Pith. sign in

REVIEW 3 major objections 6 minor 50 references

Graph Counterfactual Explainable AI via Latent Space Traversal

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Latent-space traversal of a permutation-equivariant graph VAE generates counterfactual explanations for any differentiable graph classifier without defining a graph distance.

desk verdict A clearly written but explicitly preliminary paper on latent-space traversal for graph counterfactuals; the core idea is plausible, the evaluation is too weak to support the robustness claim. read the letter →

arxiv 2501.08850 v1 pith:JWB2JF35 submitted 2025-01-15 cs.LG cs.AIstat.ML

Andreas Abildtrup Hansen , Paraskevas Pegios , Anna Calissano , Aasa Feragen This is my paper · ORCID
classification cs.LGcs.AIstat.ML
keywords counterfactualexplanationsgraphneuralnetworksvariationalautoencoderpermutationequivariancelatentspacetraversalexplainabilitymoleculargraphs
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

The paper proposes a way to explain a graph classifier's decision by producing counterfactual graphs: alternative inputs that look as close as possible to the original while flipping the predicted class. The method, CGCF, encodes a factual graph with a permutation-equivariant variational autoencoder, then walks the latent code in the direction that most increases the classifier's probability of the desired class, decoding at each step until the label flips. Because the walk happens in a learned latent space, the paper never has to choose a graph distance, which is normally the arbitrary part of counterfactual explanation. On NCI1, Mutagenicity, and AIDS, the classifier-guided walks reached the highest flip-ratio among the compared methods while staying competitive on identity preservation. The authors present the procedure as a general recipe for any differentiable graph classifier, contingent on the latent space actually representing realistic in-distribution graphs.

What carries the argument

The central object is the permutation-equivariant graph variational autoencoder (PEGVAE), whose latent space is traversed to reach the counterfactual. Equivariant linear layers, built from a fixed set of Bell-number basis elements, make the encoder and decoder commute with node permutations, so a relabeled input produces a relabeled counterfactual and the factual/counterfactual alignment is preserved. The traversal itself is the update $z_{i+1} = z_i - \epsilon \nabla \mathcal{L}(z_i, y_D)$, where $\mathcal{L}$ is cross-entropy against the desired class $y_D$ plus an $\ell^2$ penalty on the latent code, and a Gumbel-Softmax approximation to $p_\theta(G \mid z)$ makes the decoder differentiable so the classifier's gradient can steer the walk.

What would settle it

Compute the chemical validity of the decoded counterfactual graphs and the latent distance from each counterfactual code to the nearest training-set encoding; if a substantial fraction of flips are chemically invalid, or if the walked codes cluster far outside the region occupied by training encodings, the premise that traversal stays in-distribution is false.

Watch

Extended reading notes

Core claim

The paper claims that classifier-guided traversal of a permutation-equivariant VAE latent space produces valid, aligned counterfactual graphs for any differentiable graph classifier, without an explicit graph metric and without training the generative model with a counterfactual-specific loss. The walk is steered by gradient descent on the cross-entropy of the decoded graph with respect to the desired class, regularized by an L2 penalty on the latent code, and a Gumbel-Softmax relaxation lets gradients flow through the discrete decoder. In the reported experiments, CGCF is consistently high-performing and more robust than the baselines: it achieves the highest flip-ratio on all three datasets, and in the trade-off between identity preservation and validity it outperforms on NCI1 and Mutagenicity while remaining competitive on AIDS. The authors also note that because the method is probabilistic, an arbitrary number of counterfactual explanations can be sampled for a given factual graph.

Load-bearing premise

The load-bearing premise is that a VAE trained only with reconstruction and KL objectives produces a latent space in which gradient steps toward the classifier's decision boundary decode to realistic in-distribution graphs, and the paper's own limitations section acknowledges that a latent counterfactual obtained far from any training point may not be reasonable.

Editorial extensions

If this is right

  • A practitioner can generate counterfactual graphs for any differentiable black-box graph classifier without choosing a graph distance; the latent metric replaces that choice.
  • Because the decoder defines a distribution $p_\theta(G \mid z)$, the method can produce arbitrarily many explanations per factual graph by sampling at the found latent code.
  • The full pipeline is permutation equivariant, so explanations respect node ordering: permuting the input nodes permutes the counterfactual in the same way.
  • If the classifier is replaced, new counterfactuals can be generated for the new decision boundary without retraining the generative model.
  • The framework is stated to extend beyond the categorical molecule attributes used in the experiments to continuous node and edge attributes.

Reading between the lines

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

  • A testable extension would measure the latent distance from each counterfactual code to the nearest training encoding; if many walked codes land far from the data manifold, the flips may reflect out-of-distribution artifacts rather than meaningful molecular changes.
  • The same traversal recipe could apply to other structured data with a group action, such as point clouds or meshes, by substituting an appropriately equivariant autoencoder and differentiable decoder.
  • Because the walk is continuous, the method could be run in reverse from the counterfactual side to trace the decision boundary, yielding per-class minimal perturbations instead of per-instance explanations.
  • The reliance on a pretrained VAE suggests a practical shortcut: when a new classifier is deployed on the same graph domain, the expensive generative model can be reused and only the short traversal needs to be repeated.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes CGCF, a method for generating counterfactual explanations of graph classifiers by traversing the latent space of a permutation-equivariant variational autoencoder. Given a factual graph, the method encodes it, then iteratively updates the latent code with gradient descent on a cross-entropy loss (with L2 regularization) that pushes the decoded graph toward a desired class, using a Gumbel-Softmax relaxation to differentiate through the discrete decoder. The method is evaluated on three molecular graph datasets (AIDS, Mutagenicity, NCI1) against three latent-space baselines (random prior samples, nearest training graph, decoded mean of k-NN), and the paper claims that CGCF is consistently high-performing and more robust than the baselines, while removing the need to define an explicit graph distance.

Significance. If the result holds, the paper offers a principled and elegant way to generate graph counterfactuals without a hand-defined graph metric, leveraging equivariance to preserve node alignment and a pre-trained generative model to stay near the data manifold. Strengths of the paper include its explicit treatment of permutation equivariance, the algorithmic clarity of Section 2.5, the acknowledgment of limitations, and the promise of public source code. However, the central evaluation is weakened by a circularity between the optimized objective and the headline metric, and the main traversal mechanism relies on an unvalidated assumption that classifier gradients computed on Gumbel-Softmax relaxed graphs align with hard-decoded counterfactual flips. These issues must be addressed before the contribution can be considered established.

major comments (3)
  1. [Section 2.5, Eq. (5)-(6); Appendix A.1.1 and A.2] The traversal minimizes the classifier's cross-entropy on Gumbel-Softmax relaxed graphs at temperature τ=1, but the classifier was trained on discrete one-hot graphs. The paper never verifies that a latent point that flips the classifier under the soft relaxation also flips it when the decoded graph is obtained by hard sampling or argmax decoding, so the reported flip-ratios may reflect a surrogate problem rather than true counterfactual flips. I request an explicit measurement of the soft-versus-hard flip agreement, and, if the gap is nontrivial, a modification such as temperature annealing or straight-through estimation with a re-evaluation of all reported metrics.
  2. [Section 3.2, Table 1, and Eq. (6)] The primary validity metric, Flip-Ratio, measures exactly the objective being optimized in Eq. (6): the traversal minimizes cross-entropy toward the desired class, and Flip-Ratio records whether that class is achieved. This makes the headline claim partly circular. I recommend adding independent validity measures, such as chemical validity of decoded molecules (e.g., via RDKit), distance of the counterfactual latent code to latent encodings of training graphs, and agreement with a held-out or differently initialized classifier, and reporting Flip-Ratio with confidence intervals.
  3. [Section 3.3, Section 3.4, and Limitations (Section 4)] The claim that CGCF is "consistently high-performing and more robust than the baselines" is not yet supported because all baselines are weak and share the same VAE latent space; no comparison is made with established graph counterfactual methods such as CLEAR, CF-GNNExplainer, or D4Explainer. Additionally, the paper's own Limitations paragraph concedes that decoded counterfactuals can be unreasonable when the latent code is far from training points, but no experiment checks the distance of the obtained counterfactual latent codes to the latent training distribution or otherwise validates the in-distribution assumption. Adding such an analysis and stronger baselines is necessary to substantiate the robustness claim.
minor comments (6)
  1. [Title] The title contains a typo: "T raversal" should be "Traversal".
  2. [Section 2.5, Eq. (6)] The regularization term is written as λ∥z_i∥; please clarify whether this is an L2 norm or a squared L2 norm, and define the indicator notation 1{D=k}(y_D) more explicitly.
  3. [Algorithm 1] The stopping condition compares the classifier output y_i to the desired label y_D, but since the classifier outputs probabilities, the threshold used to declare a flip should be specified; also, once the desired label is achieved, the algorithm freezes further updates but still returns the graph from the final iteration, which may differ from the first successful counterfactual.
  4. [Table 1] Flip-Ratio is reported as a single scalar without variance or sample size, unlike the identity-preservation metrics; please provide confidence intervals or standard errors.
  5. [Appendix A.2] The hyperparameters τ, λ, and the number of iterations N are fixed a priori without sensitivity analysis; a brief study of their influence on flip-ratio and identity preservation would strengthen the practical claims.
  6. [Discussion, 'arbitrary number of explanations'] The claim that the method can produce an unlimited number of counterfactuals is only true if one samples from the decoder likelihood after the traversal; the traversal itself is deterministic given the factual graph and hyperparameters, so please clarify the intended stochasticity.

Circularity Check

1 steps flagged · score 4.0 of 10

Flip-ratio, used as the main validity claim, is the same objective that CGCF optimizes during latent traversal; identity-preservation metrics provide partial independent grounding.

  1. self definitional [Section 2.5 (Eq. 6 and Algorithm 1) vs Section 3.2 (Flip-Ratio)]
    "The loss function L is defined as a cross entropy loss with L2-regularization limiting the size of zi, i.e.: L(zi, yD) = − P K k=1 1{D=k}(yD) log(yi) + λ∥zi∥ ... The updates stop when either yi = yD, or when the maximum number of iterations, which is set as a hyperparameter, has been reached. ... Flip-Ratio (FR) refers to the number proportion of generated counterfactual explanations, which are assigned the desired class by the classifier."

    CGCF's traversal directly optimizes the classifier output toward the desired label yD through the cross-entropy loss in Eq. (6), and Algorithm 1 masks further updates for each counterfactual as soon as C(D(zi)) = yD. The Flip-Ratio metric then counts exactly the proportion of generated graphs that the classifier assigns to yD. A converged run therefore satisfies the FR criterion by construction; the near-1 FR values largely report whether the optimizer met its own stopping condition rather than an independent quality of the counterfactual. The metric is not a prediction from a fitted parameter, but it is the optimization objective renamed as a validity result.

full rationale

The paper's core derivation chain is otherwise self-contained: a permutation-equivariant VAE produces a latent representation, and Eq. (5)–(6) traverse that space using the classifier gradient; no fitted quantity is renamed as a prediction, and no load-bearing result is imported solely from the authors' prior work. The use of Hansen et al. [8] for the equivariant architecture is a normal methodological citation, not an imported uniqueness theorem or ansatz. The main circularity is in the evaluation: Flip-Ratio, the central validity metric used to claim consistent superiority over the baselines, is the same cross-entropy objective that the generation loop optimizes until the desired class is assigned. Thus the headline 'consistently high-performing and more robust' rests substantially on a self-referential success criterion. The identity-preservation metrics are independent and mitigate the issue, as does the honest Limitations passage about possibly unreasonable decoded counterfactuals far from latent training points. The soft-versus-hard Gumbel-Softmax distribution gap flagged by a skeptic is a correctness and robustness risk rather than a circular step, since the reported AUROC does not validate classifier behavior on relaxed inputs but this does not make the derivation equivalent to its inputs. Overall, partial circularity of the validity evaluation warrants a score of 4, not higher, because the identity-preservation results and the fixed-iteration termination keep the claims from being entirely forced by construction.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The method introduces no new physical or mathematical entities. It relies on a set of hand-chosen hyperparameters (beta, lambda, tau, learning rate, iteration cap, k) and on the unvalidated assumption that a standard beta-VAE latent space is smooth enough for classifier-guided traversal to yield realistic, in-distribution graphs. The equivariance and factorization assumptions are standard but load-bearing for the method's interpretability claim.

free parameters (6)
  • Beta (KL weight in beta-VAE) = 0.1 (AIDS), 0.5 (NCI1, Mutagenicity)
    Per-dataset weighting of the KL term in the beta-VAE objective; chosen by validation, affects latent space structure and hence counterfactual quality.
  • Lambda (L2 regularization in traversal loss, Eq. 6) = 1
    Controls how strongly the latent code is penalized for magnitude during traversal; chosen without reported sensitivity analysis.
  • Gumbel-Softmax temperature tau = 1
    Controls the relaxation sharpness for backpropagating through the discrete graph decoder; set to 1 without search.
  • Learning rate for latent traversal = 0.05
    Step size for latent updates in Algorithm 1; chosen as a hyperparameter, no sensitivity analysis reported.
  • Maximum iterations N = 1000
    Termination cap for latent updates; chosen as a hyperparameter, affects whether a counterfactual is found.
  • k for k-NN baseline = 10
    Number of nearest training-set latent codes averaged for the Decoded Mean of k-NN baseline; fixed without analysis.
assumptions (5)
  • domain assumption A meaningful latent graph representation has been obtained by a standard beta-VAE training, so Euclidean distance in latent space corresponds to semantically meaningful graph variation.
    Invoked in the opening of Section 2 and throughout the traversal procedure; the paper's Limitations section itself warns that latent points far from training data can decode to unreasonable graphs.
  • domain assumption The pretrained classifier and VAE are differentiable through the Gumbel-Softmax relaxation, and gradient steps in latent space stay on the data manifold.
    Required by Algorithm 1 and Eq. (5); the relaxation is an approximation and the manifold assumption is acknowledged as the main limitation.
  • domain assumption The permutation equivariance of the VAE and invariance of the classifier are sufficient to preserve node alignment between factual and counterfactual graphs.
    Used to justify that counterfactuals are interpretable with respect to the original node ordering (Section 2.2, Eq. (2)).
  • standard math The decoder factorizes the graph likelihood as p(B|z)p(V|z,B)p(A|z,B,V)p(E|z,B,V,A) with a Gaussian prior, which is a standard VAE modeling assumption.
    Section 2.3, Eq. (3); this factorization is standard and does not need independent verification.
  • domain assumption Dataset filtering rules from Huang et al. (frequency > 50, node limits) produce a balanced, representative evaluation set.
    Adopted in Section 3.1 from Ref. 50; the filtering changes the distribution and may affect the generality of the results.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Graph Counterfactual Explainable AI via Latent Space Traversal." pith.science (2026). https://pith.science/paper/JWB2JF35

@misc{pith2026250108850,
  author       = {Pith},
  title        = {Pith review of: Graph Counterfactual Explainable AI via Latent Space Traversal},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JWB2JF35}},
  note         = {Machine review of arXiv:2501.08850}
}
read the original abstract

Explaining the predictions of a deep neural network is a nontrivial task, yet high-quality explanations for predictions are often a prerequisite for practitioners to trust these models. Counterfactual explanations aim to explain predictions by finding the ''nearest'' in-distribution alternative input whose prediction changes in a pre-specified way. However, it remains an open question how to define this nearest alternative input, whose solution depends on both the domain (e.g. images, graphs, tabular data, etc.) and the specific application considered. For graphs, this problem is complicated i) by their discrete nature, as opposed to the continuous nature of state-of-the-art graph classifiers; and ii) by the node permutation group acting on the graphs. We propose a method to generate counterfactual explanations for any differentiable black-box graph classifier, utilizing a case-specific permutation equivariant graph variational autoencoder. We generate counterfactual explanations in a continuous fashion by traversing the latent space of the autoencoder across the classification boundary of the classifier, allowing for seamless integration of discrete graph structure and continuous graph attributes. We empirically validate the approach on three graph datasets, showing that our model is consistently high-performing and more robust than the baselines.

Figures

Figures reproduced from arXiv: 2501.08850 by the authors.

Figure 1
Figure 1. Top: The classifier architecture and the PEGVAE. Bottom: The counterfactual graph generation. of: A linear equivariant layer as specified by Maron et al. [39] and a non-linearity consisting of node- and edge-wise aggregations implemented as convolutions with a kernel size of one and the ReLU activation function. Batch normalization is applied before the activation. We will refer to this construction as an equivarian… view at source ↗
Figure 2
Figure 2. Trade-off between metrics for Identity Preservation and Validity. counterfactual graphs. Latent Euclidean Distance (LED) refers to the Euclidean distance between the latent representation of two graphs. Cosine Similar￾ity (CS) is computed between the graph embeddings extracted from the classifier. Validity. Two validity measures are used. Flip￾Ratio (FR) refers to the number proportion of gener￾ated counterfactual e… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 40 canonical work pages

  1. [1]

    A permutation testing framework to compare groups of brain networks

    Sean L Simpson, Robert G Lyday, Satoru Hayasaka, Anthony P Marsh, and Paul J Lau- rienti. A permutation testing framework to compare groups of brain networks. Front. Com- put. Neurosci., 7:171, November 2013

    work page 2013

  2. [2]

    Nonparametric bayes mod- eling of populations of networks

    Daniele Durante, David B Dunson, and Joshua T Vogelstein. Nonparametric bayes mod- eling of populations of networks. J. Am. Stat. Assoc., 112(520):1516–1530, October 2017

    work page 2017

  3. [3]

    Graph alignment exploiting the spatial organization improves the similarity of brain networks

    Anna Calissano, Theodore Papadopoulo, Xavier Pennec, and Samuel Deslauriers- Gauthier. Graph alignment exploiting the spatial organization improves the similarity of brain networks. Human Brain Mapping , 45(1): e26554, 2024

    work page 2024

  4. [4]

    Statistical shape analysis of brain arterial networks (BAN)

    Xiaoyang Guo, Aditi Basu Bal, Tom Needham, and Anuj Srivastava. Statistical shape analysis of brain arterial networks (BAN). The Annals of Applied Statistics , 16(2):1130 – 1150, 2022. doi: 10.1214/21-AOAS1536. URL https:// doi.org/10.1214/21-AOAS1536

    work page doi:10.1214/21-aoas1536 2022

  5. [5]

    Haonan Wang and J. S. Marron. Object ori- ented data analysis: Sets of trees. The An- nals of Statistics , 35(5):1849 – 1873, 2007. doi: 10.1214/009053607000000217. URL https: //doi.org/10.1214/009053607000000217

    work page doi:10.1214/009053607000000217 2007

  6. [6]

    Tree-space statistics and approximations for large-scale analysis of anatomical trees

    Aasa Feragen, Megan Owen, Jens Petersen, Mathilde M W Wille, Laura H Thomsen, Asger Dirksen, and Marleen de Bruijne. Tree-space statistics and approximations for large-scale analysis of anatomical trees. Inf. Process. Med. Imaging, 23:74–85, 2013

    work page 2013

  7. [7]

    Public transport networks: em- pirical analysis and modeling

    C von Ferber, T Holovatch, Yu Holovatch, and V Palchykov. Public transport networks: em- pirical analysis and modeling. Eur. Phys. J. B , 68(2):261–275, March 2009

    work page 2009

  8. [8]

    Interpreting equivariant rep- resentations

    Andreas Abildtrup Hansen, Anna Calissano, and Aasa Feragen. Interpreting equivariant rep- resentations. arXiv preprint arXiv:2401.12588 , 2024

    arXiv 2024

Show all 50 references
  1. [9]

    Beyond trivial counterfactual explanations with diverse valu- able explanations

    Pau Rodriguez, Massimo Caccia, Alexandre La- coste, Lee Zamparo, Issam Laradji, Laurent Charlin, and David Vazquez. Beyond trivial counterfactual explanations with diverse valu- able explanations. In 2021 IEEE/CVF Interna- tional Conference on Computer Vision (ICCV) , pages 10...

    2021

  2. [10]

    Learning model-agnostic coun- terfactual explanations for tabular data

    Martin Pawelczyk, Klaus Broelemann, and Gjergji Kasneci. Learning model-agnostic coun- terfactual explanations for tabular data. In Proceedings of The Web Conference 2020 , New York, NY, USA, April 2020. ACM

    2020

  3. [11]

    Counterfactual explanations and algorith- mic recourses for machine learning: A review

    Sahil Verma, Varich Boonsanong, Minh Hoang, Keegan E Hines, John P Dickerson, and Chirag Shah. Counterfactual explanations and algorith- mic recourses for machine learning: A review. arXiv preprint arXiv:2010.10596 , 2020

    2010 arXiv

  4. [12]

    Gnnx-bench: Unravelling the utility of perturbation-based gnn explain- ers through in-depth benchmarking

    Mert Kosan, Samidha Verma, Burouj Armgaan, Khushbu Pahwa, Ambuj Singh, Sourav Medya, and Sayan Ranu. Gnnx-bench: Unravelling the utility of perturbation-based gnn explain- ers through in-depth benchmarking. In The Twelfth International Conference on Learning Representations, 2023

    2023

  5. [13]

    A survey on graph counterfactual explanations: defini- tions, methods, evaluation, and research chal- lenges

    Mario Alfonso Prado-Romero, Bardh Prenkaj, Giovanni Stilo, and Fosca Giannotti. A survey on graph counterfactual explanations: defini- tions, methods, evaluation, and research chal- lenges. ACM Computing Surveys , 2023

    2023

  6. [14]

    Local rule-based explanations of black box decision systems

    Riccardo Guidotti, Anna Monreale, Salvatore Ruggieri, Dino Pedreschi, Franco Turini, and Fosca Giannotti. Local rule-based explanations of black box decision systems. arXiv preprint arXiv:1805.10820, 2018

    2018 arXiv

  7. [15]

    Multi-objective coun- terfactual explanations

    Susanne Dandl, Christoph Molnar, Martin Binder, and Bernd Bischl. Multi-objective coun- terfactual explanations. In International Con- ference on Parallel Problem Solving from Na- ture, pages 448–469. Springer, 2020

    2020

  8. [16]

    Actionable recourse in linear classification

    Berk Ustun, Alexander Spangher, and Yang Liu. Actionable recourse in linear classification. In Proceedings of the conference on fairness, accountability, and transparency, pages 10–19, 2019

    2019

  9. [17]

    Face: feasible and actionable counterfactual expla- nations

    Rafael Poyiadzi, Kacper Sokol, Raul Santos- Rodriguez, Tijl De Bie, and Peter Flach. Face: feasible and actionable counterfactual expla- nations. In Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , pages 344–350, 2020

    2020

  10. [18]

    Efficient search for diverse co- herent explanations

    Chris Russell. Efficient search for diverse co- herent explanations. In Proceedings of the con- ference on fairness, accountability, and trans- parency, pages 20–28, 2019. 7

    2019

  11. [19]

    Explaining machine learning classifiers through diverse counterfactual expla- nations

    Ramaravind K Mothilal, Amit Sharma, and Chenhao Tan. Explaining machine learning classifiers through diverse counterfactual expla- nations. In Proceedings of the 2020 conference on fairness, accountability, and transparency , pages 607–617, 2020

    2020

  12. [20]

    Preserving causal constraints in coun- terfactual explanations for machine learning classifiers

    Divyat Mahajan, Chenhao Tan, and Amit Sharma. Preserving causal constraints in coun- terfactual explanations for machine learning classifiers. arXiv preprint arXiv:1912.03277 , 2019

    1912 arXiv

  13. [21]

    Algo- rithmic recourse under imperfect causal knowl- edge: a probabilistic approach

    Amir-Hossein Karimi, Julius Von K¨ ugelgen, Bernhard Sch¨ olkopf, and Isabel Valera. Algo- rithmic recourse under imperfect causal knowl- edge: a probabilistic approach. Advances in neural information processing systems , 33:265– 277, 2020

    2020

  14. [22]

    Explanations based on the missing: Towards contrastive ex- planations with pertinent negatives

    Amit Dhurandhar, Pin-Yu Chen, Ronny Luss, Chun-Chen Tu, Paishun Ting, Karthikeyan Shanmugam, and Payel Das. Explanations based on the missing: Towards contrastive ex- planations with pertinent negatives. Advances in neural information processing systems , 31, 2018

    2018

  15. [23]

    Towards realistic individual recourse and action- able explanations in black-box decision mak- ing systems

    Shalmali Joshi, Oluwasanmi Koyejo, Warut Vi- jitbenjaronk, Been Kim, and Joydeep Ghosh. Towards realistic individual recourse and action- able explanations in black-box decision mak- ing systems. arXiv preprint arXiv:1907.09615 , 2019

    1907 arXiv

  16. [24]

    Counterfactual explanations via riemannian latent space traversal

    Paraskevas Pegios, Aasa Feragen, An- dreas Abildtrup Hansen, and Georgios Arvanitidis. Counterfactual explanations via riemannian latent space traversal. arXiv preprint arXiv:2411.02259, 2024

    2024 arXiv

  17. [25]

    Diffeomorphic explanations with normalizing flows

    Ann-Kathrin Dombrowski, Jan E Gerken, and Pan Kessel. Diffeomorphic explanations with normalizing flows. In ICML Workshop on In- vertible Neural Networks, Normalizing Flows, and Explicit Likelihood Models , 2021

    2021

  18. [26]

    Ceflow: A robust and efficient counterfactual ex- planation framework for tabular data using nor- malizing flows

    Tri Dung Duong, Qian Li, and Guandong Xu. Ceflow: A robust and efficient counterfactual ex- planation framework for tabular data using nor- malizing flows. In Pacific-Asia Conference on Knowledge Discovery and Data Mining , pages 133–144. Springer, 2023

    2023

  19. [27]

    Diffusion models for counterfactual expla- nations

    Guillaume Jeanneret, Lo ¨ ıc Simon, and Fr´ ed´ eric Jurie. Diffusion models for counterfactual expla- nations. In Proceedings of the Asian Conference on Computer Vision , pages 858–876, 2022

    2022

  20. [28]

    Fast diffusion-based counterfactuals for shortcut re- moval and generation

    Nina Weng, Paraskevas Pegios, Eike Petersen, Aasa Feragen, and Siavash Bigdeli. Fast diffusion-based counterfactuals for shortcut re- moval and generation. In European Conference on Computer Vision , pages 338–357. Springer, 2025

    2025

  21. [29]

    Diffusion-based iterative counterfactual expla- nations for fetal ultrasound image quality assess- ment

    Paraskevas Pegios, Manxi Lin, Nina Weng, Morten Bo Søndergaard Svendsen, Zahra Bashir, Siavash Bigdeli, Anders Nymark Chris- tensen, Martin Tolsgaard, and Aasa Feragen. Diffusion-based iterative counterfactual expla- nations for fetal ultrasound image quality assess- ment. arX...

    2024 arXiv

  22. [30]

    Global counterfactual explainer for graph neural networks

    Zexi Huang, Mert Kosan, Sourav Medya, Sayan Ranu, and Ambuj Singh. Global counterfactual explainer for graph neural networks. In Pro- ceedings of the Sixteenth ACM International Conference on Web Search and Data Mining , pages 141–149, 2023

    2023

  23. [31]

    D4explainer: In-distribution ex- planations of graph neural network via dis- crete denoising diffusion

    Jialin Chen, Shirley Wu, Abhijit Gupta, and Rex Ying. D4explainer: In-distribution ex- planations of graph neural network via dis- crete denoising diffusion. In A. Oh, T. Neu- mann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, editors, Advances in Neural Infor- mation Pro...

    2023

  24. [32]

    Robust counterfactual explana- tions on graph neural networks

    Mohit Bajaj, Lingyang Chu, Zi Yu Xue, Jian Pei, Lanjun Wang, Peter Cho-Ho Lam, and Yong Zhang. Robust counterfactual explana- tions on graph neural networks. Advances in Neural Information Processing Systems , 34: 5644–5655, 2021

    2021

  25. [33]

    Clear: Gen- erative counterfactual explanations on graphs

    Jing Ma, Ruocheng Guo, Saumitra Mishra, Aidong Zhang, and Jundong Li. Clear: Gen- erative counterfactual explanations on graphs. Advances in neural information processing sys- tems, 35:25895–25907, 2022

    2022

  26. [34]

    GraphV AE: Towards generation of small graphs using variational autoencoders

    Martin Simonovsky and Nikos Komodakis. GraphV AE: Towards generation of small graphs using variational autoencoders. In Artifi- cial Neural Networks and Machine Learning – ICANN 2018 , pages 412–422. Springer Inter- national Publishing, 2018

    2018

  27. [35]

    Auto- Encoding variational bayes

    Diederik P Kingma and Max Welling. Auto- Encoding variational bayes. December 2013

    2013

  28. [36]

    Variational graph Auto-Encoders

    Thomas N Kipf and Max Welling. Variational graph Auto-Encoders. November 2016

    2016

  29. [37]

    On the universality of invariant networks

    Haggai Maron, Ethan Fetaya, Nimrod Segol, and Yaron Lipman. On the universality of invariant networks. In International conference on machine learning , pages 4363–4371. PMLR, 2019. 8

    2019

  30. [38]

    Permutation equivariant layers for higher order interactions

    Horace Pan and Risi Kondor. Permutation equivariant layers for higher order interactions. In International Conference on Artificial Intel- ligence and Statistics , pages 5987–6001. PMLR, 2022

    2022

  31. [39]

    Invariant and equivariant graph networks

    Haggai Maron, Heli Ben-Hamu, Nadav Shamir, and Yaron Lipman. Invariant and equivariant graph networks. In International Conference on Learning Representations, 2018

    2018

  32. [40]

    Multireso- lution graph variational autoencoder

    Truong Son Hy and Risi Kondor. Multireso- lution graph variational autoencoder. arXiv preprint arXiv:2106.00967, 2021

    2021 arXiv

  33. [41]

    Geometric deep learning: Grids, groups, graphs, geodesics, and gauges

    Michael M Bronstein, Joan Bruna, Taco Cohen, and Petar Veliˇ ckovi´ c. Geometric deep learning: Grids, groups, graphs, geodesics, and gauges. April 2021

    2021

  34. [42]

    The general theory of permu- tation equivarant neural networks and higher order graph variational encoders

    Erik Henning Thiede, Truong-Son Hy, and Risi Kondor. The general theory of permu- tation equivarant neural networks and higher order graph variational encoders. CoRR, abs/2004.03990, 2020. URL https://arxiv. org/abs/2004.03990

    2004 arXiv

  35. [43]

    beta-vae: Learning basic visual concepts with a constrained variational framework

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

    2017

  36. [44]

    Categorical reparameterization with gumbel- softmax

    Eric Jang, Shixiang Gu, and Ben Poole. Categorical reparameterization with gumbel- softmax. arXiv [stat.ML] , November 2016

    2016

  37. [45]

    Comparison of descriptor spaces for chemical compound retrieval and classification

    Nikil Wale, Ian A Watson, and George Karypis. Comparison of descriptor spaces for chemical compound retrieval and classification. Knowl- edge and Information Systems , 14:347–375, 2008

    2008

  38. [46]

    Derivation and validation of toxicophores for mutagenicity prediction

    Jeroen Kazius, Ross McGuire, and Roberta Bursi. Derivation and validation of toxicophores for mutagenicity prediction. Journal of medici- nal chemistry , 48(1):312–320, 2005

    2005

  39. [47]

    Iam graph database repository for graph based pattern recognition and machine learning

    Kaspar Riesen and Horst Bunke. Iam graph database repository for graph based pattern recognition and machine learning. In Structural, Syntactic, and Statistical Pattern Recognition: Joint IAPR International Workshop, SSPR & SPR 2008, Orlando, USA, December 4-6, 2008. Proceedin...

    2008

  40. [48]

    Understanding isomorphism bias in graph data sets

    Sergei Ivanov, Sergei Sviridov, and Evgeny Bur- naev. Understanding isomorphism bias in graph data sets. arXiv preprint arXiv:1910.12091 , 2019

    1910 arXiv

  41. [49]

    Kriege, Franka Bause, Kristian Kersting, Petra Mutzel, and Marion Neumann

    Christopher Morris, Nils M. Kriege, Franka Bause, Kristian Kersting, Petra Mutzel, and Marion Neumann. Tudataset: A collection of benchmark datasets for learning with graphs. In ICML 2020 Workshop on Graph Representation Learning and Beyond (GRL+ 2020), 2020. URL www.graphlearning.io

    2020

  42. [50]

    Global counterfactual explainer for graph neural networks

    Zexi Huang, Mert Kosan, Sourav Medya, Sayan Ranu, and Ambuj Singh. Global counterfactual explainer for graph neural networks. In Proceed- ings of the Sixteenth ACM International Confer- ence on Web Search and Data Mining , WSDM ’23, pages 141–149. Association for Computing Mac...

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.