Disentangled Graph Autoencoder for Treatment Effect Estimation

Chuanhou Gao; Di Fan; Renlei Jiang; Yunhao Wen

REVIEW 5 major objections 6 minor 53 references

Disentangled Graph Autoencoder for Treatment Effect Estimation

T0 review · 5 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A graph autoencoder that splits hidden causes into four roles reports the most accurate individual treatment effect estimates on networked benchmarks.

desk verdict A coherent, well-written incremental method whose headline claim about recovering hidden factors from proxies is not actually tested by its own synthetic experiments; still deserves a serious referee. read the letter →

arxiv 2412.14497 v2 pith:POXFCL7F submitted 2024-12-19 cs.LG cs.AIstat.ML

Di Fan , Renlei Jiang , Yunhao Wen , Chuanhou Gao This is my paper · ORCID

classification cs.LGcs.AIstat.ML

keywords individualtreatmenteffectnetworkedobservationaldatalatentconfoundersdisentangledrepresentationvariationalgraphautoencoderHilbert-Schmidtindependencecriterioncausalinference

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

This paper claims that individual treatment effects can be estimated more accurately from networked observational data when the hidden causes of treatment and outcome are separated into four roles rather than treated as one undifferentiated confounder. On a network where each unit has covariates, a binary treatment, an outcome, and edges to other units, the proposed model TNDVGA learns latent instrumental, confounding, adjustment, and noise factors, and uses only the confounder and adjustment factors to predict potential outcomes. The reason this would matter is that real covariates often include features that affect only treatment, only outcome, or neither, and forcing all of them into a confounder representation introduces bias. The paper proves a conditional identifiability result for this setup and reports the lowest PEHE and ATE errors among the compared methods on synthetic data and on the BlogCatalog and Flickr semisynthetic benchmarks. A sympathetic reader would take the core claim to be that explicit four-way disentanglement, enforced by independence penalties and an adjustment-factor balancing loss, is what drives the improved estimates.

What carries the argument

The load-bearing object is the four-way disentangled latent representation inside a variational graph autoencoder. Four graph-convolutional encoders map the covariates x and adjacency matrix A to Gaussian posteriors for latent instrumental factors z_t, confounders z_c, adjustment factors z_y, and noise z_o; the generative model reconstructs x from all four, treatment from z_t and z_c, and outcome from t, z_c, and z_y. Disentanglement is enforced by pairwise Hilbert-Schmidt Independence Criterion penalties, a kernel-based measure that is zero exactly when two random variables are independent for characteristic kernels, and by a Wasserstein balance loss that forces the adjustment factor z_y to have the same distribution under treatment and control. The identifiability argument works through the Markov properties of the implied causal graph, which remove z_t and z_o from the outcome expression and leave only z_c and z_y in the ITE formula.

What would settle it

Take a synthetic dataset with known latent factors and known treatment and outcome functions, fit TNDVGA, and compare the learned z_c and z_y with the true simulated factors; if the model reports low PEHE while the recovered factors do not match the true confounder-adjustment split (for example, the correlation between learned and true factors is near zero), then the claimed recovery of the posterior is not what drives the reported accuracy.

Watch

Extended reading notes

Core claim

TNDVGA's central assertion is Theorem 4.1: if a model recovers both the posterior p(z_c, z_y | x, A) over latent confounders and adjustment factors and the outcome distribution p(y | t, z_c, z_y), then the individual treatment effect is identifiable from networked observational data, even with unmeasured confounders. The construction that is claimed to achieve this is a variational graph autoencoder with four separate graph-convolutional encoders, one for each latent factor group, a decoder that reconstructs covariates, treatment, and outcome from the appropriate factor subsets, a Hilbert-Schmidt Independence Criterion penalty that drives the four groups toward statistical independence, and an integral probability metric loss that balances the adjustment factor distributions between treated and control units. Empirically, the paper reports that on BlogCatalog and Flickr, TNDVGA attains the smallest root-mean-squared PEHE and ATE absolute error among BART, Causal Forest, CFR variants, TARNet, CEVAE, Intact-VAE, TEDVAE, NetDeconf, and GIAL variants, with ablations showing that removing either the HSIC penalty or the balancing loss degrades performance.

Load-bearing premise

The load-bearing premise is that the observed covariates and network really are generated from exactly four independent hidden factor types and that the model's training procedure recovers the true confounder and adjustment factors rather than some other split that merely predicts outcomes well.

Editorial extensions

If this is right

If the central claim is right, networked observational studies can drop the assumption that every observable proxy is a confounder; covariates that affect only treatment, only outcome, or neither can be routed to their own latent factors instead of contaminating the confounder estimate.
The method implies that the outcome decoder should be built from the confounder and adjustment factors alone, so treatment-effect predictions no longer depend on instrumental or noise factors.
The ablations imply that both the HSIC independence penalty and the adjustment-factor balancing loss are doing real work, since dropping either worsens PEHE.
On the synthetic data, the model's error stays low as selection bias grows, which suggests the disentangling approach is most valuable precisely when treatment assignment and outcome are strongly confounded.

Reading between the lines

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

Beyond the paper: the identifiability theorem is conditional on recovering the true posterior over z_c and z_y, and no injectivity or distributional identifiability conditions are stated for the encoder, so the reported gains could in principle come from flexible outcome prediction rather than from genuinely recovering the true factors; measuring how well learned factors match ground truth on synt
Beyond the paper: the same four-group scheme could be tested in settings with continuous or multiple treatments, or with time-varying networks, by keeping the HSIC independence structure and changing only the treatment and outcome decoders.
Beyond the paper: because the method assumes no interference between units, applying it to settings with spillover effects would require conditioning on neighbors' treatments; otherwise the estimated ITE would not correspond to the standard potential-outcome contrast.

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.

Desk Editor's Note

read the letter

Bottom line: TNDVGA is a sensible composition—variational graph autoencoder with four-way disentanglement, HSIC independence penalties, and a balance loss on the adjustment factor—and it posts the best numbers on BlogCatalog and Flickr. That part is real and worth taking seriously. But the stress-test note is right: the central claim about inferring unobserved factors from proxy variables is never tested, because the synthetic data generates x as the concatenation of the true latent factors. The encoders are effectively copying coordinates. The experiments show the model can exploit known factors plus regularizers; they do not show it can discover hidden factors from noisy proxies. That is a meaningful gap. The theoretical result is also conditional in the way the reader says: if you recover p(z_c,z_y|x,A) and the outcome distribution, then ITE follows from the assumed graph. That is a restatement of the causal model, not an identifiability result under a specific data-generating process. The mean-field posterior is a further mismatch: for a general nonlinear mixing x=g(z)+epsilon, z_c and z_y are dependent given x, so the factorized encoder cannot represent the posterior the theorem requires. Those are real soft spots. Smaller ones: no error bars in the main tables, baselines for BlogCatalog and Flickr are borrowed from the GIAL paper rather than rerun, and the synthetic experiments use oracle latent dimensions. The gains over GIAL-GCN are small, though consistent. The ablations do show the HSIC and balance terms matter, which is a point in the paper's favor. The writing is clear, the related work is handled honestly, and the authors do not oversell the empirical margin; they just oversell what the synthetic tests establish. Who is this for? People working on networked treatment effect estimation or disentangled representation for causal inference will want to know about it as another baseline. It deserves a serious referee, not a desk reject. I would send it out and ask for: code and error bars, a synthetic experiment with a nonlinear/noisy mixing that the encoders have to invert, and an honest rewrite of Theorem 4.1 as a conditional statement rather than an identifiability guarantee.

Referee Report

5 major / 6 minor

Summary. The paper proposes TNDVGA, a variational graph autoencoder for estimating individual treatment effects from networked observational data. The model uses four separate GCN-based encoders to approximate posterior distributions over four latent factor groups—instrumental, confounding, adjustment, and noise—and trains them with an ELBO objective, auxiliary treatment and outcome prediction losses, an HSIC-based independence penalty, and a balance loss applied to the adjustment factor. The authors prove a conditional identifiability statement (Theorem 4.1), evaluate the method on one synthetic and two semi-synthetic networked datasets, and report that TNDVGA outperforms existing baselines on PEHE and ATE metrics.

Significance. If the latent-factor recovery claim were convincingly demonstrated, the paper would make a useful contribution to ITE estimation on networked data, an area where explicit disentanglement of instrumental, confounding, adjustment, and noise factors is still underdeveloped. The model architecture is coherent, and the application of HSIC independence penalties to a VGAE is a reasonable design choice. However, the central theoretical result is a conditional equivalence that does not address identifiability of the latent factors themselves, and the experimental design does not actually test the proxy-recovery regime that motivates the model. The paper also lacks error bars and reuses baseline numbers from prior work. These issues leave the main empirical claim weaker than the text suggests.

major comments (5)

[Section 5.1.1, Eq. (22)] The synthetic data generator sets x = Concat(z_t, z_c, z_y, z_o), so every latent factor is a directly observed coordinate block of x. In this setting, recovering p(z_c, z_y | x, A) is trivial coordinate copying, and the network A is not needed for confounding adjustment. These experiments therefore do not test the paper's motivating regime, stated in Section 4.1, in which x and A are imperfect proxies for unobserved factors. I request synthetic experiments with nonlinear or noisy mixing, for example x = g(z) + epsilon with g a multilayer perceptron, where the true factors are not observable coordinate blocks, and a quantitative comparison of the learned factors against the true ones (for instance, with MMD or R^2 after aligning dimensions). Without such experiments, the improvements in Figs. 3 and 4 cannot be attributed to successful discovery of hidden disentangled factors.
[Section 4.2.1, Eq. (10)] The variational posterior is mean-field, factorized as q(z_t|x,A) q(z_c|x,A) q(z_y|x,A) q(z_o|x,A). For a general proxy model of the form x = g(z_t, z_c, z_y, z_o) + epsilon, the true posterior p(z_c, z_y | x, A) generally has z_c and z_y dependent after conditioning on x, so the factorized family cannot represent it even with perfect optimization. Thus the inference model does not match the premise of Theorem 4.1 in the general proxy setting. The paper should either restrict the generative model to a class in which conditional independence holds (for example, conditionally factorial decoders with additive noise), add auxiliary latent variables or a normalizing flow to capture posterior dependence, or state the narrower conditions under which Theorem 4.1 applies to the mean-field family actually used.
[Tables 2 and 3; Section 5.5] The comparison against BART, Causal Forest, CFR, TARNet, CEVAE, NetDeconf, and GIAL uses results taken from a previous paper [9], as stated in Section 5.5, rather than re-running those baselines under the same splits and hyperparameter selection. At the same time, the paper reports only averages for TNDVGA, with no standard deviations or confidence intervals. Several reported differences are small, for example 3.969 versus 4.023 for BlogCatalog at kappa2=0.5, and could be within run-to-run noise. Please report mean and standard deviation over the 10 runs for TNDVGA and, whenever feasible, re-run the baselines in the same framework, or at least provide a direct statistical comparison such as paired tests with matched folds.
[Section 5.4] The parameter settings state that, unless otherwise noted, the latent variable dimensions d_z_t, d_z_c, d_z_y, d_z_o are set to their true values on the synthetic data. This gives TNDVGA oracle knowledge of the factor structure that a real method would not have. The radar-chart experiment in Section 5.5.1 only tests setting one dimension to zero, never overestimating or misspecifying dimensions. The synthetic evaluation would be more convincing if it included a hyperparameter sweep over latent dimensions, with dimensions selected on the validation set, and if results were reported for misspecified dimensions as well as oracle dimensions.
[Section 4.3.4, Eq. (20)] The balance loss L_disc is applied to z_y, based on the assumption that p(z_y | t=0) = p(z_y | t=1). If z_y is a true adjustment factor, this equality should hold by construction when z_y is learned correctly, but the loss can also be minimized by making z_y uninformative, for example by driving it toward a constant or toward pure noise, which would not satisfy p(z_y|t=0)=p(z_y|t=1) in a meaningful sense. The paper should verify that z_y retains predictive information for the outcome after balancing, for instance by measuring the mutual information or prediction accuracy of z_y for y in both treatment groups, or by showing that the two-headed outcome decoder still relies on z_y. Otherwise the balance loss may be discarding useful adjustment information rather than removing confounding.

minor comments (6)

[Section 2, VGAE background, Eq. (2)] The product over i appears twice in the VGAE likelihood expression; it should likely be a product over i and j, or a single product over node pairs.
[Section 4.2.3, Eq. (15)] The matrices are called 'Grammer matrices'; the correct spelling is 'Gram matrices'.
[Section 5.5.2, Table 4] In the BlogCatalog ablation table, the TNDVGA row at kappa2=2 reports sqrt-PEHE = 0.651, which is far smaller than the corresponding value in Table 2 (6.066) and appears to be a typo or a formatting error; please correct and verify.
[Section 5.5] The heading 'Perfomance Comparision' contains typos; it should read 'Performance Comparison'.
[Section 5.1.2] The sentence 'In this dataset, the hyperparameters kappa1 >= 0 and kappa2 >= 0 control the strength of the selection bias introduced by the blogger's topics and the topics of their neighbors, respectively' defines kappa1 but the experiments only vary kappa2; please clarify the role and value of kappa1.
[Theorem 4.1 proof] The proof uses a consistency assumption between steps (ii) and (iii), but this assumption is not listed among the numbered assumptions in Section 3; please state it explicitly.

Circularity Check

1 steps flagged · score 4.0 of 10

Synthetic validation of latent-factor recovery is self-fulfilling: x is defined as the concatenation of the latent factors, so 'recovery' reduces to copying observed coordinate blocks; the BlogCatalog/Flickr benchmarks give independent external support and keep the paper from being wholly circular.

self definitional [Section 5.1.1 (Eq. 22); Section 5.4; Section 5.5.1 (radar-chart ablation)]
"x = Concat(z_t, z_c, z_y, z_o); ... For the synthetic datasets, we use the same parameter selection approach as for the semi-synthetic datasets. Unless stated otherwise, the latent variable dimensions for the different factors are set to their true values. ... If TNDVGA performs better when considering the disentanglement of all latent factors compared to when any one latent factor is ignored, then it can be concluded that TNDVGA can recover latent factors."

In the synthetic data, the latent factors the model is claimed to recover are literally the observable covariates: z_t, z_c, z_y, and z_o are coordinate blocks of x by construction, and the model is additionally given their true dimensions. Thus p(z_c,z_y|x,A) is degenerate/deterministic on observed coordinates, and 'recovering' the latent factors is equivalent to copying input blocks, not to inferring unobserved factors from proxies x and A. The synthetic experiments therefore do not test the paper's motivating regime or Theorem 4.1's precondition; the ablation's performance drops simply reflect deleting input information. The claimed recovery is true by construction, not by empirical discovery.

full rationale

The paper's real-data empirical claims on BlogCatalog and Flickr are tested against external baselines with ground-truth ITE generated from the procedures of [16,41], so the main PEHE/ATE comparisons have independent content and are not reduced to the model's own inputs. Nor are there load-bearing self-citations: the inspiration from TEDVAE [51] is acknowledged as inspiration, and no uniqueness or identifiability theorem is imported from the authors' own prior work. What lowers the score is the synthetic validation (Section 5.1.1, Eq. 22): because x = Concat(z_t,z_c,z_y,z_o) and the latent dimensions are set to the true values, the experiments that purport to show TNDVGA 'can recover latent factors' are self-fulfilling. Also, Theorem 4.1 itself is a conditional sufficiency statement: it shows that ITE is computable from the model's target distributions if they are recovered, but the paper does not prove recoverability under a general proxy mixing x = g(z) + noise, and the mean-field encoder family of Eq. (10) may be unable to represent the true joint posterior in that regime. Those are correctness/identifiability gaps rather than circular derivations, so they are noted but not scored as circular steps. Overall, the circularity is partial and confined to the synthetic support for the recovery claim, while the central real-data benchmark remains externally grounded.

Assumptions & free parameters 5 free parameters · 6 assumptions · 1 invented entities

The central claim rests on the assumption that a four-factor latent structure exists and is recoverable from proxies, plus several tuned hyperparameters. The synthetic validation is generated from the same structure the model is designed to learn, so it does not independently confirm that the model discovers the true factors in real data.

free parameters (5)

latent factor dimensions d_z_t, d_z_c, d_z_y, d_z_o = synthetic: true dims (4 or 8); semi-synthetic: grid searched in {10, 20, 30, 40, 50}
The dimension of each latent factor is a model choice that materially affects disentanglement capacity; using true dims in synthetic experiments is an oracle setting, and the semi-synthetic performance varies with this choice.
HSIC weight alpha_1 = tuned over {0.01, 0.1, 1, 10, 100}
Controls the strength of independence regularization; the paper reports performance degrades when alpha_1 >= 10.
balance loss weight alpha_2 = tuned over {0.01, 0.1, 1, 10, 100}
Controls the IPM balance loss on z_y; the paper reports performance degrades when alpha_2 >= 10.
treatment/outcome auxiliary loss weights alpha_t, alpha_y = 100
Set to 100 in all experiments; these weights force the latent factors to be predictive of t and y during training.
L2 weight lambda = 5e-5
Weight decay coefficient chosen by grid search to avoid overfitting.

assumptions (6)

domain assumption SUTVA: potential outcomes of a unit are unaffected by treatments of other units (Assumption 1)
Section 3 states this assumption; it rules out interference even though the data are networked and the network is used for inference.
domain assumption Overlap: 0 < P(t=1|x) < 1 (Assumption 2)
Standard positivity assumption required for identifying individual treatment effects.
domain assumption Unconfoundedness given latent confounders: t is independent of (y0, y1) conditional on z_c (Assumption 3)
Section 4.1; this relaxed ignorability assumption is the core of the identifiability proof.
ad hoc to paper Proxy assumption: observed covariates x and network A are generated from the four latent factors, and the factorization is recoverable
Section 4.1 states 'we assume that x and A serve as proxy variables for all latent variables'. The paper gives no identifiability conditions under which the variational objective recovers the true factors.
domain assumption Conditional independence y is independent of z_t, z_o given t, z_c, z_y, and x is independent of A given z (Markov properties of Figure 1)
Used in Equation (4) step (iv) and in the decoder simplification in Section 4.2.1; these follow from the assumed generative graph.
domain assumption Adjustment factors are independent of treatment: z_y is independent of t
Enforced by the balanced representation loss in Section 4.3.4; this is what characterizes z_y as an adjustment factor, but it is imposed rather than verified.

invented entities (1)

four-way latent factor decomposition (z_t, z_c, z_y, z_o)
purpose: To separate instrumental, confounding, adjustment, and noise causes so that only confounders and adjusters are used for outcome prediction.
The four factor types are postulated by the model. The paper provides no falsifiable handle outside the model: synthetic experiments construct data from the same decomposition and give the true dimensions, while semi-synthetic experiments never measure the recovered factors against ground truth.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Disentangled Graph Autoencoder for Treatment Effect Estimation." pith.science (2026). https://pith.science/paper/POXFCL7F

@misc{pith2026241214497,
  author       = {Pith},
  title        = {Pith review of: Disentangled Graph Autoencoder for Treatment Effect Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/POXFCL7F}},
  note         = {Machine review of arXiv:2412.14497}
}

read the original abstract

Treatment effect estimation from observational data has attracted significant attention across various research fields. However, many widely used methods rely on the unconfoundedness assumption, which is often unrealistic due to the inability to observe all confounders, thereby overlooking the influence of latent confounders. To address this limitation, recent approaches have utilized auxiliary network information to infer latent confounders, relaxing this assumption. However, these methods often treat observed variables and networks as proxies only for latent confounders, which can result in inaccuracies when certain variables influence treatment without affecting outcomes, or vice versa. This conflation of distinct latent factors undermines the precision of treatment effect estimation. To overcome this challenge, we propose a novel disentangled variational graph autoencoder for treatment effect estimation on networked observational data. Our graph encoder disentangles latent factors into instrumental, confounding, adjustment, and noisy factors, while enforcing factor independence using the Hilbert-Schmidt Independence Criterion. Extensive experiments on multiple networked datasets demonstrate that our method outperforms state-of-the-art approaches.

Figures

Figures reproduced from arXiv: 2412.14497 by the authors.

**Figure 2.** The overall architecture of TNDVGA consists of a generative network and an inference network for [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** Experimental results of different methods in ITE estimation under different levels of selection bias. As [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: In the radar chart, each vertex of the polygon is labeled with a sequence of latent factor dimensions [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]

**Figure 5.** Figure 5: Hyperparameter analysis on BlogCatalog across different [PITH_FULL_IMAGE:figures/full_fig_p019_5.png]

**Figure 6.** Figure 6: Hyperparameter analysis on Flickr across different [PITH_FULL_IMAGE:figures/full_fig_p019_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 37 canonical work pages

[9]

Zhixuan Chu, Stephen L Rathbun, and Sheng Li. 2021. Graph infomax adversarial learning for treatment effect estimation with networked observational data. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining . 176–184

work page 2021
[1]

Alberto Abadie and Guido W Imbens. 2006. Large sample properties of matching estimators for average treatment effects. econometrica 74, 1 (2006), 235–267

work page 2006
[2]

David Arbour, Dan Garant, and David Jensen. 2016. Inferring network effects from observational data. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining . 715–724

work page 2016
[3]

Onur Atan, James Jordon, and Mihaela Van der Schaar. 2018. Deep-treat: Learning optimal personalized treatments from observational data using neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 32

work page 2018
[4]

Susan Athey and Guido Imbens. 2016. Recursive partitioning for heterogeneous causal effects. Proceedings of the National Academy of Sciences 113, 27 (2016), 7353–7360

2016
[5]

Qingsen Bao, Zeyong Mao, and Lei Chen. 2022. Learning Disentangled Latent Factors for Individual Treatment Effect Estimation Using Variational Generative Adversarial Nets. In 2022 IEEE 25th International Conference on Computer Supported Cooperative Work in Design (CSCWD) . IEEE, 347–352

work page 2022
[6]

Andrew Bennett and Nathan Kallus. 2019. Policy evaluation with latent confounders via optimal balance. Advances in neural information processing systems 32 (2019)

work page 2019
[7]

Mingyuan Cheng, Xinru Liao, Quan Liu, Bin Ma, Jian Xu, and Bo Zheng. 2022. Learning disentangled representations for counterfactual regression via mutual information minimization. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval . 1802–1806

2022

Show all 53 references

[8]

Hugh A Chipman, Edward I George, and Robert E McCulloch. 2010. BART: Bayesian additive regression trees. (2010)

2010
[10]

Marco Cuturi and Arnaud Doucet. 2014. Fast computation of Wasserstein barycenters. In International conference on machine learning. PMLR, 685–693

2014
[11]

Weili Ding and Steven F Lehrer. 2010. Estimating treatment effects from contaminated multiperiod education experiments: The dynamic impacts of class size reductions. The Review of Economics and Statistics 92, 1 (2010), 31–42

2010
[12]

Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. 2014. Generative adversarial nets. Advances in neural information processing systems 27 (2014). Disentangled Graph Autoencoder for Treatment Effect ...

2014
[13]

Arthur Gretton, Olivier Bousquet, Alex Smola, and Bernhard Schölkopf. 2005. Measuring statistical dependence with Hilbert-Schmidt norms. In International conference on algorithmic learning theory . Springer, 63–77

2005
[14]

Ruocheng Guo, Jundong Li, Yichuan Li, K Selçuk Candan, Adrienne Raglin, and Huan Liu. 2021. Ignite: A minimax game toward learning individual treatment effects from networked observational data. In Proceedings of the Twenty-Ninth International Conference on International Joint...

2021
[15]

Ruocheng Guo, Jundong Li, and Huan Liu. 2020. Counterfactual evaluation of treatment assignment functions with networked observational data. InProceedings of the 2020 SIAM International Conference on Data Mining. SIAM, 271–279

2020
[16]

Ruocheng Guo, Jundong Li, and Huan Liu. 2020. Learning individual causal effects from networked observational data. In Proceedings of the 13th international conference on web search and data mining . 232–240

2020
[17]

Jenny Häggström. 2018. Data-driven confounder selection via Markov and Bayesian networks. Biometrics 74, 2 (2018), 389–398

2018
[18]

Jason Hartford, Greg Lewis, Kevin Leyton-Brown, and Matt Taddy. 2017. Deep IV: A flexible approach for counterfactual prediction. In International Conference on Machine Learning . PMLR, 1414–1423

2017
[19]

Negar Hassanpour and Russell Greiner. 2019. Learning disentangled representations for counterfactual regression. In International Conference on Learning Representations

2019
[20]

Qiang Huang, Jing Ma, Jundong Li, Ruocheng Guo, Huiyan Sun, and Yi Chang. 2023. Modeling Interference for Individual Treatment Effect Estimation from Networked Observational Data.ACM Transactions on Knowledge Discovery from Data 18, 3 (2023), 1–21

2023
[21]

Guido W Imbens and Donald B Rubin. 2015. Causal inference in statistics, social, and biomedical sciences . Cambridge university press

2015
[22]

Fredrik Johansson, Uri Shalit, and David Sontag. 2016. Learning representations for counterfactual inference. In International conference on machine learning . PMLR, 3020–3029

2016
[23]

Diederik P Kingma. 2013. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114 (2013)

2013 arXiv
[24]

Diederik P Kingma. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 (2014)

2014 arXiv
[25]

Thomas N Kipf and Max Welling. 2016. Semi-supervised classification with graph convolutional networks. arXiv preprint arXiv:1609.02907 (2016)

2016 arXiv
[26]

Thomas N Kipf and Max Welling. 2016. Variational graph auto-encoders. arXiv preprint arXiv:1611.07308 (2016)

2016 arXiv
[27]

Kun Kuang, Peng Cui, Bo Li, Meng Jiang, Shiqiang Yang, and Fei Wang. 2017. Treatment effect estimation with data-driven variable decomposition. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 31

2017
[28]

Kun Kuang, Peng Cui, Hao Zou, Bo Li, Jianrong Tao, Fei Wu, and Shiqiang Yang. 2020. Data-driven variable decomposition for treatment effect estimation. IEEE Transactions on Knowledge and Data Engineering 34, 5 (2020), 2120–2134

2020
[29]

Solomon Kullback and Richard A Leibler. 1951. On information and sufficiency. The Annals of Mathematical Statistics 22, 1 (1951), 79–86

1951
[30]

Yu Liu, Jian Wang, and Bing Li. 2024. EDVAE: Disentangled latent factors models in counterfactual reasoning for individual treatment effects estimation. Information Sciences 652 (2024), 119578

2024
[31]

Christos Louizos, Uri Shalit, Joris M Mooij, David Sontag, Richard Zemel, and Max Welling. 2017. Causal effect inference with deep latent-variable models. Advances in neural information processing systems 30 (2017)

2017
[32]

Judea Pearl. 2009. Causal inference in statistics: An overview. (2009)

2009
[33]

Judea Pearl. 2009. Causality. Cambridge university press

2009
[34]

Paul R Rosenbaum and Donald B Rubin. 1983. The central role of the propensity score in observational studies for causal effects. Biometrika 70, 1 (1983), 41–55

1983
[35]

Donald B Rubin. 1978. Bayesian inference for causal effects: The role of randomization. The Annals of statistics (1978), 34–58

1978
[36]

Donald B Rubin. 2005. Causal inference using potential outcomes: Design, modeling, decisions. J. Amer. Statist. Assoc. 100, 469 (2005), 322–331

2005
[37]

Bernhard Schölkopf, Francesco Locatello, Stefan Bauer, Nan Rosemary Ke, Nal Kalchbrenner, Anirudh Goyal, and Yoshua Bengio. 2021. Toward causal representation learning. Proc. IEEE 109, 5 (2021), 612–634

2021
[38]

Uri Shalit, Fredrik D Johansson, and David Sontag. 2017. Estimating individual treatment effect: generalization bounds and algorithms. In International conference on machine learning . PMLR, 3076–3085

2017
[39]

Le Song, Alex Smola, Arthur Gretton, Justin Bedo, and Karsten Borgwardt. 2012. Feature Selection via Dependence Maximization. Journal of Machine Learning Research 13, 5 (2012)

2012
[40]

Bharath K Sriperumbudur, Kenji Fukumizu, Arthur Gretton, Bernhard Schölkopf, and Gert RG Lanckriet. 2012. On the empirical estimation of integral probability metrics. (2012)

2012
[41]

Lei Tang and Huan Liu. 2011. Leveraging social media networks for classification.Data mining and knowledge discovery 23 (2011), 447–478. 111:22 Fan et al

2011
[42]

Abhinav Thorat, Ravi Kolla, Niranjan Pedanekar, and Naoyuki Onoe. 2023. Estimation of individual causal effects in network setup for multiple treatments. arXiv preprint arXiv:2312.11573 (2023)

2023 arXiv
[43]

Victor Veitch, Yixin Wang, and David Blei. 2019. Using embeddings to correct for unobserved confounding in networks. Advances in Neural Information Processing Systems 32 (2019)

2019
[44]

Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, and Yoshua Bengio. 2017. Graph attention networks. arXiv preprint arXiv:1710.10903 (2017)

2017 arXiv
[45]

Matthew J Vowels, Necati Cihan Camgoz, and Richard Bowden. 2021. Targeted VAE: Variational and targeted learning for causal inference. In 2021 IEEE International Conference on Smart Data Services (SMDS) . IEEE, 132–141

2021
[46]

Stefan Wager and Susan Athey. 2018. Estimation and inference of heterogeneous treatment effects using random forests. J. Amer. Statist. Assoc. 113, 523 (2018), 1228–1242

2018
[47]

Anpeng Wu, Junkun Yuan, Kun Kuang, Bo Li, Runze Wu, Qiang Zhu, Yueting Zhuang, and Fei Wu. 2022. Learning decomposed representations for treatment effect estimation. IEEE Transactions on Knowledge and Data Engineering 35, 5 (2022), 4989–5001

2022
[48]

Pengzhou Wu and Kenji Fukumizu. 2021. Intact-VAE: Estimating treatment effects under unobserved confounding. arXiv preprint arXiv:2101.06662 (2021)

2021 arXiv
[49]

Liuyi Yao, Zhixuan Chu, Sheng Li, Yaliang Li, Jing Gao, and Aidong Zhang. 2021. A survey on causal inference. ACM Transactions on Knowledge Discovery from Data (TKDD) 15, 5 (2021), 1–46

2021
[50]

Liuyi Yao, Sheng Li, Yaliang Li, Mengdi Huai, Jing Gao, and Aidong Zhang. 2018. Representation learning for treatment effect estimation from observational data. Advances in neural information processing systems 31 (2018)

2018
[51]

Weijia Zhang, Lin Liu, and Jiuyong Li. 2021. Treatment effect estimation with disentangled latent factors. InProceedings of the AAAI Conference on Artificial Intelligence , Vol. 35. 10923–10930

2021
[52]

Yang Zhang, Fuli Feng, Xiangnan He, Tianxin Wei, Chonggang Song, Guohui Ling, and Yongdong Zhang. 2021. Causal intervention for leveraging popularity bias in recommendation. In Proceedings of the 44th international ACM SIGIR conference on research and development in informatio...

2021
[53]

Ziyu Zhao, Anpeng Wu, Kun Kuang, Ruoxuan Xiong, Bo Li, Zhihua Wang, and Fei Wu. 2024. Networked Instrumental Variable for Treatment Effect Estimation with Unobserved Confounders. IEEE Transactions on Knowledge and Data Engineering (2024). Received 20 February 2007; revised 12 ...

2024

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [9]

Zhixuan Chu, Stephen L Rathbun, and Sheng Li. 2021. Graph infomax adversarial learning for treatment effect estimation with networked observational data. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining . 176–184

work page 2021

[2] [1]

Alberto Abadie and Guido W Imbens. 2006. Large sample properties of matching estimators for average treatment effects. econometrica 74, 1 (2006), 235–267

work page 2006

[3] [2]

David Arbour, Dan Garant, and David Jensen. 2016. Inferring network effects from observational data. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining . 715–724

work page 2016

[4] [3]

Onur Atan, James Jordon, and Mihaela Van der Schaar. 2018. Deep-treat: Learning optimal personalized treatments from observational data using neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence , Vol. 32

work page 2018

[5] [4]

Susan Athey and Guido Imbens. 2016. Recursive partitioning for heterogeneous causal effects. Proceedings of the National Academy of Sciences 113, 27 (2016), 7353–7360

2016

[6] [5]

Qingsen Bao, Zeyong Mao, and Lei Chen. 2022. Learning Disentangled Latent Factors for Individual Treatment Effect Estimation Using Variational Generative Adversarial Nets. In 2022 IEEE 25th International Conference on Computer Supported Cooperative Work in Design (CSCWD) . IEEE, 347–352

work page 2022

[7] [6]

Andrew Bennett and Nathan Kallus. 2019. Policy evaluation with latent confounders via optimal balance. Advances in neural information processing systems 32 (2019)

work page 2019

[8] [7]

Mingyuan Cheng, Xinru Liao, Quan Liu, Bin Ma, Jian Xu, and Bo Zheng. 2022. Learning disentangled representations for counterfactual regression via mutual information minimization. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval . 1802–1806

2022