Pith. sign in

REVIEW 3 major objections 6 minor 46 references

Towards Fair Graph Neural Networks via Graph Counterfactual without Sensitive Attributes

T0 review · 3 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 The paper claims that GNN fairness can be enforced without sensitive attributes by learning pseudo-sensitive proxies and minimizing a mutual-information upper bound on the hidden protected attribute.

desk verdict The method is a reasonable combination of known ideas with modest empirical gains, but the theoretical guarantee that anchors the paper doesn't survive contact with the actual architecture. read the letter →

arxiv 2412.09947 v1 pith:OCHYWM3L submitted 2024-12-13 cs.LG

Xuemin Wang , Tianlong Gu , Xuguang Bao , Liang Chang This is my paper · ORCID
classification cs.LG
keywords graphneuralnetworkscounterfactualfairnesswithoutsensitiveattributespseudo-sensitivenodeclassificationrepresentationlearning
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

Fairwos aims to make graph neural network predictions fair even when sensitive attributes such as race or gender are unavailable at training time. It first learns low-dimensional pseudo-sensitive attributes from the non-sensitive features and graph structure, then finds graph counterfactuals by searching the real dataset for same-label nodes with different pseudo-attributes, and trains the GNN so original and counterfactual embeddings are close. The paper's central theoretical claim is that reducing the mutual information between pseudo-sensitive attributes and the learned representation also reduces an upper bound on the mutual information between the hidden sensitive attribute and the prediction. This matters because privacy laws often forbid collecting protected attributes, yet unfairness can still flow through correlated features and network structure. If the bound holds, Fairwos offers a principled, attribute-free route to counterfactual-style fairness for GNNs.

What carries the argument

The load-bearing objects are the pseudo-sensitive attributes $x^0_u = \mathrm{Encoder}(G; \Theta^*_{\mathrm{enc}})$, low-dimensional representations of non-sensitive features and graph structure, and the counterfactual-consistency loss that minimizes the distance between the embedding of the original subgraph and the embeddings of its top-$K$ real-data counterfactuals. The identity that carries the argument is Theorem 1's mutual-information chain, which connects the disparity loss to the hidden sensitive attribute's influence on predictions. A per-attribute weight $\lambda_i$, updated by a KKT-based scheme, concentrates fairness regularization on the pseudo-attributes that contribute most to bias while preserving utility.

What would settle it

Train Fairwos on a public graph benchmark with the true sensitive attribute withheld and with it available only for evaluation; then estimate $I(s;z_u)$ and $\sum_t I(x^0_t;z_u)$ on test nodes and test conditional independence of $z_u$ and $G_u$ given $x^0_u$. A trained model that violates the inequality, or whose embedding still depends on graph structure beyond the pseudo-attributes, shows the theorem's hypothesis is not met.

Watch

Extended reading notes

Core claim

The central discovery, on the paper's own terms, is that the gap between the hidden sensitive attribute $s$ and the prediction $\hat y_u$ is controlled by the dependence of the GNN embedding $z_u$ on the pseudo-sensitive attributes $x^0_u$. Theorem 1 states the chain $0 \le I(s;\hat y_u) \le I(s;z_u) \le I(G_u;z_u) \le \sum_t I(x^0_t;z_u)$, and the framework is built to shrink the rightmost term: learn $x^0$ from non-sensitive attributes and graph structure, find real graph counterfactuals by nearest-neighbor search under the constraints $y_i = y_j$ and $x^0_i \neq x^0_j$, and train with a weighted disparity loss between original and counterfactual embeddings. Because the counterfactual comes from real data rather than by perturbing a protected attribute, the method avoids unrealistic counterfactuals while still attacking the root cause of bias.

Load-bearing premise

The guarantee assumes that a node's learned embedding depends on its graph neighborhood only through the pseudo-sensitive attributes, even though the actual GNN consumes the full graph, so this mediation link is asserted rather than tested.

Editorial extensions

If this is right

  • If the representation ignores the pseudo-sensitive attributes, Theorem 1 forces the prediction's dependence on the hidden sensitive attribute toward zero through the chain of inequalities.
  • Sensitive attributes are not needed anywhere in training, only for evaluating fairness on the test set.
  • Counterfactuals are selected from the real dataset, so the method does not rely on unrealistic perturbations of a protected variable.
  • The per-pseudo-attribute weights $\lambda_i$ are re-estimated during training, so fairness regularization concentrates on the attributes most responsible for bias while preserving utility.
  • The framework is compatible with different GNN backbones (GCN and GIN are tested), so it can be added to existing architectures.

Reading between the lines

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

  • Beyond the paper, the bound suggests a practical fairness audit: with true protected labels available only for evaluation, one can estimate $I(s;z_u)$ and the pseudo-attribute sum on held-out nodes and use any violation as a warning that the fairness guarantee is not operating.
  • Beyond the paper, the real-data counterfactual search could replace sensitive-attribute perturbation in other counterfactual fairness pipelines, since it only needs pseudo-attributes and labels.
  • Beyond the paper, the dynamic weight update defines a simplex-constrained trade-off; generalizing $\lambda_i$ to continuous or probabilistic pseudo-attributes would let the same idea apply to non-categorical protected proxies.
  • Beyond the paper, the Markov-chain assumption could be stress-tested by measuring residual dependence of $z_u$ on the graph after conditioning on $x^0_u$; if the residual is large, an additional debiasing step on graph structure would be needed.
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 Fairwos, a framework for training fair graph neural networks when sensitive attributes are not available during training. The method has three main components: an encoder that maps node attributes and graph structure to low-dimensional pseudo-sensitive attributes X(0); a counterfactual search procedure that, for each node, finds real nodes with the same label but different pseudo-sensitive attributes and uses those as graph counterfactuals; and a fair representation learning step that penalizes the distance between original and counterfactual embeddings, with per-attribute weights updated through a KKT-based scheme. The authors claim a theoretical guarantee that reducing the mutual information between pseudo-sensitive attributes and GNN embeddings bounds the unfairness of predictions, and they report experiments on six datasets showing improved fairness-utility trade-offs relative to several baselines.

Significance. If the theoretical guarantee were valid, the paper would make a meaningful contribution: it addresses a practical and timely problem, fairness without sensitive attributes for graph data, and it proposes a concrete mechanism, pseudo-sensitive attributes generated by an encoder, with graph counterfactuals selected from real data rather than by unrealistic perturbation. The experimental study is broad, covering six datasets and two backbone GNNs, and the reported results are generally favorable. However, the central theoretical claim is the load-bearing justification for the method, and it rests on a conditional-independence assumption that the proposed architecture does not satisfy. The empirical results alone, while suggestive, do not establish the fairness mechanism, especially because the standard deviations in Table II are large and no statistical significance tests are reported. The paper should therefore not be accepted in its current form.

major comments (3)
  1. [Section IV.A, Theorem 1] The proof of Theorem 1 relies on the Markov chain s -> G_u -> x0_u -> z_u -> yhat_u, which requires z_u to be conditionally independent of G_u given x0_u. This does not hold for the actual architecture: in Section III.C, the GNN classifier takes G = (V, E, X(0)) as input and computes z_u through message passing over the graph adjacency (Eqs. 7-8), so z_u carries information about G_u beyond x0_u. Consequently, the inequality I(s; yhat_u) <= I(G_u; z_u) <= sum_t I(x0_t; z_u) does not follow from the stated assumptions. Since the abstract and conclusion rest on this theoretical guarantee, the central justification for Fairwos is unsupported unless the mediation assumption is derived from the architecture or explicitly tested.
  2. [Section IV.A, Theorem 2] The proof of Theorem 2 drops the neighborhood aggregation term in Eq. (28) with the claim that the second term is 0. This is inconsistent with the algorithm: graph counterfactuals are found by searching for different real nodes (Eq. 12), so the neighborhoods N(u) and N(tilde u) are generally different, and the aggregation difference term is generally nonzero. Additionally, the proof assumes ||tilde x0_u - x0_u|| = 1 for a single perturbed attribute, but the real-node search changes the pseudo-sensitive attribute vector in a data-dependent way, not as a unit-norm single-coordinate perturbation. Thus Theorem 2 does not provide a valid bound for the actual counterfactual generation procedure.
  3. [Section III.D and fair representation learning] There is a circularity concern in the fairness argument. The pseudo-sensitive attributes X(0) are produced by an encoder trained with labels (Eqs. 4-5), and the counterfactual search in Eq. (12) uses both these pseudo-sensitive attributes and the (pseudo-)labels. The fairness loss in Eq. (15) then aligns original and counterfactual embeddings, effectively encouraging the model to be invariant to variations in the same pseudo-sensitive attributes that were used to select the counterfactuals. This only reduces unfairness if pseudo-sensitive attributes mediate all sensitive influence on predictions. That mediation assumption is precisely what the paper needs to establish, but it is asserted rather than proven or tested.
minor comments (6)
  1. [Section III.F, Algorithm 1] Algorithm 1 contains incorrect equation references: line 1 says 'optimizing (2)' but the encoder is trained with Eq. (5), and line 4 says 'optimizing Eq. (7)' but the GNN classifier loss is Eq. (10).
  2. [Section II.B] Definitions labeled 3.1 and 3.2 appear in Section II.B but are numbered as if they belong to Section III; the definition numbering should be corrected.
  3. [Section III.F, Eqs. (19)-(21)] The KKT derivation for the lambda update is missing the factor alpha: the first-order condition of Eq. (19) is alpha D_i^K + 2 lambda_i - a_i + b = 0, but Eq. (20) omits alpha, and this omission propagates to the closed-form update in Eq. (24).
  4. [Section IV.A, Theorem 1 statement] The notation in Theorem 1 is inconsistent: the final bound is written as sum_t I(x0_i; z_u), mixing indices i and t, and the role of T is not defined precisely.
  5. [Table II] The first dataset column is labeled 'Recidivism' while the dataset is called 'Bail' in Table I and Section V.A.1; the table header should be consistent with the dataset name.
  6. [Section V.B] The claims of superior performance are based on means with overlapping standard deviations (e.g., Fairwos versus FairGKD\S on several datasets in Table II); reporting statistical significance tests or confidence intervals would make the empirical comparison more convincing.

Circularity Check

1 steps flagged · score 6.0 of 10

Theorem 1's fairness bound is derived by assuming the very mediation property the architecture does not implement: z is computed from the full graph G, not from x0 alone.

  1. self definitional [Section IV.A, Theorem 1 proof, contrasted with Section III.B Eq. (6) and Section III.C Eqs. (7)-(8)]
    "Due to the existence of an encoder, the Markov chain is represented as s → Gu → x0_u → zu → ˆyu. ... Furthermore I (Gu; zu) ≤ P_T_t I (x0_i; zu) indicate that the bias is determined by x0_u."

    The proof's Markov chain s → G → x0 → z → ŷ requires z_u ⊥ G_u | x0_u, i.e., that the pseudo-sensitive attributes x0_u mediate all graph information used by the GNN. But the paper's own architecture does not implement this: Eq. (6) computes x0 = Encoder(G), and then the GNN classifier takes the full graph G = (V, E, X(0)) as input, with every layer aggregating over the node's neighborhoods (Eqs. 7-8). Thus z is a function of both G and x0. The bound I(G_u; z_u) ≤ Σ_t I(x0_t; z_u) is exactly the claim that x0 is a sufficient statistic for the graph information in z, which is the premise the theorem needs to prove.

full rationale

The paper has a genuine empirical component: the experiments compare Fairwos against baselines on six datasets, with ablations and hyperparameter studies, and those parts are self-contained and not circular. However, the central theoretical claim advertised in the abstract and conclusion — that minimizing the relation between pseudo-sensitive attributes and predictions provably enables fairness — rests on Theorem 1. The proof's crucial step is the asserted Markov chain s → G_u → x0_u → z_u → ŷ_u, which requires the GNN embedding z_u to depend on the graph only through the encoder output x0_u. The architecture contradicts this: x0_u is itself computed from G, and then the GNN consumes both the processed node features and the graph adjacency/message-passing structure at every layer (Eqs. 7-8). Therefore the inequality I(G_u; z_u) ≤ Σ_t I(x0_t; z_u) does not follow from the model; it is an unproven sufficiency assumption. Because the paper defines pseudo-sensitive attributes as representations of the graph structure and then re-imports that definition as the Markov chain in the proof, the central theoretical guarantee is circular rather than derived. Theorem 2 also contains an unsupported step (dropping the neighborhood-difference term by declaring it zero), but I do not treat that as circularity; it is a separate correctness concern. No self-citation chain is load-bearing here, so the score reflects the self-definitional nature of the main theoretical result while acknowledging the independent empirical evaluation.

Assumptions & free parameters 3 free parameters · 5 assumptions · 1 invented entities

The central claim rests on the assumption that encoder-derived pseudo-sensitive attributes faithfully proxy the unobserved sensitive attribute and that a Markov chain through these attributes bounds the unfairness. Both assumptions are unproven; the Markov chain is false for the actual GNN architecture because the classifier consumes the full graph. The counterfactual search further assumes that real same-label nodes can serve as counterfactuals, and Theorem 2 assumes unit-norm perturbations and a zero neighborhood term. These load-bearing assumptions receive no quantitative validation beyond a t-SNE figure.

free parameters (3)
  • Encoder output dimension (number of pseudo-sensitive attributes) = Grid searched over {2, 4, 8, 16, 32}; validation-selected
    The dimensionality of X(0) determines the granularity of pseudo-sensitive attributes and strongly affects the fairness-utility trade-off (Section V.D).
  • α (fairness regularization weight) = Main experiments searched over {0.01, 0.05, 1, 2, 5}; sensitivity over {0.01, 0.02, 0.04, 0.08}
    Controls the strength of the representation-consistency term in Eq. (15).
  • K (number of graph counterfactuals per node) = Main experiments searched over {1, 2, 5, 10, 20}; sensitivity over {1, 2, 3, 4}
    Number of nearest same-label nodes with different pseudo-sensitive attributes used in the fairness loss; strongly affects utility.
assumptions (5)
  • domain assumption Sensitive attributes influence predictions only through non-sensitive attributes and graph structure, and the encoder output X(0) captures this influence.
    Stated in Section III.B and Fig. 3; used to justify treating every dimension of X(0) as a pseudo-sensitive attribute. No quantitative validation that X(0) tracks the sensitive attribute.
  • ad hoc to paper Markov chain s → G_u → x0_u → z_u → ŷ_u holds, i.e., z_u is conditionally independent of G_u given x0_u.
    Invoked in the proof of Theorem 1 (Section IV.A). False in the actual model because the GNN consumes the graph directly, so z_u depends on G_u beyond x0_u.
  • ad hoc to paper Minimizing L2 distance between original and counterfactual embeddings reduces mutual information I(x0_i; z).
    Asserted in Section IV.A after Theorem 1; no derivation connects the distance loss to mutual information.
  • ad hoc to paper The neighbor aggregation difference term in Eq. (28) is zero.
    Assumed in the proof of Theorem 2 to drop the second term; in general Σ_{v∈N(ũ)} z^{k-1}_v ≠ Σ_{v∈N(u)} z^{k-1}_v, and for different real nodes the neighborhoods differ.
  • domain assumption Real nodes with the same label but different pseudo-sensitive attributes serve as valid graph counterfactuals.
    Section III.D: the counterfactual search selects real nodes; this equates a factual different individual with a counterfactual of the original, which is not a counterfactual in the causal sense.
invented entities (1)
  • Pseudo-sensitive attributes (X(0))
    purpose: Low-dimensional encoder outputs used as proxies for unobserved sensitive attributes to guide counterfactual search and fairness regularization.
    The paper provides only t-SNE visualization (Fig. 7) that clusters correlate with sensitive groups on two datasets; no quantitative assessment (e.g., prediction accuracy of sensitive attribute from X(0)) and no causal validation. The construct is introduced by the paper but lacks external falsifiable support.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Fair Graph Neural Networks via Graph Counterfactual without Sensitive Attributes." pith.science (2026). https://pith.science/paper/OCHYWM3L

@misc{pith2026241209947,
  author       = {Pith},
  title        = {Pith review of: Towards Fair Graph Neural Networks via Graph Counterfactual without Sensitive Attributes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OCHYWM3L}},
  note         = {Machine review of arXiv:2412.09947}
}
read the original abstract

Graph-structured data is ubiquitous in today's connected world, driving extensive research in graph analysis. Graph Neural Networks (GNNs) have shown great success in this field, leading to growing interest in developing fair GNNs for critical applications. However, most existing fair GNNs focus on statistical fairness notions, which may be insufficient when dealing with statistical anomalies. Hence, motivated by causal theory, there has been growing attention to mitigating root causes of unfairness utilizing graph counterfactuals. Unfortunately, existing methods for generating graph counterfactuals invariably require the sensitive attribute. Nevertheless, in many real-world applications, it is usually infeasible to obtain sensitive attributes due to privacy or legal issues, which challenge existing methods. In this paper, we propose a framework named Fairwos (improving Fairness without sensitive attributes). In particular, we first propose a mechanism to generate pseudo-sensitive attributes to remedy the problem of missing sensitive attributes, and then design a strategy for finding graph counterfactuals from the real dataset. To train fair GNNs, we propose a method to ensure that the embeddings from the original data are consistent with those from the graph counterfactuals, and dynamically adjust the weight of each pseudo-sensitive attribute to balance its contribution to fairness and utility. Furthermore, we theoretically demonstrate that minimizing the relation between these pseudo-sensitive attributes and the prediction can enable the fairness of GNNs. Experimental results on six real-world datasets show that our approach outperforms state-of-the-art methods in balancing utility and fairness.

Figures

Figures reproduced from arXiv: 2412.09947 by the authors.

Figure 1
Figure 1. Running example of node classification in a loan [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. An illustration of our proposed framework [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. A causal relationship exists between sensitive attributes [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Comparisons between Fairwos and its variants on [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Impacts of the dimension of the encoder to Fairwos [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Hyper-parameter study on Bail dataset (a) NBA dataset (b) Occupation dataset [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Visualizations of pseudo-Sensitive attributes for NBA [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 42 canonical work pages

  1. [1]

    Mhednbi, and S

    A. Mhednbi, and S. Salihoglu, ”Modern Techniques for Querying Graph- Structured Relations: Foundations, System Implementations, and Open Challenges,” in Proc. VLDB Endow. (VLDB) , 2022, pp. 3762-3765

    work page 2022

  2. [2]

    Z. Li, X. Shen, Y . Jiao, X. Pan, P. Zhou, X. Meng, C. Yao, and J. Bu, ”Hierarchical Bipartite Graph Neural Networks: Towards Large-Scale E-commerce Applications,” in Proc. 36th IEEE Int. Conf. Data Eng. (ICDE), 2020, pp. 1677-1688

    work page 2020

  3. [3]

    Z. Wang, T. Xia, R. Jiang, X Liu, K. Kim, X. Song, and R. Shibasaki, ”Forecasting Ambulance Demand with Profiled Human Mobility via Heterogeneous Multi-Graph Neural Networks,” in Proc. 37th IEEE Int. Conf. Data Eng. (ICDE) , 2021, pp. 1751-1762

    work page 2021

  4. [4]

    H. Yuan, G. Li, Z. Bao, and L. Feng, ”An Effective Joint Prediction Model for Travel Demands and Traffic Flows,” in Proc. 37th IEEE Int. Conf. Data Eng. (ICDE) , 2021, pp. 348-359

    work page 2021

  5. [5]

    K. M. Saifuddin, B. Bumgardner, F. Tanvir, and E. Akbas, ”HyGNN: Drug-Drug Interaction Prediction via Hypergraph Neural Network,” in Proc. 39th IEEE Int. Conf. Data Eng. (ICDE) , 2023, pp. 1503-1516

    work page 2023

  6. [6]

    W. Chen, H. Wan, Y . Wu, S. Zhao, J. Cheng, Y . Li, and Y . Lin, ”Local- global history-aware contrastive learning for temporal knowledge graph reasoning,” in Proc. 40th IEEE Int. Conf. Data Eng. (ICDE) , 2024, pp. 733-746

    work page 2024

  7. [7]

    T. Peng, W. Wu, H. Yuan, Z. Bao, P. Zhao, X. Yu, X. Lin, Y . Liang, and Y . Pu, ”GraphRARE: Reinforcement Learning Enhanced Graph Neural Network with Relative Entropy,” in Proc. 40th IEEE Int. Conf. Data Eng. (ICDE), 2024, pp. 2489-2502

    work page 2024

  8. [8]

    H. Chen, H. Yin, X. Sun, T. Chen, B. Gabrys, and K. Musial, ”Multi- level Graph Convolutional Networks for Cross-platform Anchor Link Prediction,” in Proc. 20th ACM SIGKDD Conf. Knowl. Discov. Data Min (KDD), 2020, pp. 1503-1511

    work page 2020

Show all 46 references
  1. [9]

    Wang, S.Di, L

    Z. Wang, S.Di, L. Chen, and X. Zhou, ”Search to Fine-Tune Pre-Trained Graph Neural Networks for Graph-Level Tasks,” Proc. 40th IEEE Int. Conf. Data Eng. (ICDE) , 2024, pp. 2805-2819

    2024

  2. [10]

    Tsitsulin, J

    A. Tsitsulin, J. Perozzi, and E. M ¨uller, ”Graph Clustering with Graph Neural Networks,” J. Mach. Learn. Res., vol. 24, pp. 127:1-127:21, 2023

    2023

  3. [11]

    H. Zhao, Q. Yao, and W. Tu, ”Search to aggregate neighborhood for graph neural network,” in Proc. 37th IEEE Int. Conf. Data Eng. , 2021, pp. 552–563

    2021

  4. [12]

    Agarwal, H

    C. Agarwal, H. Lakkaraju, and M. Zitnik, ”Towards a Unified Frame- work for Fair and Stable Graph Representation Learning,” in Proc. Thirty-Seventh Conf. Uncertainty in Artificial Intelligence , 2021, pp. 2114-2124

    2021

  5. [13]

    Dai and S

    E. Dai and S. Wang, ”Learning Fair Graph Neural Networks With Lim- ited and Private Sensitive Attribute Information,” IEEE Trans. Knowl. Data Eng., vol. 35, no. 7, pp. 7103-7117, 2023

    2023

  6. [14]

    Y . Wang, Y . Zhao, Y . Dong, H. Chen, J. Li, and T. Derr, ”Improving Fairness in Graph Neural Networks via Mitigating Sensitive Attribute Leakage,” in Proc. 28th ACM SIGKDD Conf. Knowl. Discov. Data Min., 2022, pp. 1938-1948

    2022

  7. [15]

    C. Yang, J. Liu, Y . Yan, and C. Shi, ”FairSIN: Achieving Fairness in Graph Neural Networks through Sensitive Information Neutraliza- tion,” in Proc. 38th AAAI Conference on Artificial Intelligence, 36th Conference on Innovative Applications of Artificial Intelligence, 14th Sy...

    2024

  8. [16]

    Z. Guo, J. Li, T. Xiao, Y . Ma, and S. Wang, ”Towards Fair Graph Neural Networks via Graph Counterfactual,” in Proc. 32nd ACM Int. Conf. Inf. Knowl. Manage., 2023, pp. 669-678

    2023

  9. [17]

    J. Hang, Z. Dong, H. Zhao, X. Song, P. Wang, and H. Zhu, ”Outside In: Market-aware Heterogeneous Graph Neural Network for Employee Turnover Prediction,” in Proc. 15th ACM Int. Conf. Web Search Data Mining, 2022, pp. 353-362

    2022

  10. [18]

    Cheng, X

    D. Cheng, X. Wang, Y . Zhang, and L. Zhang, ”Graph Neural Network for Fraud Detection via Spatial-Temporal Attention,” IEEE Trans. Knowl. Data Eng., vol. 34, no. 8, pp. 3800-3813, 2022

    2022

  11. [19]

    Loveland, J

    D. Loveland, J. Pan, A. F. Bhathena, and Y . Lu, ”FairEdit: Preserving Fairness in Graph Neural Networks through Greedy Graph Editing,” arXiv preprint arXiv: 2201.03681, 2022

    2022 arXiv

  12. [20]

    M. A. Prado-Romero, B. Prenkaj, G. Stilo, F. Giannotti, ”A Survey on Graph Counterfactual Explanations: Definitions, Methods, Evaluation, and Research Challenges,” ACM Comput. Surv., vol 56, no. 7, pp. 171:1– 171:37

  13. [21]

    J. Ma, R. Guo, M. Wan, L. Yang, A. Zhang, and J. Li, ”Learning Fair Node Representations with Graph Counterfactual Fairness,” in Proc. 15th ACM Int. Conf. Web Search Data Mining , 2022, pp. 695-703

    2022

  14. [22]

    Coston, K

    A. Coston, K. N. Ramamurthy, D. Wei, K. R. Varshney, S. Speakman, Z. Mustahsan, and S. Chakraborty, ”Fair Transfer Learning with Missing Protected Attributes,” in Proc. 2019 AAAI/ACM Conf. AI, Ethics, Soc. , 2019, pp. 91-98

    2019

  15. [23]

    Lahoti, A

    P. Lahoti, A. Beutel, J. Chen, K. Lee, F. Prost, N. Thain, X. Wang, and E. H. Chi, ”Fairness without Demographics through Adversarially Reweighted Learning,” in Advances in Neural Inf. Process. Syst. 33 , 2020

    2020

  16. [24]

    T. Zhao, E. Dai, K. Shu, and S. Wang, ”Towards Fair Classifiers Without Sensitive Attributes: Exploring Biases in Related attributes,” in Proc. 15th ACM Int. Conf. Web Search Data Mining , 2022, pp. 1433-1442

    2022

  17. [25]

    V oigt and A

    P. V oigt and A. V on dem Bussche, The EU General Data Protection Regulation (GDPR). Cham: Springer International Publishing, 2017

    2017

  18. [26]

    H. Zhu, E. Dai, H. Liu, and S. Wang, ”Learning fair models without sensitive attributes: A generative approach,” Neurocomputing, vol. 561, pp. 126841, 2023

    2023

  19. [27]

    Chai and X

    J. Chai and X. Wang, ”Self-Supervised Fair Representation Learning without Demographics,” in Advances in Neural Inf. Process. Syst. 35 , 2022

    2022

  20. [28]

    Y . Zhu, J. Li, L. Chen, and Z. Zheng, ”The Devil is in the Data: Learning Fair Graph Neural Networks via Partial Knowledge Distillation,” in Proc. 17th ACM Int. Conf. Web Search Data Mining , Merida, Mexico, 2024, pp. 1012-1021, doi: 10.1145/3616855.3635768

    2024

  21. [29]

    Dwork, M

    C. Dwork, M. Hardt, T. Pitassi, O. Reingold, and R. S. Zemel, ”Fairness through awareness,” in Proc. Innovations in Theoretical Com- puter Science 2012 , Cambridge, MA, USA, 2012, pp. 214-226, doi: 10.1145/2090236.2090255

    2012

  22. [30]

    Hardt, E

    M. Hardt, E. Price, and N. Srebro, ”Equality of Opportunity in Super- vised Learning,” in Proc. Advances in Neural Inf. Process. Syst. 29 , 2016, pp. 3315-3323

    2016

  23. [31]

    G. V . Demirci, A. Haldar, and H. Ferhatosmanoglu, ”Scalable Graph Convolutional Network Training on Distributed-Memory Systems,” Proc. VLDB Endow., vol. 16, no. 4, pp. 711-724, 2022

    2022

  24. [32]

    K. Xu, W. Hu, J. Leskovec, and S. Jegelka, ”How Powerful are Graph Neural Networks?,” in Proc. 7th Int. Conf. Learn. Representations, 2019

    2019

  25. [33]

    K. L. Jordan and T. L. Freiburger, ”The Effect of Race/Ethnicity on Sentencing: Examining Sentence Type, Jail Length, and Prison Length,” J. Ethn. Crim. Justice , vol. 13, no. 3, pp. 179-196, 2015

    2015

  26. [34]

    Yeh and C.-H

    I.-C. Yeh and C.-H. Lien, ”The comparisons of data mining techniques for the predictive accuracy of probability of default of credit card clients,” Expert Syst. Appl. , vol. 36, no. 2, pp. 2473-2480, 2009

    2009

  27. [35]

    Takac and M

    L. Takac and M. Zabovsky, ”Data analysis in public social networks,” in Int. Sci. Conf. Workshop Present Day Trends Innov., 2012, vol. 1, no. 6

    2012

  28. [36]

    X. Qian, Z. Guo, J. Li, H. Mao, B. Li, S. Wang, and Y . Ma, ”Addressing Shortcomings in Fair Graph Learning Datasets: Towards a New Bench- mark,” CoRR, vol. abs/2403.06017, 2024

    2024 arXiv

  29. [37]

    R. S. Zemel, Y . Wu, K. Swersky, T. Pitassi, and C. Dwork, ”Learning Fair Representations,” in Proc. 30th Int. Conf. Mach. Learn. , 2013, pp. 325-333

    2013

  30. [38]

    Yan, H.-T

    S. Yan, H.-T. Kao, and E. Ferrara, ”Fair Class Balancing: Enhancing Model Fairness without Observing Sensitive Attributes,” in Proc. 29th ACM Int. Conf. Inf. Knowl. Manage. , 2020, pp. 1715-1724

    2020

  31. [39]

    Van, and G

    L. Van, and G. Hinton, ”Visualizing data using t-SNE,” J. Mach. Learn. Res., vol. 9, pp.2579-2605

  32. [40]

    Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, and P. S. Yu, ”A Comprehensive Survey on Graph Neural Networks,”IEEE Trans. Neural Networks Learn. Syst. , vol. 32, no. 1, pp. 4-24, 2021

    2021

  33. [41]

    Bruna, W

    J. Bruna, W. Zaremba, A. Szlam, and Y . LeCun, ”Spectral Networks and Locally Connected Networks on Graphs,” in Proc. 2nd Int. Conf. Learn. Representations, 2014

    2014

  34. [42]

    Velickovic, G

    P. Velickovic, G. Cucurull, A. Casanova, A. Romero, P. Li `o, and Y . Bengio, ”Graph Attention Networks,” in Proc. 6th Int. Conf. Learn. Representations, 2018

    2018

  35. [43]

    W. L. Hamilton, Z. Ying, and J. Leskovec, ”Inductive Representation Learning on Large Graphs,” in Advances in Neural Inf. Process. Syst. 30, 2017, pp. 1024-1034

    2017

  36. [44]

    Y . Dong, J. Kang, H. Tong, and J. Li, ”Individual Fairness for Graph Neural Networks: A Ranking based Approach,” in Proc. 27th ACM SIGKDD Conf. Knowl. Discov. Data Min. , 2021, pp. 300-310

    2021

  37. [45]

    Y . Dong, N. Liu, B. Jalaian, and J. Li, ”EDITS: Modeling and Mitigating Data Bias for Graph Neural Networks,” in Proc. ACM Web Conf., 2022, pp. 1259-1269

    2022

  38. [46]

    T. N. Kipf and M. Welling, ”Variational Graph Auto-Encoders,” CoRR, vol. abs/1611.07308, 2016

    2016 arXiv

Pith tools

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