeds GraCE-VAE's (0.9452), contradicting the text's 'outperforms all baselines on R2'; this is an empirical overclaim rather than a circularity, but it further weakens support. The empirical double-intervention experiments are independent and not circular, but the central theoretical claim reduces to a self-citation chain and a defective proof step, warranting a score of 6.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The paper introduces no new physical entities. The latent causal programs U are a standard latent SCM construct inherited from CMVAE rather than a newly postulated ontological object. The main free parameters are the latent dimension and the loss/capacity hyperparameters chosen by validation. The axioms are almost all inherited from Zhang et al. (2023), and one mathematical claim used in the appendix is false.

free parameters (6)
  • Latent dimension p = 105 (NORMAN), 414 (REPLOGLE-SMALL), 300 (REPLOGLE-LARGE)
    Set to the number of distinct intervention targets for NORMAN and REPLOGLE-SMALL, and grid-searched over {100,200,300} for REPLOGLE-LARGE. Determines the capacity and structure of the latent causal graph.
  • MMD alignment coefficient alpha_max = 8, 10, 12, or 15 depending on dataset/model
    Tuned on validation data; controls the strength of interventional distribution alignment in the loss.
  • KL coefficient beta_max = 2, 3, or 4 depending on dataset/model
    Tuned on validation data; controls the strength of the prior regularizer.
  • DAG sparsity penalty lambda = 0.0001 or 0.001
    Tuned on validation data; promotes sparsity of the learned latent adjacency matrix M_theta.
  • Softmax temperature Temp = 4, 5, or 6 depending on dataset/model
    Tuned on validation data; controls the sharpness of the intervention encoder's soft one-hot target assignment.
  • GNN architecture and layer count = 1-layer GraphSAGE chosen; 1 or 3 layers in ablations
    Selected by performance on validation; deeper GNNs did not improve results, so layer count is an empirical choice.
assumptions (6)
  • domain assumption Assumption 1: full-row-rank polynomial decoder f with non-empty interior support of PU
    Imported from Zhang et al. (2023). The experiments use an MLP decoder with leaky ReLU, which is not globally polynomial, so this assumption is not verified for the implemented model.
  • domain assumption Assumption 2: linear interventional faithfulness
    Imported from Zhang et al. (2023). No empirical check is performed on the CRISPR datasets; the paper assumes it holds.
  • domain assumption Assumption 3: total separation for every edge
    Imported from Zhang et al. (2023). This is a strong structural condition and is not verified.
  • domain assumption Faithfulness: latent graph G and latent distribution PU are faithful to each other
    Stated in Problem Setup; needed to map conditional independences in PU to d-separations in G.
  • domain assumption Each latent variable is targeted by at least one intervention
    Stated in Problem Setup, citing Squires et al. (2023). Needed for identifiability of intervention targets.
  • standard math Polynomial automorphisms that compose to the identity must have degree 1
    Invoked in the proof of Lemma 1. This is false for p>=2, e.g., (x, y+x^2) is an invertible polynomial map of degree 2, so the proof as written is invalid.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Causal Representation Learning from Network Data." pith.science (2026). https://pith.science/paper/QG257I7O

@misc{pith2026250901916,
  author       = {Pith},
  title        = {Pith review of: Causal Representation Learning from Network Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QG257I7O}},
  note         = {Machine review of arXiv:2509.01916}
}
read the original abstract

Causal disentanglement from soft interventions is identifiable under the assumptions of linear interventional faithfulness and availability of both observational and interventional data. Prior work has focused on unstructured observations without leveraging known relational context among measured entities. In many scientific applications, however, the measured variables come with an observed interaction network that provides structured context, such as protein-protein interactions and pathway-gene membership. We propose GraCE-VAE, a graph-aware causal discrepancy variational autoencoder that treats pathway-level information as an auxiliary view of the latent causal programs. The graph neural network encoder conditions on this auxiliary pathway view and the biological graph to improve amortized inference, while the causal decoder remains a latent SCM with soft interventions. Assuming samples are i.i.d. within each intervention regime, we show that GraCE-VAE inherits the identifiability guarantees of causal discrepancy VAEs and identifies the latent causal graph and intervention targets up to the standard equivalence class. Experiments on three CRISPR perturbation datasets demonstrate that leveraging structured biological context improves prediction of interventional outcomes, including unseen perturbation combinations.

Figures

Figures reproduced from arXiv: 2509.01916 by the authors.

Figure 1
Figure 1. Overview of GRACE-VAE’s architecture. GRACE-VAE consists of a graph-aware encoder, intervention encoder, and a causal decoder. Given entities X, H, adjacency structure A, and intervention I, GRACE-VAE learns a causal directed acyclic graph of the latent variables U to generate observational Xˆ and interventional XˆI data. GRACE-VAE is flexible with any GNN layer propagate neural messages over G. Here, we instantiate… view at source ↗
Figure 2
Figure 2. DAG of the learned latent program by GRACE￾VAE for NORMAN. Each node is annotated with its most representative gene and edge weights correspond to the ab￾solute values of the learned DAG coefficients. GRACE-VAE avoids parameter inflation introduced by large multihot encodings (Appendix [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. A detailed visualization of the latent program for [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Comparison of generated sample distributions under all single and all double interventions. We visualize the generated [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Comparison of generated samples (yellow) versus actual (or ground-truth) observed samples (blue) under single [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Comparison of generated samples (yellow) versus actual (or ground-truth) observed samples (blue) under single [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Comparison of generated samples (yellow) versus actual (or ground-truth) observed samples (blue) under double [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: Comparison of generated samples (yellow) versus actual (or ground-truth) observed samples (blue) under double [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 25 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    A.; Tang, H.; Wu, G.; Marjan, S.; and Hussain, A

    Bhatti, U. A.; Tang, H.; Wu, G.; Marjan, S.; and Hussain, A. 2023. Deep learning with graph convolutional networks: An overview and latest applications in computational intelligence. International Journal of Intelligent Systems, 2023(1): 8342104

  4. [4]

    Brehmer, J.; De Haan, P.; Lippe, P.; and Cohen, T. S. 2022. Weakly supervised causal representation learning. Advances in Neural Information Processing Systems, 35: 38319--38331

  5. [5]

    Buchholz, S.; Rajendran, G.; Rosenfeld, E.; Aragam, B.; Sch \"o lkopf, B.; and Ravikumar, P. 2024. Learning linear causal representations from interventions under general nonlinear mixing. Advances in Neural Information Processing Systems, 36

  6. [6]

    Chickering, D. M. 2002. Optimal structure identification with greedy search. Journal of machine learning research, 3(Nov): 507--554

  7. [7]

    Cussens, J. 2020. GOBNILP: Learning Bayesian network structure with integer programming. In International Conference on Probabilistic Graphical Models, 605--608. PMLR

  8. [8]

    Feng, J.; Zhang, L.; and Yang, L. 2023. Concept-free causal disentanglement with variational graph auto-encoder. arXiv preprint arXiv:2311.10638

Show all 34 references
  1. [9]

    Fine, B.; and Rosenberger, G. 1997. The fundamental theorem of algebra. Springer Science & Business Media

  2. [10]

    M.; Rasch, M

    Gretton, A.; Borgwardt, K. M.; Rasch, M. J.; Sch \"o lkopf, B.; and Smola, A. 2012. A kernel two-sample test. The journal of machine learning research, 13(1): 723--773

  3. [11]

    Hamilton, W.; Ying, Z.; and Leskovec, J. 2017. Inductive representation learning on large graphs. Advances in neural information processing systems, 30

  4. [12]

    Higgins, I.; Amos, D.; Pfau, D.; Racaniere, S.; Matthey, L.; Rezende, D.; and Lerchner, A. 2018. Towards a definition of disentangled representations. arXiv preprint arXiv:1812.02230

  5. [13]

    P.; Glorot, X.; Botvinick, M

    Higgins, I.; Matthey, L.; Pal, A.; Burgess, C. P.; Glorot, X.; Botvinick, M. M.; Mohamed, S.; and Lerchner, A. 2017. beta-vae: Learning basic visual concepts with a constrained variational framework. ICLR (Poster), 3

  6. [14]

    Jassal, B.; Matthews, L.; Viteri, G.; Gong, C.; Lorente, P.; Fabregat, A.; Sidiropoulos, K.; Cook, J.; Gillespie, M.; Haw, R.; et al. 2020. The reactome pathway knowledgebase. Nucleic acids research, 48(D1): D498--D503

  7. [15]

    Khemakhem, I.; Monti, R.; Leech, R.; and Hyvarinen, A. 2021. Causal autoregressive flows. In International conference on artificial intelligence and statistics, 3520--3528. PMLR

  8. [16]

    N.; and Welling, M

    Kipf, T. N.; and Welling, M. 2016. Variational graph auto-encoders. arXiv preprint arXiv:1611.07308

  9. [17]

    Kong, L.; Huang, B.; Xie, F.; Xing, E.; Chi, Y.; and Zhang, K. 2023. Identification of nonlinear latent hierarchical models. Advances in Neural Information Processing Systems, 36: 2010--2032

  10. [18]

    Kumar, A.; Sattigeri, P.; and Balakrishnan, A. 2017. Variational inference of disentangled latent concepts from unlabeled observations. arXiv preprint arXiv:1711.00848

  11. [19]

    M.; Huang, Y.; Sumathipala, M.; Liang, M

    Li, M. M.; Huang, Y.; Sumathipala, M.; Liang, M. Q.; Valdeolivas, A.; Ananthakrishnan, A. N.; Liao, K.; Marbach, D.; and Zitnik, M. 2024. Contextual AI models for single-cell protein biology. Nature Methods, 21(8): 1546--1557

  12. [20]

    M.; Cohen, T.; and Gavves, S

    Lippe, P.; Magliacane, S.; L \"o we, S.; Asano, Y. M.; Cohen, T.; and Gavves, S. 2022. Citris: Causal identifiability from temporal intervened sequences. In International Conference on Machine Learning, 13557--13603. PMLR

  13. [21]

    Massidda, R.; Geiger, A.; Icard, T.; and Bacciu, D. 2023. Causal abstraction with soft interventions. In Conference on Causal Learning and Reasoning, 68--87. PMLR

  14. [22]

    Pearl, J. 2000. Causality: Models, Reasoning, and Inference. Cambridge University Press

  15. [23]

    Raskutti, G.; and Uhler, C. 2018. Learning directed acyclic graph models based on sparsest permutations. Stat, 7(1): e183

  16. [24]

    M.; Saunders, R

    Replogle, J. M.; Saunders, R. A.; Pogson, A. N.; Hussmann, J. A.; Lenail, A.; Guna, A.; Mascibroda, L.; Wagner, E. J.; Adelman, K.; Lithwick-Yanai, G.; et al. 2022. Mapping information-rich genotype-phenotype landscapes with genome-scale Perturb-seq. Cell, 185(14): 2559--2575

  17. [25]

    Sanchez-Martin, P.; Rateike, M.; and Valera, I. 2021. Vaca: Design of variational graph autoencoders for interventional and counterfactual queries. arXiv preprint arXiv:2110.14690

  18. [26]

    Solus, L.; Wang, Y.; and Uhler, C. 2021. Consistency guarantees for greedy permutation-based causal inference algorithms. Biometrika, 108(4): 795--814

  19. [27]

    S.; and Uhler, C

    Squires, C.; Seigal, A.; Bhate, S. S.; and Uhler, C. 2023. Linear causal disentanglement via interventions. In International Conference on Machine Learning, 32540--32560. PMLR

  20. [28]

    Suter, R.; Miladinovic, D.; Sch \"o lkopf, B.; and Bauer, S. 2019. Robustly disentangled causal mechanisms: Validating deep representations for interventional robustness. In International Conference on Machine Learning, 6056--6065. PMLR

  21. [29]

    Tao, H.; Yu, H.; and Li, J. 2024. DeepITE: Designing Variational Graph Autoencoders for Intervention Target Estimation. Advances in Neural Information Processing Systems, 37: 89978--90008

  22. [30]

    Varici, B.; Acarturk, E.; Shanmugam, K.; Kumar, A.; and Tajer, A. 2023. Score-based causal representation learning with interventions. arXiv preprint arXiv:2301.08230

  23. [31]

    Veli c kovi \'c , P.; Cucurull, G.; Casanova, A.; Romero, A.; Lio, P.; and Bengio, Y. 2017. Graph attention networks. arXiv preprint arXiv:1710.10903

  24. [32]

    Yang, M.; Liu, F.; Chen, Z.; Shen, X.; Hao, J.; and Wang, J. 2021. Causalvae: Disentangled representation learning via neural structural causal models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 9593--9602

  25. [33]

    Zhang, J.; Greenewald, K.; Squires, C.; Srivastava, A.; Shanmugam, K.; and Uhler, C. 2023. Identifiability guarantees for causal disentanglement from soft interventions. Advances in Neural Information Processing Systems, 36: 50254--50292

  26. [34]

    Zhu, X.; Xu, C.; and Tao, D. 2021. Commutative lie group vae for disentanglement learning. In International Conference on Machine Learning, 12924--12934. PMLR

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.