Homophily Enhanced Graph Domain Adaptation

Bingheng Li; Boyu Wang; Charles Ling; Gezheng Xu; Jingyu Zhao; Qiuhao Zeng; Ruiyi Fang; Ruizhi Pu

REVIEW 4 major objections 4 minor 1 cited by

Homophily Enhanced Graph Domain Adaptation

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper argues that mismatched homophily between source and target graphs is a hidden driver of graph domain adaptation failure, and that aligning homophilic, heterophilic, and attribute signals separately restores performance.

desk verdict The empirical motivation is fresh and the three-view alignment is a reasonable design, but Theorem 1's proof has a reversed Wasserstein inequality that collapses the theoretical claim, and the appendix tables contradict the main results. read the letter →

arxiv 2505.20089 v3 pith:UHLR74PM submitted 2025-05-26 cs.SI cs.AI

Ruiyi Fang , Bingheng Li , Jingyu Zhao , Ruizhi Pu , Qiuhao Zeng , Gezheng Xu , Charles Ling , Boyu Wang This is my paper

classification cs.SIcs.AI

keywords graphdomainadaptationhomophilyheterophilyneuralnetworkscross-networknodeclassificationdistributionshiftsignalfiltering

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper sets out to show that graph homophily, the tendency of connected nodes to share labels, is a decisive and overlooked factor in graph domain adaptation (GDA), where a classifier is learned on a labeled source graph and applied to an unlabeled target graph. It documents that source and target graphs in standard benchmarks differ more sharply within homophilic and heterophilic subgroups than in their aggregate homophily, and that target accuracy falls as subgroup homophily divergence rises. It then proves a probabilistic generalization bound stating that target loss is controlled by four separate shifts: homophilic signal, attribute signal, heterophilic signal, and the graph's heterophily distribution. Accordingly it proposes HGDA, which applies homophilic, full-pass, and heterophilic filters to each graph and aligns the corresponding channels with KL-divergence losses. If the claim holds, GDA methods should stop aligning a single blended representation and instead match these three graph-signal channels separately, and HGDA should keep its reported accuracy gains across Airport, ACM, Citation, Blog, Twitch, and MAG.

What carries the argument

The central object is the four-term KL decomposition of feature-distribution shift: $D_{\mathrm{KL}}(P_F^S \parallel P_F^T) \leq D_{\mathrm{KL}}(A_S X_S \parallel A_T X_T) + D_{\mathrm{KL}}(X_S \parallel X_T) + D_{\mathrm{KL}}(L_S X_S \parallel L_T X_T)$, together with the graph-level heterophily distribution shift $D_{\mathrm{KL}}(P_H^S \parallel P_H^T)$. Carried by Theorem 1, this bound says a deterministic classifier's target loss is controlled by source margin loss, a complexity term, and the sum of the square roots of those KL terms. HGDA operationalizes the decomposition with a homophilic filter (the normalized adjacency $\tilde{A}$), a full-pass filter (the identity $I$), and a heterophilic filter (the normalized Laplacian $\tilde{L}$), aligning the resulting embeddings channel-wise with three KL losses.

What would settle it

Pick a benchmark pair such as U to B, estimate each KL term empirically, and check whether $D_{\mathrm{KL}}(P_F^S \parallel P_F^T) \leq D_{\mathrm{KL}}(A_S X_S \parallel A_T X_T) + D_{\mathrm{KL}}(X_S \parallel X_T) + D_{\mathrm{KL}}(L_S X_S \parallel L_T X_T)$ holds, and separately whether the asserted bound on $D_{\mathrm{KL}}(P_H^S \parallel P_H^T)$ holds; failure of either inequality would show the stated proof chain does not go through.

Watch

Extended reading notes

Core claim

The paper's central claim is that GDA performance is governed by four independent shift terms: homophilic signal, attribute signal, heterophilic signal, and graph-level node heterophily distribution. Theorem 1 upper-bounds target classification loss by source margin loss plus a penalty proportional to the square roots of the KL divergences of these four terms. To exploit the bound, HGDA applies three filters, the normalized adjacency, the identity, and the normalized Laplacian, to the source and target graphs and adds pairwise KL alignment losses between the corresponding channels. Empirically, homophily subgroup ratios negatively correlate with target accuracy, and HGDA achieves the best average accuracy across Airport, ACM, Citation, Blog, Twitch, and MAG benchmarks.

Load-bearing premise

The proof assumes, without stating it in the main text, that adjacency-filtered and Laplacian-filtered features are conditionally independent given the raw features, an assumption that fails when the same features and graph structure feed both filters; it also asserts the heterophily KL bound in Eq. (52) without derivation.

Editorial extensions

If this is right

Methods that align only a single blended representation will keep failing in exactly the node subgroups whose homophily ratios diverge most between source and target graphs.
The four-term bound predicts that reducing any one of the four shifts, homophilic, attribute, heterophilic, or heterophily distribution, reduces the upper bound on target loss, so channel-wise alignment is the natural objective.
Because the heterophily-distribution term is a fixed graph statistic, two graphs can look globally similar yet still require adaptation; benchmark pairs such as ACM3 to ACM4 are cases in point.
The consistent gains of the three-filter combination over each single-filter variant imply that homophilic and heterophilic information are complementary and both need explicit alignment.

Reading between the lines

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

The subgroup homophily divergence could be measured before training and used as a cheap diagnostic of whether standard GDA is likely to fail on a given source-target pair.
The conditional-independence assumption behind the KL decomposition is testable; if it fails, a corrected bound would need an extra cross-term between adjacency- and Laplacian-filtered features, which would not necessarily invalidate the method's empirical gains.
Because the three filters decompose the graph signal into low-, full-, and high-frequency channels, the same channel-wise alignment idea could be carried over to arbitrary spectral filter banks, including learnable or dataset-adaptive filters.
Matching the graph-level heterophily distribution directly, for example by reweighting target nodes or constraining label entropy conditional on degree, could complement the filter alignment and further tighten the bound.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

I read this with the stress-test note in hand. My verdict: the central theoretical claim does not hold as written, and the experimental reporting has internal inconsistencies, but the empirical observation about local homophily distribution shift is new and plausible.

What's new: nobody in the cited GDA literature studies local homophily distribution shift or proposes aligning homophilic, heterophilic, and attribute signals separately. The empirical analysis in Figures 1 and 2 is a reasonable motivation, and the three-filter architecture is a simple, sensible design. The gains on MAG (8.5 points over PA) are large and consistent across the ten transfers, which suggests the method is doing something real—though with the caveat below.

The proof of Theorem 1 is not sound. Proposition 1 (Eq. 26) assumes the double sum over node pairs approximates N_S N_T W1. That is wrong: W1 is an infimum over couplings, and the uniform product coupling is just one feasible coupling, so the average pairwise distance is always at least W1. The very next display runs the inequality in the opposite direction. So the bound on D^{γ/2}_{S,T} does not follow. Theorem 1 inherits this error. The additional step bounding D_KL(P^H_S || P^H_T) by the same three signal KL terms (Eq. 52) is asserted without derivation. Together these mean the paper's headline theoretical claim—homophily discrepancies degrade GDA performance—is not established by this proof. This is an internal math problem, not a taste disagreement, and it cannot be fixed by changing a constant.

Two smaller issues: the method section says the overall loss is L = L_H + α L_S + β L_T, but L_D is referenced and never defined. And Appendix D Table 4 gives HGDA accuracy 0.326 for A→C (and C→D) while Table 2 reports 0.829 and 0.779. No standard deviations are reported anywhere despite the appendix saying five runs.

The empirical story might survive a corrected proof—the correlation plot in Figure 2 is the kind of evidence that could stand alone—but as submitted, the load-bearing theoretical claim is unsupported and the tables are inconsistent. The method specification gap is minor by comparison, but real.

Who this is for: someone working on GDA who cares about heterophily-aware alignment. It deserves a serious referee because the problem is well-motivated, the idea is new within the subfield, and the empirical results are promising enough to warrant a careful revision.

Recommendation: send to peer review, but expect major revision—the proof needs to be corrected or removed/re-scoped as empirical motivation, L_D needs to be defined, and the appendix tables need to be reconciled. I would not desk-reject, but I also would not trust the current Theorem 1 or the efficiency table.

Referee Report

4 major / 4 minor

Summary. The paper addresses unsupervised graph domain adaptation (GDA) under node homophily discrepancy. It presents empirical evidence that subgroups with differing homophily ratios show a negative correlation between homophily distribution divergence and target classification accuracy. The authors propose HGDA, which applies homophilic (A-tilde), full-pass (I), and heterophilic (L-tilde) filters to source and target node features, aligns the resulting embeddings via three KL divergence terms, and trains with a source classifier loss plus target entropy loss. The central theoretical claim is Theorem 1, a PAC-Bayes upper bound on target margin loss in terms of homophilic signal shift, attribute shift, heterophilic signal shift, and graph-level node heterophily distribution shift. Experiments compare HGDA and three ablations against GDA baselines on Airport, ACM, Citation, Blog, Twitch, and MAG.

Significance. If Theorem 1 and the empirical results were sound, the paper would provide a useful conceptual contribution: explicit separation and alignment of homophilic, attribute, and heterophilic signals for GDA, with a simple and computationally light architecture. The paper also ships extensive benchmark comparisons and an efficiency analysis. However, the proof of the core bound contains a reversed inequality involving the 1-Wasserstein distance, and an appendix table contradicts the main accuracy tables; the theoretical justification therefore fails as written, and the empirical claims are not internally consistent.

major comments (4)

[Appendix A, Proposition 1, Eqs. (26)-(27)] The proof of Proposition 1 uses the uniform product coupling to relate the pairwise feature distance sum to W_1(P^F_S, P^F_T). Since W_1 is the infimum over all couplings and the product coupling is one feasible coupling, the average pairwise distance is always at least W_1. The text's assertion that the sum can be approximated by N_S N_T W_1, followed by Eq. (27) '<=' N_S W_1, has the inequality in the wrong direction. Consequently the chain (31)-(39) does not establish the claimed bound on D^{gamma/2}_{S,T}(P;lambda), and Theorem 1, which relies on this proposition, is unproven. This is a load-bearing error: the paper's main theoretical conclusion that homophily discrepancies degrade GDA performance is not supported.
[Appendix A, Corollary 4, Eqs. (40)-(44)] The proof assumes conditional independence between adjacency-transformed and Laplacian-transformed features. Both are deterministic functions of the same node features X and graph structure, so conditional independence does not hold. Moreover, the chain-rule step Eq. (43) conditions on X_S in the first factor and X_T in the second, so the decomposition cannot yield the unconditional KL sum in Eq. (41). Since Corollary 4 is the bridge from feature-distribution KL to the three signal KL terms in Theorem 1, this step also fails.
[Appendix A, Eq. (52)] In the proof of Theorem 5, the bound DKL(P_H_S||P_H_T) <= DKL(A_S X_S||A_T X_T) + DKL(X_S||X_T) + DKL(L_S X_S||L_T X_T) is asserted without derivation and is not a consequence of the previous arguments: P_H is the distribution of label-derived node heterophily values, not a feature-signal distribution. Additionally, the final display Eq. (54) omits the heterophily KL term, whereas the theorem statements (Theorems 1 and 5) include it; the proof is internally inconsistent about which terms the bound contains.
[Appendix D, Table 4] The efficiency experiment reports HGDA A->C accuracy 0.326, while Table 2 reports HGDA A->C accuracy 0.829; other entries disagree similarly (e.g., PA A->D accuracy 0.562 in Table 4 versus 0.752 in Table 2). The table therefore cannot be read as a reliable report of the same experiments. Furthermore, no standard deviations are given anywhere, and the per-task hyperparameter choices in Table 5 are not accompanied by a validation protocol, leaving the possibility of selection on the test set unaddressed. These issues undermine confidence in the empirical claims.

minor comments (4)

[Section 4.2] The correspondence between the three KL alignment terms and the theoretical terms is mislabeled: Z_H is introduced in Section 4.1 as the output of the heterophilic filter and Z_F as the full-pass filter, but Section 4.2 says KL(Z_S_H||Z_T_H) aligns 'graph attributes' (DKL(X_S||X_T)) and KL(Z_S_F||Z_T_F) aligns 'heterophilic signal' (DKL(L_S X_S||L_T X_T)); these labels are swapped relative to the definitions.
[Section 4.3, Eq. (14)] The overall loss is given as L = L_H + alpha L_S + beta L_T, but the surrounding text mentions combining L_H, L_S, 'LD' and L_T; L_D is never defined.
[General typos] Section 4 heading 'Methology' should be 'Methodology'; 'According to Theorem 7' in Section 4.2 should refer to Theorem 1; Algorithm 1 line 1 has an incomplete sentence ('by running .'); and Figure 5 caption lists 'HGDAL' twice, presumably with the fourth variant intended as HGDAH.
[Appendix D, Table 4] The table is titled 'Airport datset' but contains Citation rows (A->D, A->C, C->D), indicating that the efficiency experiments were not cleanly separated by dataset in the presentation.

Circularity Check

3 steps flagged · score 5.0 of 10

Partial circularity: Theorem 1's heterophily bound is assumed via an unproved 'similarly' step, and the core PAC-Bayes bound is imported from prior work including the authors' own, while the benchmark experiments remain independent.

self citation load bearing [Appendix A, Proposition 1 and its proof, first inequality.]
"Proposition 1 (Bound for DγS,T(P;λ), Adaptation from (Ma et al., 2021; Mao et al., 2024a; Fang et al., 2025)). ... From the cited works (Ma et al., 2021; Mao et al., 2024a; Fang et al., 2025), it follows that the expected loss discrepancy measure D_{S,T}^{γ/2}(P;λ) can be bounded by: [1/max(N_S,N_T)] Σ_i Σ_j (ln 3 + λCρ/(√(2π)σ)(∥f_i−f_j∥ + |h_i−h_j|·ρ))."

The decisive step that turns the PAC-Bayes domain discrepancy into the pairwise feature-plus-heterophily expression is not proved in this paper; it is explicitly 'Adaptation from' prior work, including the authors' own Fang et al. (2025). The new heterophily term |h_i−h_j|·ρ is inserted into the imported bound as though the same citation justified it. Because Theorem 1 inherits the whole chain from Proposition 1, the paper's theoretical justification of homophily alignment rests on a self-citation whose proof is not exhibited here. The presence of non-self citations (Ma et al. 2021, Mao et al. 2024a) gives some independent support, so this is load-bearing but not exclusively self-citational.
other [Appendix A, proof of Theorem 5, Eq. (52).]
"Similarly, for the heterophily-based KL divergence: D_{KL}(P^H_S ∥ P^H_T) ≤ D_{KL}(A_S X_S ∥ A_T X_T) + D_{KL}(X_S ∥ X_T) + D_{KL}(L_S X_S ∥ L_T X_T)."

Corollary 4 derived an analogous decomposition for the feature distribution P^F under a conditional-independence assumption, but no derivation is given for Eq. (52). The heterophily distribution P^H is defined from labels in Definition 1 and is not a function of the three filtered signals A_S X_S, X_S, and L_S X_S. Asserting 'similarly' that its KL is bounded by the same three signal KLs assumes exactly what the theorem is meant to establish: that homophily/heterophily discrepancy is controlled by the three aligned channels. The method's loss in Eq. (11) then minimizes those three DKL terms, so the bound and the algorithm mirror each other by construction rather than by independent derivation.

1 more flagged steps

self definitional [Appendix A, Proposition 1 proof, Eqs. (26)-(27).]
"Since the sum over all node pairs approximates an expectation over a joint coupling of the distributions, we approximate: Σ_{i∈V_S} Σ_{j∈V_T} ∥f_i − f_j∥ ≈ N_S N_T W_1(P^F_S, P^F_T). ... Substituting this into our sum, we obtain: Σ_i Σ_j (1/N_T) ∥f_i − f_j∥ ≤ N_S W_1(P^F_S, P^F_T)."

By Corollary 2, W_1 is the infimum over couplings, so the double sum under the uniform product coupling is one feasible coupling and is always at least W_1, never at most. Eq. (27) therefore reverses the definitional inequality. This is a definitional error rather than a classic circular definition, but it is a specific reduction that fails by construction: the claimed bound does not follow from the definition of W_1 and invalidates the subsequent chain of inequalities leading to Theorem 1.

full rationale

The empirical contribution is not circular: HGDA is evaluated against external GDA baselines on six benchmarks, and the reported accuracy gains are independent of Theorem 1 as a fitted input. The circularity is confined to the theoretical derivation chain. Proposition 1, which supplies the central reduction from PAC-Bayes discrepancy to pairwise feature/heterophily terms, is explicitly adapted from prior work including the authors' own Fang et al. (2025), and the added heterophily term is not re-derived. Eq. (52) then asserts, without proof, that the label-defined heterophily distribution KL is bounded by the three filtered-signal KLs that the algorithm's loss minimizes, so the theorem's new term is assumed rather than derived. Finally, the proof replaces the Wasserstein-1 infimum by the uniform product-coupling sum, reversing the definitional inequality. These issues make Theorem 1 a post-hoc justification of the three-channel design rather than a first-principles bound, but they do not taint the benchmark experiments.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The theoretical claim rests on a decomposition (Corollary 4) that assumes conditional independence, plus an unproved inequality for the heterophily distribution term (Eq. 52). The empirical claim rests on per-task hyperparameter choices that are not validated with target labels. No new physical or mathematical entities are introduced.

free parameters (2)

alpha_l, alpha_f, alpha_h (filter scalars) = Learned during training; no final values reported
Equations 8 to 10 introduce learnable scalar parameters for each filter; they are trained as part of the model, not fixed constants.
alpha and beta trade-off hyperparameters = Per-task values listed in Appendix Table 5, ranging from 0.01 to 0.5
Chosen per task from small grids without a validation set; in an unsupervised target setting this is effectively parameter fitting if the selection uses target labels.

assumptions (4)

standard math Bobkov-Gotze inequality bounding Wasserstein-1 distance by KL divergence (Lemma 1)
Used in Appendix A to convert Wasserstein bounds into sqrt-KL terms; this is an accepted standard result.
domain assumption Local node homophily ratio is the relevant distributional summary for GDA transfer
The paper assumes shifts in local homophily ratios are a key driver of performance loss; this is a modeling choice rather than a proven fact.
ad hoc to paper Conditional independence of adjacency-transformed and Laplacian-transformed features in Corollary 4
Needed for the KL decomposition, but not generally true because both transforms share the same node features and graph structure.
ad hoc to paper DKL(PH_S || PH_T) is bounded by DKL(ASXS || ATXT) + DKL(XS || XT) + DKL(LSXS || LTXT), Eq. (52)
Asserted without derivation and used to keep the heterophily distribution term in the theorem; no supporting argument is given.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Homophily Enhanced Graph Domain Adaptation." pith.science (2026). https://pith.science/paper/UHLR74PM

@misc{pith2026250520089,
  author       = {Pith},
  title        = {Pith review of: Homophily Enhanced Graph Domain Adaptation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UHLR74PM}},
  note         = {Machine review of arXiv:2505.20089}
}

read the original abstract

Graph Domain Adaptation (GDA) transfers knowledge from labeled source graphs to unlabeled target graphs, addressing the challenge of label scarcity. In this paper, we highlight the significance of graph homophily, a pivotal factor for graph domain alignment, which, however, has long been overlooked in existing approaches. Specifically, our analysis first reveals that homophily discrepancies exist in benchmarks. Moreover, we also show that homophily discrepancies degrade GDA performance from both empirical and theoretical aspects, which further underscores the importance of homophily alignment in GDA. Inspired by this finding, we propose a novel homophily alignment algorithm that employs mixed filters to smooth graph signals, thereby effectively capturing and mitigating homophily discrepancies between graphs. Experimental results on a variety of benchmarks verify the effectiveness of our method.

Figures

Figures reproduced from arXiv: 2505.20089 by the authors.

**Figure 1.** This represents node homophily distibution in two benchmark. The dotted line represents the overall node homophily ratio of the entire graph. This shows the local node homophily distibution shift is existing in various levels of homophilic groups. Homophily means that similar nodes are prone to connect to each other. In this work, we highlight the role of graph homophily discrepancies and their impact on GDA perf… view at source ↗

**Figure 2.** Performance of cross-network classification tasks in different homophilic ratio ranges. For a fair comparison, we use a two-layer GCN with standard unsupervised GDA settings as our evaluation model. The X-axis represents node subgroups categorized by specific ranges of homophily ratios. The bars in the vertical bar chart represent the proportion of nodes in each subgroup, defined by a specific homophily ratio range … view at source ↗

**Figure 3.** An overview of HGDA. HGDA optimizes heterophily, attribute, and homophily graph signals while minimizing homophily distribution shifts between the source and target graphs. graph. In practice, we can treat A as homophilc filter and L as heterophilic filter (Nt & Maehara, 2019). Thus, as shown in [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (6 more)

**Figure 4.** Figure 4: Visualizing source and target node embeddings via T-SNE. Each color represents a class, while dark and light shades of the same color correspond to nodes of the same class from the source and target domains, respectively. (a) U → E (b) U → B (c) B → E (d) A4 → A3 [PIT…

**Figure 5.** Figure 5: Classification accuracy of HGDA, HGDAL, HGDAF , and HGDAL across different subgroups in Airport and ACM datasets. 0.01 0.05 0.1 0.5 1 0.5 0.56 0.62 0.68 0.74 0.8 Accuracy(%) 𝜶𝜶 𝜷𝜷 (a) A → D 0.01 0.05 0.1 0.5 1 0.3 0.4 0.5 0.6 0.7 0.8 Accuracy(%) 𝜶𝜶 𝜷𝜷 (b) U → B 0.01 0.…

**Figure 6.** Figure 6: The influence of parameters α and β on Citation, Airport, ACM and MAG datasets. 5.4. Ablation Study Among the four variants of HGDA, HGDA performs the best in most cases. HGDAL, HGDAF , and HGDAH achieve varying performance levels across different groups of datasets, i…

**Figure 7.** Figure 7: This shows the local node homophily distibution shift is existing in MAG, Citation, Blog and Twitch datasets. C. Empirical Study: Homophily Divergence Effect to Target classification Performance In this appendix, we extend our analysis by presenting the homophily distr…

**Figure 8.** Figure 8: GCN performance on cross-network classification tasks across different homophily ratio ranges for the MAG, Twitch, and Blog datasets. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]

**Figure 9.** Figure 9: The influence of hyper-parameters α and β on Citation, Airport, ACM and MAG datasets. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_9.png]

Discussion (0). Sign in 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

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

26 extracted references · 18 canonical work pages · cited by 1 Pith paper

[1]

+ ln 1 δ + λ2 4NS +D γ/2 S,T (P;λ) . Proposition 1(Bound for Dγ S,T (P;λ) , Adaptation from (Ma et al., 2021; Mao et al., 2024a; Fang et al., 2025)).For any γ≥0 , and under the assumption that the prior distribution P over the classification function family Φ is defined, we establish a bound for the domain discrepancy measureD γ/2 S,T (P;λ). Specifically,...

work page 2021
[5]

Mamba-Based Graph Convolutional Networks: Tackling Over-smoothing with Selective State Space

He, X., Wang, Y ., Fan, W., Shen, X., Juan, X., Miao, R., and Wang, X. Mamba-based graph convolutional networks: Tackling over-smoothing with selective state space.arXiv preprint arXiv:2501.15461,

work page Pith review arXiv
[6]

Robust mid-pass filtering graph convolutional networks

Huang, J., Du, L., Chen, X., Fu, Q., Han, S., and Zhang, D. Robust mid-pass filtering graph convolutional networks. InProceedings of the ACM Web Conference 2023, pp. 328–338,

work page 2023
[10]

Rethinking propagation for unsupervised graph domain adaptation

Liu, M., Fang, Z., Zhang, Z., Gu, M., Zhou, S., Wang, X., and Bu, J. Rethinking propagation for unsupervised graph domain adaptation. InProceedings of the AAAI Confer- ence on Artificial Intelligence, volume 38, pp. 13963– 13971, 2024a. Liu, M., Zhang, Z., Tang, J., Bu, J., He, B., and Zhou, S. Re- visiting, benchmarking and understanding unsupervised gra...

arXiv 2024
[11]

Demystifying structural disparity in graph neural networks: Can one size fit all?Advances in neural information processing systems, 36, 2024a

Mao, H., Chen, Z., Jin, W., Han, H., Ma, Y ., Zhao, T., Shah, N., and Tang, J. Demystifying structural disparity in graph neural networks: Can one size fit all?Advances in neural information processing systems, 36, 2024a. Mao, H., Du, L., Zheng, Y ., Fu, Q., Li, Z., Chen, X., Han, S., and Zhang, D. Source free unsupervised graph domain adaptation.The 17th...

work page 2024
[12]

URL https:// openreview.net/forum?id=zrQIc9mQQN. Nt, H. and Maehara, T. Revisiting graph neural net- works: All we have is low-pass filters.arXiv preprint arXiv:1905.09550,

arXiv 1905
[14]

Adversarial deep network embedding for cross-network node classification

Shen, X., Dai, Q., Chung, F.-l., Lu, W., and Choi, K.-S. Adversarial deep network embedding for cross-network node classification. InProceedings of the AAAI confer- ence on artificial intelligence, volume 34, pp. 2991–2999, 2020a. Shen, X., Dai, Q., Mao, S., Chung, F.-l., and Choi, K.-S. Network together: Node classification via cross-network deep network...

work page 1935
[16]

Unsu- pervised domain adaptive graph convolutional networks

Wu, M., Pan, S., Zhou, C., Chang, X., and Zhu, X. Unsu- pervised domain adaptive graph convolutional networks. InProceedings of The Web Conference 2020, pp. 1457– 1467,

work page 2020

Show all 26 references

[18]

Intersectional unfairness discovery.arXiv preprint arXiv:2405.20790, 2024a

Xu, G., Chen, Q., Ling, C., Wang, B., and Shui, C. Intersectional unfairness discovery.arXiv preprint arXiv:2405.20790, 2024a. Xu, G., Guo, H., Yi, L., Ling, C., Wang, B., and Yi, G. Re- visiting source-free domain adaptation: a new perspective via uncertainty control. InThe T...

arXiv
[19]

Graph-relational domain adaptation.arXiv preprint arXiv:2202.03628,

Xu, Z., He, H., Lee, G.-H., Wang, Y ., and Wang, H. Graph-relational domain adaptation.arXiv preprint arXiv:2202.03628,

arXiv
[20]

and Wang, C

Yan, B. and Wang, C. Graphae: adaptive embedding across graphs. In2020 IEEE 36th International Conference on Data Engineering (ICDE), pp. 1958–1961. IEEE,

1958
[21]

Diverse message passing for attribute with het- erophily

Yang, L., Li, M., Liu, L., Niu, B., Wang, C., Cao, X., and Guo, Y . Diverse message passing for attribute with het- erophily. InAdvances in Neural Information Processing Systems 34, NeurIPS 2021,, pp. 4751–4763,

2021
[22]

Cluster-guided contrastive graph clustering network

Yang, X., Liu, Y ., Zhou, S., Wang, S., Tu, W., Zheng, Q., Liu, X., Fang, L., and Zhu, E. Cluster-guided contrastive graph clustering network. InProceedings of the AAAI con- ference on artificial intelligence, volume 37, pp. 10834– 10842, 2023a. Yang, X., Tan, C., Liu, Y ., Li...

2024
[23]

Dane: Domain adaptive network embedding.arXiv preprint arXiv:1906.00684,

Zhang, Y ., Song, G., Du, L., Yang, S., and Jin, Y . Dane: Domain adaptive network embedding.arXiv preprint arXiv:1906.00684,

1906 arXiv
[24]

Pygda: A python li- brary for graph domain adaptation.arXiv preprint arXiv:2503.10284,

Zhang, Z., Liu, M., and He, B. Pygda: A python li- brary for graph domain adaptation.arXiv preprint arXiv:2503.10284,

arXiv
[1999]

Let P and Q be two probability distributions on a metric space (X, d), where P≪Q (i.e., P is absolutely continuous with respect to Q). Assume that: •Xhas bounded support with diameterD, i.e.,d(x, y)≤Dfor allx, y∈ X, •There exists a constantC 0 >0such that all 1-Lipschitz funct...

2021
[2015]

9 Homophily Enhanced Graph Domain Adaptation Kipf, T. N. and Welling, M. Semi-supervised classifica- tion with graph convolutional networks.arXiv preprint arXiv:1609.02907,

arXiv
[2016]

Pc-conv: Unifying homophily and heterophily with two-fold filtering

Li, B., Pan, E., and Kang, Z. Pc-conv: Unifying homophily and heterophily with two-fold filtering. InProceedings of the AAAI conference on artificial intelligence, volume 38, pp. 13437–13445, 2024a. Li, B., Chen, Z., Han, H., Zeng, S., Liu, J., and Tang, J. Unveiling mode conn...

2022 arXiv
[2017]

and Chung, F

Shen, X. and Chung, F. L. Network embedding for cross-network node classification.arXiv preprint arXiv:1901.07264,

1901 arXiv
[2019]

Text-space graph foundation models: Comprehensive benchmarks and new insights.arXiv preprint arXiv:2406.10727,

Chen, Z., Mao, H., Liu, J., Song, Y ., Li, B., Jin, W., Fatemi, B., Tsitsulin, A., Perozzi, B., Liu, H., et al. Text-space graph foundation models: Comprehensive benchmarks and new insights.arXiv preprint arXiv:2406.10727,

arXiv
[2020]

Prov- able filter for real-world graph clustering.arXiv preprint arXiv:2403.03666,

Xie, X., Pan, E., Kang, Z., Chen, W., and Li, B. Prov- able filter for real-world graph clustering.arXiv preprint arXiv:2403.03666,

arXiv
[2021]

Higher-order structure boosts link prediction on temporal graphs.arXiv preprint arXiv:2505.15746,

Liu, J., Hua, Z., Xie, Y ., Li, B., Shomer, H., Song, Y ., Hassani, K., and Tang, J. Higher-order structure boosts link prediction on temporal graphs.arXiv preprint arXiv:2505.15746,

arXiv
[2022]

One2multi graph autoencoder for multi-view graph clustering

Fan, S., Wang, X., Shi, C., Lu, E., Lin, K., and Wang, B. One2multi graph autoencoder for multi-view graph clustering. Inproceedings of the web conference 2020, pp. 3070–3076,

2020
[2023]

Graph domain adaptation: Challenges, progress and prospects.arXiv preprint arXiv:2402.00904,

Shi, B., Wang, Y ., Guo, F., Xu, B., Shen, H., and Cheng, X. Graph domain adaptation: Challenges, progress and prospects.arXiv preprint arXiv:2402.00904,

arXiv
[2024]

Adaptive uni- versal generalized pagerank graph neural network

Chien, E., Peng, J., Li, P., and Milenkovic, O. Adaptive uni- versal generalized pagerank graph neural network. In9th International Conference on Learning Representations, ICLR 2021,,

2021
[2025]

Magnn: Metapath aggregated graph neural network for heterogeneous graph embedding

Fu, X., Zhang, J., Meng, Z., and King, I. Magnn: Metapath aggregated graph neural network for heterogeneous graph embedding. InProceedings of the web conference 2020, pp. 2331–2341,

2020

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

+ ln 1 δ + λ2 4NS +D γ/2 S,T (P;λ) . Proposition 1(Bound for Dγ S,T (P;λ) , Adaptation from (Ma et al., 2021; Mao et al., 2024a; Fang et al., 2025)).For any γ≥0 , and under the assumption that the prior distribution P over the classification function family Φ is defined, we establish a bound for the domain discrepancy measureD γ/2 S,T (P;λ). Specifically,...

work page 2021

[2] [5]

Mamba-Based Graph Convolutional Networks: Tackling Over-smoothing with Selective State Space

He, X., Wang, Y ., Fan, W., Shen, X., Juan, X., Miao, R., and Wang, X. Mamba-based graph convolutional networks: Tackling over-smoothing with selective state space.arXiv preprint arXiv:2501.15461,

work page Pith review arXiv

[3] [6]

Robust mid-pass filtering graph convolutional networks

Huang, J., Du, L., Chen, X., Fu, Q., Han, S., and Zhang, D. Robust mid-pass filtering graph convolutional networks. InProceedings of the ACM Web Conference 2023, pp. 328–338,

work page 2023

[4] [10]

Rethinking propagation for unsupervised graph domain adaptation

Liu, M., Fang, Z., Zhang, Z., Gu, M., Zhou, S., Wang, X., and Bu, J. Rethinking propagation for unsupervised graph domain adaptation. InProceedings of the AAAI Confer- ence on Artificial Intelligence, volume 38, pp. 13963– 13971, 2024a. Liu, M., Zhang, Z., Tang, J., Bu, J., He, B., and Zhou, S. Re- visiting, benchmarking and understanding unsupervised gra...

arXiv 2024

[5] [11]

Demystifying structural disparity in graph neural networks: Can one size fit all?Advances in neural information processing systems, 36, 2024a

Mao, H., Chen, Z., Jin, W., Han, H., Ma, Y ., Zhao, T., Shah, N., and Tang, J. Demystifying structural disparity in graph neural networks: Can one size fit all?Advances in neural information processing systems, 36, 2024a. Mao, H., Du, L., Zheng, Y ., Fu, Q., Li, Z., Chen, X., Han, S., and Zhang, D. Source free unsupervised graph domain adaptation.The 17th...

work page 2024

[6] [12]

URL https:// openreview.net/forum?id=zrQIc9mQQN. Nt, H. and Maehara, T. Revisiting graph neural net- works: All we have is low-pass filters.arXiv preprint arXiv:1905.09550,

arXiv 1905

[7] [14]

Adversarial deep network embedding for cross-network node classification

Shen, X., Dai, Q., Chung, F.-l., Lu, W., and Choi, K.-S. Adversarial deep network embedding for cross-network node classification. InProceedings of the AAAI confer- ence on artificial intelligence, volume 34, pp. 2991–2999, 2020a. Shen, X., Dai, Q., Mao, S., Chung, F.-l., and Choi, K.-S. Network together: Node classification via cross-network deep network...

work page 1935

[8] [16]

Unsu- pervised domain adaptive graph convolutional networks

Wu, M., Pan, S., Zhou, C., Chang, X., and Zhu, X. Unsu- pervised domain adaptive graph convolutional networks. InProceedings of The Web Conference 2020, pp. 1457– 1467,

work page 2020