Pith. sign in

REVIEW 6 major objections 6 minor 1 cited by

On the Benefits of Attribute-Driven Graph Domain Adaptation

T0 review · 6 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper argues that node attributes are a distinct, often dominant, driver of graph domain shift and derives a bound that makes attribute alignment part of the transfer guarantee.

desk verdict The paper's central theory is broken—the proof of Proposition 1 has a sign error—so the attribute-importance claim is unproven; the method and experiments are reasonable but the evaluation is too loose. read the letter →

arxiv 2502.06808 v3 pith:U3UOZHKT submitted 2025-02-04 cs.LG cs.AI

Ruiyi Fang , Bingheng Li , Zhao Kang , Qiuhao Zeng , Nima Hosseini Dashtbayaz , Ruizhi Pu , Boyu Wang , Charles Ling This is my paper · ORCID
classification cs.LGcs.AI
keywords graphdomainadaptationnodeattributesPAC-Bayesbounddiscrepancyneuralnetworksattributeshifttopologyfeature
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 claims that graph domain adaptation methods that only align topology miss half of the problem: node attributes are a separate, and often larger, source of cross-graph distribution shift. To show this, it proves Proposition 1, an upper bound on the PAC-Bayesian domain discrepancy that contains two explicit terms—one for structure $(A_S X_S)_i$ versus $(A_T X_T)_j$ and one for raw attributes $X^S_i$ versus $X^T_j$—and it offers an empirical comparison of average feature magnitudes suggesting the attribute term is the bigger gap. Based on that, it proposes GAA, a cross-channel model that aligns both a topology view and an attribute view, the latter built as a $k$-nearest-neighbour graph on node features, refined by a cross-view similarity matrix. If the claim is right, future GDA systems should allocate explicit capacity to attribute alignment rather than treating structure as the only domain difference.

What carries the argument

The load-bearing object is the expected-loss discrepancy $D^{\gamma/2}_{S,T}(P;\lambda)$ of Definition 1, a PAC-Bayesian measure of how much a classifier's expected margin loss changes between the source and target graphs. Proposition 1 bounds it by a sum over source–target node pairs of two squared Euclidean distances: $\|(A_S X_S)_i - (A_T X_T)_j\|_2^2$ for topology-filtered features and $\|X^S_i - X^T_j\|_2^2$ for raw attributes. That factorisation is what lets the paper say attribute divergence belongs in the transfer bound. On the method side, the attribute view is realised by a $k$-nearest-neighbour graph $\hat G$ built from $X$ via cosine similarity, and the cross-view similarity matrix $S$ then filters each channel's embeddings before the two MSE alignment losses are applied.

What would settle it

On synthetic graphs with constant attributes and varying topology (the paper's own SBM setup in Appendix F.2), compute the two sums in Proposition 1; if target error rises while the attribute term stays at zero and the topology term does not track the error, the bound as stated does not explain transfer difficulty.

Watch

Extended reading notes

Core claim

The paper's central claim is that node attributes are not a side issue for graph domain adaptation but a term in the generalization bound itself. Proposition 1 upper-bounds the domain discrepancy $D^{\gamma/2}_{S,T}(P;\lambda)$ by the sum over all source-target node pairs of the squared Euclidean distance between topology-filtered features $(A_S X_S)_i$ and $(A_T X_T)_j$, plus the squared distance between raw attributes $X^S_i$ and $X^T_j$. The paper reads this as proof that both graph structure and node attributes bound the transfer gap, and its companion empirical comparison of average feature magnitudes is offered as evidence that attribute divergence is the larger of the two. On that basis it proposes GAA, a domain-adaptation model with two channels—the original graph and a $k$-nearest-neighbour graph built from attributes—aligned by a cross-view similarity filter.

Load-bearing premise

The proof's key premise is the ordering $L^{\gamma/2}(h_j,k) \le L^\gamma(h_i,k)$, which lets the difference of expected losses be converted into a sum of node-pair distances; if that ordering fails for the classifier family, the bound does not follow.

Editorial extensions

If this is right

  • A GDA method that aligns only aggregated topology features leaves the attribute term of Proposition 1 untouched, so its target-risk guarantee is strictly weaker than one that also aligns raw node attributes.
  • The two-term bound gives a pre-training diagnostic: compute the two pairwise sums over source and target nodes to forecast how hard a given graph pair will be to adapt.
  • The empirical claim that attribute shift exceeds topology shift, if it holds on standard benchmarks, means attribute alignment should yield larger accuracy gains than structural alignment, matching the reported ablations.
  • GAA's architecture—original graph plus $k$NN feature graph, aligned with MSE losses and refined by a cross-view similarity matrix—is a concrete recipe for turning the bound into an algorithm.

Reading between the lines

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

  • We infer that the two-term bound supports a cheap source-selection heuristic: rank candidate source graphs by the ratio of attribute distance to topology distance before training any model.
  • We infer that if the attribute-dominance finding replicates on larger real-world graphs, GDA research effort should shift from designing structural alignment losses to learning transferable attribute representations; GAA is an early instance of that direction.
  • We infer a testable variant: rebuild or learn the $k$NN feature graph during training rather than fixing it, and check whether the attribute term of the bound shrinks further than with the fixed graph.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

6 major / 6 minor

Summary. The paper addresses graph domain adaptation (GDA) and argues that node attributes, not only graph topology, must be aligned across domains. It presents a PAC-Bayes generalization bound (Proposition 1) claiming to show that the domain discrepancy D^{γ/2}_{S,T}(P;λ) is upper-bounded by sums of squared differences of aggregated topology features and raw node attributes, and it reports empirical evidence (Figure 1) that attribute shift is larger than topology shift. Based on this motivation, the authors propose GAA, which builds a kNN feature graph, applies GCN encoders to both topology and feature views, refines the views with a cross-view similarity matrix, and aligns source and target embeddings via an attribute-driven loss plus adversarial domain adaptation. Experiments on Airport, Citation, Blog/Twitch, and MAG datasets report accuracy improvements over several baselines.

Significance. The question of whether node attributes or topology dominate graph domain shift is well motivated and could guide GDA algorithm design. The paper includes extensive benchmark comparisons, ablations, synthetic experiments probing attribute/topology divergence, and an efficiency analysis. If the theoretical bound and the empirical discrepancy measurement were correct, this would be a useful contribution. However, the central theoretical claim is not proven, the proposed objective contains a sign error and a dimension mismatch, and the empirical support for the main premise is based on a non-distributional summary statistic. These issues are load-bearing: the algorithm, the narrative, and the comparisons all depend on them. The paper cannot be recommended for publication in its present form.

major comments (6)
  1. [Appendix A, Eq. (28)] The step from Eq. (27) to Eq. (28) uses an inequality in the wrong direction. The quantity a = L^{γ/2}(h_j,k) − L^γ(h_i,k) is non-positive because γ/2 < γ, so for η_k(j) ∈ [0,1] one has η_k(j)a ≥ a, not η_k(j)a ≤ a. The text's justification (both factors upper-bounded by 1 and L^{γ/2} ≤ L^γ) would only support an upper bound if a were non-negative. As a concrete counterexample, take a = −1, η_k(j) = 0.5, η_k(i) = 0.6, and L^γ(h_i,k) = 1: the left-hand side of the per-k term in Eq. (28) is −0.6, while the proposed replacement is −0.99, contradicting the claimed inequality. Proposition 1 is therefore not proven.
  2. [Appendix A, Theorem 2 and final step of Proposition 2 proof] The proof of Theorem 2 contains unsubstantiated steps: the denominator in Eq. (21) is asserted to be bounded without proof or justification, and the transition from Eq. (21) to Eq. (22) is a Taylor-type expansion whose error terms are not controlled. More importantly, the proof of Proposition 2 ends by replacing the abstract aggregated features f_u and f_v with (A_S X_S)_i and X_i (or X_j), but no argument is given that the GNN features appearing in Theorem 2 coincide with these quantities or are even comparable to them. Consequently, the final bound in Eq. (24) does not follow from the preceding lemmas.
  3. [Eq. (12)] The attribute-driven alignment loss is defined as L_A = −(||att^S − att^T||_2^2 + ||att^S_f − att^T_f||_2^2). With this sign, minimizing L_A drives the source and target embeddings apart rather than aligning them. The paper refers to L_A as an MSE loss, but the formula is the negative of an MSE. Unless a gradient-reversal mechanism is intended for this specific term—which is not stated; the GRL is used only for L_D—the overall objective in Eq. (15) would maximize discrepancy in the attribute and topology views.
  4. [Eq. (11)] The Hadamard product att^S = att^S ⊙ S^S is dimensionally invalid. The attention output att^S is an N×d matrix (from softmax(K^T Q/√d)M^T with Q,K,M ∈ R^{d×N}), whereas S^S is an N×N similarity matrix. The same issue affects att^S_f ⊙ S^S, att^T ⊙ S^T, and att^T_f ⊙ S^T. The intended refinement operation—matrix multiplication, broadcasting, or a different dimension convention—must be specified.
  5. [Section 5.6 and Appendix D] Table 5 reports different values of α, β, τ, and k for every source–target task, and Appendix D motivates these choices from dataset properties. No validation procedure is described (e.g., a held-out target split, a model-selection criterion, or a fixed development protocol). Tuning on the test tasks makes the reported accuracies optimistic and the comparison against baselines unfair. The statement in Section 5.3 that 'we use the same parameter settings for all the cross-domain node classification methods' is contradicted by the per-task hyperparameters in Table 5.
  6. [Figure 1 and Appendix B] The comparison of 'topology' and 'attribute' feature values uses the mean absolute entry of A X and of the kNN-filtered feature matrix (F and F_f in Appendix B). This is a measure of average signal magnitude, not a discrepancy between the source and target distributions. Plotting these magnitudes per dataset does not establish that attribute shift is larger than topology shift in any distributional sense, so the empirical motivation in Section 1 is not supported.
minor comments (6)
  1. [General] Typos and formatting issues: 'further underscore' should be 'further underscores' (Abstract); 'Atrribute' in Figure 1 should be 'Attribute'; 'benchmark' in the Abstract should be plural; several table captions lack terminal periods.
  2. [Section 5.6] Section 5.6 refers to 'Figure.5' when discussing parameter sensitivity, but the parameter-sensitivity figure in the main text is Figure 4; Figure 5 is in the appendix. Please correct the cross-reference.
  3. [Proposition numbering] The discrepancy bound is labeled Proposition 1 in the main text but Proposition 2 in Appendix A; use consistent numbering to avoid confusion.
  4. [Definition 1] Definition 1 is used for two different notions: the Expected Loss Discrepancy in Section 3 and the Gaussian data assumption in Appendix A. Rename one to eliminate the conflict.
  5. [Appendix G] Appendix G reports a variant 'GAAo' without defining it; if it is an ablation or a more efficient version, it should be described before being compared against GAA.
  6. [Tables 2–4] The text states that 'some results are directly taken from (Shi et al., 2023; Pang et al., 2023)' but does not indicate which entries are borrowed; specifying the source of each baseline result would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: Proposition 1 is analytically derived, not fitted to its conclusion; empirical claims are independent observations.

full rationale

Proposition 1 is presented as an analytic PAC-Bayes-style bound: the discrepancy D^{γ/2}_{S,T}(P;λ) is defined through expected margin losses, and the proof attempts to upper-bound it by sums of squared feature differences using Theorem 2 and a Gaussian data assumption. Nothing in this chain is fitted to the quantity it claims to bound; the right-hand side is not defined in terms of D, and the bound is not produced by training the proposed model. The GAA loss in Eq. (12) is motivated by, and directly minimizes, the two terms appearing on the right-hand side of Proposition 1, but that is a design choice guided by the theory rather than a case of the "prediction" being equivalent to its input. The empirical claim that attribute shift is larger than topology shift is based on average feature magnitudes (Appendix B), not on the theoretical bound; it is an independent observation, even though its metric is questionable. Self-citations (e.g., Fang et al. 2022, Li et al. 2025, Qian et al. 2024) are used for auxiliary motivation and implementation details, not to exclude alternatives or to justify the central bound, so they are not load-bearing. Separately, the proof of Proposition 1 appears to contain a real algebraic error in the transition from Eq. (27) to Eq. (28): with a = L^{γ/2}(h_j,k) − L^γ(h_i,k) ≤ 0 and η_k(j) ∈ [0,1], η_k(j)·a ≥ a, so the claimed upper-bound replacement is invalid; and the final substitution of f_u by (A_S X_S)_i and X_i is unproved. These are correctness and soundness concerns, not circularity, and they do not raise the circularity score.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central proof relies on several unproved or unjustified assumptions, most notably the margin-loss comparison used in the wrong direction and the final leap from GNN features to AX and X terms. The method also introduces tuned hyperparameters per task, which act as free parameters in the reported results.

free parameters (4)
  • alpha (α) = per-task values from {0.005,0.01,0.1,0.5,1,5} (Table 5)
    Trade-off weight for the source classification loss; tuned separately for each experiment with no validation split reported.
  • beta (β) = per-task values from {0.005,0.01,0.1,0.5,1,5} (Table 5)
    Trade-off weight for the domain classifier loss; tuned per task without a validation split.
  • tau (τ) = per-task values from {0.005,0.01,0.1,0.5,1,5} (Table 5)
    Trade-off weight for the target entropy loss; tuned per task, and per-task values vary widely (e.g., 0.01 to 0.5).
  • k (kNN neighbors) = per-task values from {1,...,10}, typically 2-6 (Table 5)
    Number of neighbors in the feature graph; chosen per task based on test-set accuracy, another tuned quantity.
assumptions (6)
  • domain assumption Node features in each class are Gaussian with class-conditional means μ_i and common variance σ^2; classes are balanced.
    Assumed in Appendix A Definition 1 and used in the proof of Theorem 2 (Eqs. 16-22).
  • domain assumption Labels are independent given aggregated features Z = f(X,G).
    Assumed in Section 3 notation and used to define expected margin losses.
  • ad hoc to paper The inequality L^{γ/2}(h_j,k) ≤ L^γ(h_i,k) holds for all relevant i,j and is used to derive Eq. (28).
    This is an unproved and generally false assumption; for margin losses, L^{γ/2} ≤ L^γ gives the opposite inequality direction, so Eq. (28) is not an upper bound.
  • ad hoc to paper The denominator in the Gaussian conditional probability expression is bounded, and a Taylor-type expansion yields the bound in Eq. (22).
    The denominator can be arbitrarily small or large; no rigorous bound is provided, and the expansion to a 1/σ^2 term is not justified.
  • ad hoc to paper The aggregated feature differences ||f_u - f_v|| can be replaced by ||(A_S X_S)_i - (A_T X_T)_j||, and the mean parameters μ_i satisfy Δμ_i ≤ ||X^S - X^T||.
    Final step of Proposition 2's proof in Appendix A; no derivation connects GNN outputs to A X and X, and the μ terms are simply asserted to vanish into the X divergence.
  • standard math Prior P over the classifier family H is independent of the training data.
    Standard PAC-Bayes requirement, cited from Ma et al. 2021 and used in Theorem 1.

how reviews work

0 comments
Cite this review

Pith. "Pith review of On the Benefits of Attribute-Driven Graph Domain Adaptation." pith.science (2026). https://pith.science/paper/U3UOZHKT

@misc{pith2026250206808,
  author       = {Pith},
  title        = {Pith review of: On the Benefits of Attribute-Driven Graph Domain Adaptation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/U3UOZHKT}},
  note         = {Machine review of arXiv:2502.06808}
}
read the original abstract

Graph Domain Adaptation (GDA) addresses a pressing challenge in cross-network learning, particularly pertinent due to the absence of labeled data in real-world graph datasets. Recent studies attempted to learn domain invariant representations by eliminating structural shifts between graphs. In this work, we show that existing methodologies have overlooked the significance of the graph node attribute, a pivotal factor for graph domain alignment. Specifically, we first reveal the impact of node attributes for GDA by theoretically proving that in addition to the graph structural divergence between the domains, the node attribute discrepancy also plays a critical role in GDA. Moreover, we also empirically show that the attribute shift is more substantial than the topology shift, which further underscores the importance of node attribute alignment in GDA. Inspired by this finding, a novel cross-channel module is developed to fuse and align both views between the source and target graphs for GDA. Experimental results on a variety of benchmarks verify the effectiveness of our method.

Figures

Figures reproduced from arXiv: 2502.06808 by the authors.

Figure 1
Figure 1. This represents feature value in two groups of datasets. This shows the feature value distribution gap in the attribute is larger than in the topology. Early works on GDA apply deep domain adaptation (DA) techniques directly, thereby (Shen et al., 2020b; Shui et al., 2023; Wu et al., 2020; Shen et al., 2020a; Dai et al., 2022) without considering the topologi￾cal structures of graphs for domain alignment. To address… view at source ↗
Figure 2
Figure 2. (a) An overview of our method. GAA gives attribute and topology graph representation, [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 4
Figure 4. The influence of parameters α, β, τ and k on Citation and Airport dataset. performance by enhancing and integrating feature information, benefiting the synergistic refinement of both attribute and topology. 5.6 PARAMETER ANALYSIS 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 A→D D→A A→C C→A C→D D→C FAA 𝑮𝑮𝑮𝑮𝑮𝑮 FAA1 𝑮𝑮𝑮𝑮𝑮𝑮 FAA2 FAA3 𝟏𝟏 𝑮𝑮𝑮𝑮𝑮𝑮𝟐𝟐 𝑮𝑮𝑮𝑮𝑮𝑮𝟑𝟑 (a) Citation 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75 U→B U→E B→U B→E E… view at source ↗
Figures from the paper (4 more)
Figure 3
Figure 3. Figure 3: The classification accuracy of GAA and its variants on citation datasets and airport dataset. In this section, we analyze the sensitivity of the parameters of our method on the Airport dataset and Citation dataset. As shown in Figure.5 in Subfigure (4), the accuracy us…
Figure 5
Figure 5. Figure 5: The influence of parameters α, β, τ and k on two social datasets. Types Datasets α β τ k Airport U→B 0.5 0.5 0.01 4 U→E 0.1 0.1 0.01 2 B→U 0.1 0.1 0.01 4 B→E 0.5 0.1 0.1 3 E→U 0.5 0.5 0.1 4 E→B 0.5 0.5 0.1 4 Citation A→D 0.1 0.1 0.1 3 D→A 0.1 0.1 0.01 4 A→C 0.5 0.5 0.0…
Figure 6
Figure 6. Figure 6: Visualization of learnt representations of different methods on D-A task of dataset. [PITH_FULL_IMAGE:figures/full_fig_p018_6.png]
Figure 7
Figure 7. Figure 7: Visualization of bound value and LA value. G MODEL EFFICIENT EXPERIMENT To further investigate the efficiency of GAAo, Table?? reports the running time comparison across various algorithms. We also compared the training time and GPU memory usage of common baselines UDA…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Cross-Resolution Semantic Learning for Graph Domain Adaptation

    cs.LG 2026-07 conditional novelty 6.0 of 10

    CReSL improves graph domain adaptation by learning cross-resolution source-to-target routing and grafting target representations toward source class prototypes.

Reference graph

Works this paper leans on

19 extracted references · 12 canonical work pages · cited by 1 Pith paper

  1. [2]

    (16) Proof

    For nodes s ∈ VS and t ∈ VT with aggregated features f = GNN(x), the following inequality holds: ||P(yu = c0|fu) − P(yv = c0|fv)|| ≤O(||fu − fv|| + ||fv − µ(S) 1 || + ||fv − µ(T ) 1 ||). (16) Proof. The conditional probability of class c0 given the aggregated feature f can be expressed using Bayes’ theorem: P(ys = c0|fs) = P(fs|ys = c0)P(ys = c0) P(fs|ys ...

    work page 2025

  2. [5]

    Rethinking propagation for unsupervised graph domain adaptation

    Meihan Liu, Zeyu Fang, Zhen Zhang, Ming Gu, Sheng Zhou, Xin Wang, and Jiajun Bu. Rethinking propagation for unsupervised graph domain adaptation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pp. 13963–13971, 2024a. Shikun Liu, Deyu Zou, Han Zhao, and Pan Li. Pairwise alignment improves graph domain adaptation. arXiv preprin...

    arXiv

  3. [6]

    Revisiting link prediction: A data perspective

    Haitao Mao, Juanhui Li, Harry Shomer, Bingheng Li, Wenqi Fan, Yao Ma, Tong Zhao, Neil Shah, and Jiliang Tang. Revisiting link prediction: A data perspective. arXiv preprint arXiv:2310.00793,

    arXiv

  4. [9]

    Adversarial deep network embedding for cross-network node classification

    Xiao Shen, Quanyu Dai, Fu-lai Chung, Wei Lu, and Kup-Sze Choi. Adversarial deep network embedding for cross-network node classification. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pp. 2991–2999, 2020a. Xiao Shen, Quanyu Dai, Sitong Mao, Fu-lai Chung, and Kup-Sze Choi. Network together: Node classification via cross-networ...

    work page 1935

  5. [11]

    Unsupervised domain adaptive graph convolutional networks

    Man Wu, Shirui Pan, Chuan Zhou, Xiaojun Chang, and Xingquan Zhu. Unsupervised domain adaptive graph convolutional networks. In Proceedings of The Web Conference 2020, pp. 1457–1467,

    work page 2020

  6. [13]

    Graph-relational domain adaptation

    12 Published as a conference paper at ICLR 2025 Zihao Xu, Hao He, Guang-He Lee, Yuyang Wang, and Hao Wang. Graph-relational domain adaptation. arXiv preprint arXiv:2202.03628,

    arXiv 2025

  7. [14]

    Graphae: adaptive embedding across graphs

    Bencheng Yan and Chaokun Wang. Graphae: adaptive embedding across graphs. In 2020 IEEE 36th International Conference on Data Engineering (ICDE), pp. 1958–1961. IEEE,

    work page 2020

  8. [15]

    Dane: Domain adaptive network embedding

    Yizhou Zhang, Guojie Song, Lun Du, Shuwen Yang, and Yilun Jin. Dane: Domain adaptive network embedding. arXiv preprint arXiv:1906.00684,

    arXiv 1906

Show all 19 references
  1. [17]

    Given the sparsity of this dataset, α, β and τ should be set relatively higher to emphasize topology alignment and capture key structural relationships

    Airport Dataset: Often contains transportation networks with fewer nodes but complex edge rela- tionships. Given the sparsity of this dataset, α, β and τ should be set relatively higher to emphasize topology alignment and capture key structural relationships. α, β and τ is sel...

    2025

  2. [19]

    As illustrated in Figure 6(b), both the bound value and the loss value of the model increase as the attribute discrepancy grows. Conversely, the classification performance declines with increasing topology discrepancy, emphasizing that the bound’s topology component is also cl...

    2025

  3. [100]

    Conversely, the classification performance declines with increasing attribute discrepancy, highlighting that the bound attribute component is closely related to the GDA performance

    As illustrated in Figure 6(a), both the bound value and the loss value of the model increase as the attribute discrepancy grows. Conversely, the classification performance declines with increasing attribute discrepancy, highlighting that the bound attribute component is closel...

    2025

  4. [2015]

    Semi-supervised classification with graph convolutional networks

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

    arXiv

  5. [2017]

    Network embedding for cross-network node classification

    Xiao Shen and Fu Lai Chung. Network embedding for cross-network node classification. arXiv preprint arXiv:1901.07264,

    1901 arXiv

  6. [2019]

    Graph transfer learning via adversarial domain adaptation with graph convolution

    10 Published as a conference paper at ICLR 2025 Quanyu Dai, Xiao-Ming Wu, Jiaren Xiao, Xiao Shen, and Dan Wang. Graph transfer learning via adversarial domain adaptation with graph convolution. IEEE Transactions on Knowledge and Data Engineering, 35(5):4908–4922,

    2025

  7. [2020]

    One node one model: Featuring the missing-half for graph clustering

    Xuanting Xie, Bingheng Li, Erlin Pan, Zhaochen Guo, Zhao Kang, and Wenyu Chen. One node one model: Featuring the missing-half for graph clustering. arXiv preprint arXiv:2412.09902, 2024a. Xuanting Xie, Erlin Pan, Zhao Kang, Wenyu Chen, and Bingheng Li. Provable filter for real...

    arXiv

  8. [2021]

    Cdc: A simple framework for complex data clustering

    Zhao Kang, Xuanting Xie, Bingheng Li, and Erlin Pan. Cdc: A simple framework for complex data clustering. arXiv preprint arXiv:2403.03670,

    arXiv

  9. [2022]

    Structure-preserving graph represen- tation learning

    Ruiyi Fang, Liangjian Wen, Zhao Kang, and Jianzhuang Liu. Structure-preserving graph represen- tation learning. In 2022 IEEE International Conference on Data Mining (ICDM), pp. 927–932. IEEE,

    2022

  10. [2023]

    Graph domain adaptation: Challenges, progress and prospects

    Boshen Shi, Yongqing Wang, Fangda Guo, Bingbing Xu, Huawei Shen, and Xueqi Cheng. Graph domain adaptation: Challenges, progress and prospects. arXiv preprint arXiv:2402.00904,

    arXiv

  11. [2025]

    Upper bounding barlow twins: A novel filter for multi-relational clustering

    11 Published as a conference paper at ICLR 2025 Xiaowei Qian, Bingheng Li, and Zhao Kang. Upper bounding barlow twins: A novel filter for multi-relational clustering. In Proceedings of the AAAI Conference on Artificial Intelligence , volume 38, pp. 14660–14668,

    2025

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.