Pith. sign in

REVIEW 4 major objections 8 minor 54 references

Hierarchical Vector Quantized Graph Autoencoder with Annealing-Based Code Selection

T0 review · 4 major / 8 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a graph autoencoder with a vector-quantization bottleneck, annealed code selection, and a second-layer clustering codebook beats 16 self-supervised baselines on link prediction and node classification.

desk verdict Link prediction wins are probably real, but the node classification SOTA claim is oversold by unmatched baselines and margins inside one standard deviation. read the letter →

arxiv 2504.12715 v1 pith:TVHBKZXJ submitted 2025-04-17 cs.LG

Long Zeng , Jianxiang Yu , Jiapeng Zhu , Qingsong Zhong , Xiang Li This is my paper · ORCID
classification cs.LG
keywords graphself-supervisedlearningvectorquantizationautoencodercodebookunderutilizationannealing-basedcodeselectionhierarchicallinkpredictionnodeclassification
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

Vector-quantized autoencoders, a staple of image and audio generation, have been slow to reach graphs. This paper tries to close that gap by showing that inserting a discrete codebook between a graph encoder and decoder helps the model capture graph topology, and by solving two obstacles that appear when quantization meets graph data: many codebook entries go unused, and the entries that are used are treated as isolated points even though graph nodes are related. The proposed model, HQA-GAE, replaces hard code selection with a temperature-annealed softmax, so training first explores many codes and later concentrates on the useful ones, and it stacks a second, smaller codebook on top of the first so similar codes are pulled together. On eight datasets the model reports the best link-prediction score on all of them and the best node-classification score on six of eight, against 16 contrastive and autoencoding baselines. If these results hold, perturbation-free graph self-supervision is a stronger route than the augmentation-heavy methods that dominate the area.

What carries the argument

The load-bearing object is the two-layer vector-quantized bottleneck between the encoder and the decoders. A first-layer codebook of $M$ embeddings quantizes each encoder output $h_i$ to the nearest code $e_{1,i}$; the node decoder reconstructs features from $e_{1,i}$ while an edge decoder scores links from $h_i$. Because hard argmax selection starves most codes, selection is softened to a probability $p_{i,j} \propto \exp(\mathrm{sim}(h_i, e_j)/T)$ with the temperature annealed as $T_k = \max(\gamma T_{k-1}, \epsilon)$, so early training explores the codebook broadly and later training sharpens onto the effective codes. To fight codebook sparsity, a second-layer codebook of $C < M$ embeddings clusters the first-layer codes by maximizing $\sum_j \sum_{i \in S_j} \mathrm{sim}(e_{1,i}, e_{2,j})$, which pulls similar codes together and keeps dissimilar codes apart. The training loss combines scaled-cosine node reconstruction, negative-sampled edge reconstruction, and two vector-quantization losses, one per codebook layer.

What would settle it

A direct test is to train VQ-GAE and a plain GAE with an MLP encoder that sees only node features, as in the paper's preliminary experiment, but on a graph with planted community structure, and measure whether nodes sharing a code but having different neighborhoods are reconstructed more accurately than the plain GAE would predict. If the gap between VQ-GAE and GAE disappears when node features are made uninformative, the claimed structural mechanism is confirmed; if the gap persists, quantization is helping in some other way.

Watch

Extended reading notes

Core claim

The central discovery is that vector quantization does useful work in graph autoencoders even though the quantized code is not the final node representation. Because the codebook is much smaller than the node set, nodes with similar raw features are forced into the same code; the paper argues that reconstruction then can only succeed if the decoder reads the structural differences between those nodes, and that this pressure feeds back through straight-through gradients to make the encoder's continuous embeddings carry more topology. On top of that mechanism, HQA-GAE addresses codebook underutilization by selecting codes with a softmax over code similarities at a temperature that starts broad and sharpens as training proceeds, and it addresses codebook sparsity by adding a second-layer codebook that effectively clusters the first-layer codes. The paper reports that this design outperforms 16 baselines in self-supervised link prediction and node classification across eight datasets, with average ranks of 1.00 and 1.25 respectively, and that the two-layer codebook improves clustering metrics over a single-layer codebook.

Load-bearing premise

The explanation for the model's advantage rests on the premise that forcing similar-featured nodes into the same code makes the decoder rely on each node's structural context, and that this pressure carries back into the encoder; if that premise fails, the empirical gains would stand without the proposed mechanism.

Editorial extensions

If this is right

  • If the reported results hold, self-supervised graph learning does not need hand-crafted augmentations or masking: a perturbation-free reconstruction objective with a discrete bottleneck matches or beats contrastive and masked-autoencoder baselines.
  • Annealing-based code selection is an alternative to Gumbel-Softmax for improving codebook utilization in graph vector-quantization models, and the experiments tie utilization and downstream accuracy to the decay factor, with performance peaking near $\gamma = 0.9$.
  • The hierarchical second-layer codebook improves k-means clustering metrics on node representations, implying that relating codes to each other yields better representation geometry for downstream tasks.
  • The method's time and space complexity are linear in the number of nodes and edges when hyperparameters are fixed, which supports scaling to the 169k-node ogbn-arxiv graph.
  • Concatenating intermediate GNN layer outputs consistently beats using only the final layer as the node representation, across GCN, GAT, GraphSAGE, and GIN encoders.

Reading between the lines

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

  • I infer the mechanism could be tested directly: track which nodes share first-layer codes and compare their neighborhoods; if same-code nodes with different structure are reconstructed accurately, the structural-differentiation story is supported, and if not, the gain must come from the discrete bottleneck acting as a regularizer rather than from topology capture.
  • The two-layer codebook resembles a learned cluster hierarchy over the first-layer codes, so a natural extension is to make the number of second-layer codes adaptive to the graph's class or community structure rather than fixed, which the sensitivity analysis already hints at.
  • Because the final embeddings are the continuous encoder outputs rather than the discrete codes, HQA-GAE is not yet a discrete graph representation; it behaves more like a regularized autoencoder, and a follow-up that uses the codes themselves as downstream features would test whether the discrete space is truly informative.
  • The annealing schedule is controlled by one decay factor, and datasets with very different sizes or feature distributions may need dataset-specific schedules; a self-tuning version that adapts temperature to codebook utilization is a concrete testable improvement.
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

4 major / 8 minor

Summary. The paper extends VQ-VAE to graph self-supervised learning, proposing HQA-GAE, which uses a GNN encoder, a softmax-based annealed code selection (Eqs. 4-5), and a hierarchical two-layer codebook (Eqs. 6-9) to address codebook underutilization and codebook space sparsity. In experiments on eight datasets, it reports average rank 1.00 on link prediction (Table 1) and 1.25 on node classification (Table 2), and it includes ablations on codebook size, temperature decay, loss weights, encoder choice, clustering quality, efficiency, and t-SNE visualizations.

Significance. The paper addresses a real gap in applying VQ-VAE to graph SSL and proposes two concrete, well-motivated mechanisms: annealed code selection to counteract winner-take-all underutilization, and a second-layer codebook to reduce sparsity. Strengths include the public code release, broad benchmark coverage, extensive ablations, a linear-complexity analysis, and a falsifiable claim about codebook utilization. If the empirical results survive matched-environment evaluation and the method specification is completed, HQA-GAE would be a competitive and conceptually novel graph SSL approach. The main limitation is that the headline empirical claim currently rests on uncontrolled baseline comparisons and untested margins, especially for node classification.

major comments (4)
  1. [§5.3, Table 2, Appendix A.3] Table 2 mixes node-classification results from at least three different sources: DGI, GIC, MVGRL, BGRL, ARVGA, GraphMAE, MaskGAE, and S2GAE are taken from the S2GAE paper; Bandana's numbers are taken from Bandana; and only VGAE, SeeGera, and HQA-GAE are run by the authors. Footnote 2 states that MaskGAE and Bandana results are 'notably affected by software versions, CUDA environments, and random seeds,' yet those published numbers are used in the table without re-running in a matched environment. On five of the eight datasets (Cora, PubMed, Photo, Computers, Physics), HQA-GAE's margin over the runner-up is within one standard deviation (e.g., Cora 88.78±1.03 vs 88.59±1.35; PubMed 88.49±0.53 vs 88.16±0.57), and HQA-GAE is second on CS and ogbn-arxiv. No paired significance test is reported. The abstract's claim that the model 'outperforms 16 representative baseline methods in self-supervised link prediction and node classification' is therefore not established for node classification; the conclusion's more cautious 'comparable performance' is closer to the evidence. The authors should either re-run all baselines under the same protocol or substantially temper the node-classification claim.
  2. [§4.2, Eqs. (4)-(5)] The annealed code selection is the central novelty, but the forward pass is incompletely specified. The paper defines a softmax probability p_{i,j} over codes and a decaying temperature T_k, but never states whether training selects a code by sampling from this distribution, by the expected codebook embedding, or by argmax (with straight-through gradients), nor how the selection is made at inference. This distinction matters because sampling and expected-embedding behave differently under the annealing schedule and under the stop-gradient used in Eq. (9). Without this specification, the method is not fully reproducible and the ablation in Figure 3 cannot be interpreted unambiguously.
  3. [§4.1, Figure 2] The paper's stated mechanism—that vector quantization 'forces the model to leverage the structural difference when reconstructing their raw features'—is supported only by the bar-chart comparison in Figure 2. That figure has no error bars, does not report the number of seeds, and reports no statistical test. Moreover, the comparison uses an MLP encoder, whereas the final HQA-GAE uses a GCN encoder, so Figure 2 does not directly establish the benefit of VQ in the model configuration actually used in Tables 1 and 2. Please add error bars or statistical tests and an ablation that removes VQ from the full GCN-based model.
  4. [§5.2, Table 1, Appendix A.3] For link prediction, most baseline numbers in Table 1 are taken from the Bandana paper rather than run under the authors' environment; only DGI and GIC are reported as author-run. Given the sensitivity to software and CUDA versions documented in Footnote 2, the link-prediction superiority claim also lacks matched-environment confirmation. The margins are larger than in node classification, which reduces the severity of this concern, but a matched re-run or an explicit statement of protocol equivalence is needed before the abstract's claim is fully supported for link prediction as well.
minor comments (8)
  1. [Eq. (4)] The denominator of the softmax in Eq. (4) is written with the summation index i, but it should sum over the codebook index j; as written, p_{i,j} is not a normalized distribution over codes.
  2. [§5.3, Appendix A.3, References] The citation to S2GAE in §5.3 and Appendix A.3 is given as [49] (Bandana) rather than [36]; additionally, the baseline name 'AGVGA' appears in §5.1 and A.3, while the tables and reference [27] use 'ARVGA'.
  3. [Table 4] Table 4 contains typos in dataset names: 'Cara', 'CitrSeer', and 'Cumputers' should be 'Cora', 'CiteSeer', and 'Computers'.
  4. [Abstract vs. Conclusion] The abstract says the model 'outperforms 16 representative baseline methods in self-supervised link prediction and node classification,' while the conclusion says it 'shows comparable performance in node classification'; these statements are inconsistent and should be aligned.
  5. [Figure 2] Figure 2 should state the number of runs and show error bars or confidence intervals; without them, the phrase 'significantly enhance' in Section 4.1 is not substantiated.
  6. [Appendix A.4] Appendix A.4 refers to the GitHub repository for the remaining hyperparameters; for archival reproducibility, the per-dataset values of M, C, T0, gamma, lambda, alpha, beta, and the negative-sampling count should be included in the paper itself.
  7. [§5.5] Section 5.5 says MaskGAE 'lags behind ours in performance by nearly 10%' on Physics link prediction; Table 1 shows MaskGAE AUC 91.94 vs HQA-GAE 98.37 (about 6.4 percentage points) and AP 86.33 vs 98.22 (about 11.9 percentage points), so the claim should specify the metric.
  8. [Eq. (8)] The negative-sampling edge reconstruction loss uses a set E− of negative edges, but the paper does not state how many negative edges are sampled per positive edge; this should be specified for reproducibility.

Circularity Check

2 steps flagged · score 2.0 of 10

Minor self-definitional ablations; central SOTA comparisons are externally benchmarked.

  1. self definitional [Section 4.3 (Eq. 6) and Section 5.4.2 (Table 3)]
    "The optimization objective, similar to k-means clustering [2], is then defined as: O = max Σ sim(e1,i, e2,j). ... node representations learned from both codebook designs are subjected to k-means clustering [2, 11] ... Table 3 shows that our hierarchical codebook generally outperforms the single-layer version, demonstrating superior clustering performance."

    Eq. (6) trains the second-layer codebook to maximize the similarity between each first-layer code and its assigned second-layer center, i.e., it is a k-means-like clustering objective over the first-layer codes. Through Lvq2 and Lvq1 (Eq. 9), encoder outputs are pulled toward the first-layer codes and the first-layer codes toward the second-layer centers. Table 3 then measures the clustering quality (NMI/ARI/SC after k-means) of the resulting node representations. The reported clustering benefit is therefore largely a restatement of the training objective; the only external component is whether the induced clusters agree with ground-truth labels. The headline link-prediction and node-classification comparisons are not affected by this construction.

  2. self definitional [Section 5.4.1, Figure 3(a)]
    "As shown in Figure 3(a), codebook utilization increases as the decay factor rises, eventually reaching saturation. This is expected, as a higher decay factor slows the annealing process, allowing for more exploration and leading to a more diverse use of the codebook space."

    Under the softmax selection in Eq. (4), p_ij = exp(s_ij/T)/Σ exp(s_ij/T), a higher T yields a more uniform selection distribution, and Eq. (5), T_k = max(γT_{k-1}, ε), keeps T high longer for larger γ. Thus the utilization-versus-decay trend in Figure 3(a) follows directly from the definitions of the mechanism and the utilization metric; the paper even labels it 'expected.' The node-classification curve in Figure 3(b) is the empirical part, so this self-definitional sanity check is not load-bearing for the main claim.

full rationale

The central derivation of HQA-GAE is self-contained: the model is trained with reconstruction and vector-quantization losses (Eqs. 8-10), and its headline claims are evaluated on external link-prediction and node-classification benchmarks against published baseline numbers. No fitted parameter is renamed as a prediction, and no load-bearing premise is imported from a self-citation; the only overlapping-author reference, SeeGera [23], serves as a baseline rather than as a justification. The two flagged items are auxiliary analyses in which a measured benefit is partly a restatement of the training objective: the second-layer codebook is explicitly a k-means-style clustering objective over first-layer codes, and codebook utilization under annealing is a direct consequence of the softmax temperature schedule. These do not affect the central SOTA claim, so the overall circularity is low. The mixed-source baseline numbers in Table 2 and footnote 2 about split sensitivity are reproducibility and statistical concerns, not circularity.

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

This is an empirical method paper. The result rests on a set of tuned hyperparameters (codebook sizes, temperature schedule, VQ loss weights, reconstruction loss exponent, architecture) and on three unproven domain assumptions about how quantization gradients and hierarchical clustering transfer to downstream tasks. No invented physical entities are introduced; the second-layer codebook is an architectural component, not a postulated external entity.

free parameters (6)
  • First-layer codebook size M = 2^4 to 2^12 (per dataset)
    Selected by grid search over 2^1 to 2^12; Figure 6 shows optimal 2^8 for Cora/CiteSeer and 2^4 for PubMed, so downstream performance guides the choice.
  • Second-layer codebook size C = 2^1 to 2^8 (per dataset)
    Selected by grid search; Figure 7 shows performance drops when C is smaller than the number of classes, so C is chosen above the class count.
  • Initial temperature T0 and decay factor gamma = T0 in repository; gamma = 0.9
    The annealing schedule in Eq. (5) is tuned against node classification accuracy; decay factor 0.9 is chosen from {0, 0.3, 0.6, 0.9, 0.99, 0.999, 0.9999}.
  • VQ loss weights alpha and beta = alpha = 1; beta in {0, 0.001, 0.01, 0.1, 1}
    Appendix B.1: alpha set to 1 across datasets; beta chosen so that clustering improves while classification stays flat.
  • Scaled cosine error exponent lambda = not stated in text
    Controls sensitivity of node reconstruction loss in Eq. (8); value is delegated to the GitHub repository.
  • Encoder/decoder architecture and training hyperparameters = GCN encoder, GAT decoder, hidden sizes and learning rate in repository
    Network capacity and optimization settings are chosen empirically; the repository is the only complete specification.
assumptions (4)
  • standard math Straight-through gradient estimation gives useful gradients for the non-differentiable code lookup
    Invoked in Sections 3.3 and 4.1; standard in VQ-VAE literature, but no convergence guarantee for graphs.
  • domain assumption Quantizing each node to one code and reconstructing features from that code forces the encoder to encode topology
    Section 4.1 mechanism claim; supported only by the GAE vs VQ-GAE experiment in Figure 2.
  • domain assumption A second-layer codebook trained with the k-means-like objective in Eq. (6) reduces codebook space sparsity and improves downstream representations
    Section 4.3; no formal proof that the global optimum of Eq. (6) transfers to node classification or link prediction.
  • domain assumption Public datasets and baseline results from the original papers are comparable despite differences in evaluation protocols
    Tables 1 and 2 mix results from Bandana and S2GAE with self-run experiments under 5-fold cross-validation; comparability is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hierarchical Vector Quantized Graph Autoencoder with Annealing-Based Code Selection." pith.science (2026). https://pith.science/paper/TVHBKZXJ

@misc{pith2026250412715,
  author       = {Pith},
  title        = {Pith review of: Hierarchical Vector Quantized Graph Autoencoder with Annealing-Based Code Selection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TVHBKZXJ}},
  note         = {Machine review of arXiv:2504.12715}
}
read the original abstract

Graph self-supervised learning has gained significant attention recently. However, many existing approaches heavily depend on perturbations, and inappropriate perturbations may corrupt the graph's inherent information. The Vector Quantized Variational Autoencoder (VQ-VAE) is a powerful autoencoder extensively used in fields such as computer vision; however, its application to graph data remains underexplored. In this paper, we provide an empirical analysis of vector quantization in the context of graph autoencoders, demonstrating its significant enhancement of the model's capacity to capture graph topology. Furthermore, we identify two key challenges associated with vector quantization when applying in graph data: codebook underutilization and codebook space sparsity. For the first challenge, we propose an annealing-based encoding strategy that promotes broad code utilization in the early stages of training, gradually shifting focus toward the most effective codes as training progresses. For the second challenge, we introduce a hierarchical two-layer codebook that captures relationships between embeddings through clustering. The second layer codebook links similar codes, encouraging the model to learn closer embeddings for nodes with similar features and structural topology in the graph. Our proposed model outperforms 16 representative baseline methods in self-supervised link prediction and node classification tasks across multiple datasets.

Figures

Figures reproduced from arXiv: 2504.12715 by the authors.

Figure 1
Figure 1. The schematic diagram of HQA-GAE, including graph encoding and decoding, annealing-based code selection, [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. The performance of node classification tasks for [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The effect of annealing-based code selection. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (5 more)
Figure 6
Figure 6. Figure 6: The impact of varying the size of the first layer [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: The effect of the size of the second layer codebook [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: The effect of 𝛽 on clustering task. 0 0.5 1.0 1.5 2.0 Value of 75.0 77.5 80.0 82.5 85.0 87.5 90.0 Accuracy (%) (a) Performace under different Cora CiteSeer 0 0.001 0.01 0.1 1.0 Value of 75.0 77.5 80.0 82.5 85.0 87.5 90.0 Accuracy (%) (b) Performace under different Cora…
Figure 9
Figure 9. Figure 9: The effect of 𝛼 and 𝛽 on node classification. B.2 Impact of Encoder Design We examine how the architecture of the encode influences the qual￾ity of representations. We employ various encoders, including GCN, GAT, GraphSAGE, and GIN, across the Cora, PubMed, and CiteSee…
Figure 10
Figure 10. Figure 10: The t-SNE visualizations of various latent representations of the CiteSeer dataset, encoded by different graph [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 13 canonical work pages

  1. [1]

    Stanley C Ahalt, Ashok K Krishnamurthy, Prakoon Chen, and Douglas E Melton

    work page

  2. [2]

    David Arthur and Sergei Vassilvitskii. 2006. k-means++: The advantages of careful seeding. Technical Report. Stanford

    2006

  3. [3]

    Peter W Battaglia, Jessica B Hamrick, Victor Bapst, Alvaro Sanchez-Gonzalez, Vinicius Zambaldi, Mateusz Malinowski, Andrea Tacchetti, David Raposo, Adam Santoro, Ryan Faulkner, et al. 2018. Relational inductive biases, deep learning, and graph networks. arXiv preprint arXiv:1806.01261 (2018)

    arXiv 2018

  4. [4]

    Yoann Boget, Magda Gregorova, and Alexandros Kalousis. 2024. Discrete Graph Auto-Encoder. Transactions on Machine Learning Research (2024)

    work page 2024

  5. [5]

    Alexandre Défossez, Jade Copet, Gabriel Synnaeve, and Yossi Adi. 2022. High fidelity neural audio compression. arXiv preprint arXiv:2210.13438 (2022)

    arXiv 2022

  6. [6]

    Mucong Ding, Kezhi Kong, Jingling Li, Chen Zhu, John Dickerson, Furong Huang, and Tom Goldstein. 2021. VQ-GNN: A universal framework to scale up graph neural networks using vector quantization. Advances in Neural Information Processing Systems 34 (2021), 6733–6746

    work page 2021

  7. [7]

    Patrick Esser, Robin Rombach, and Bjorn Ommer. 2021. Taming transformers for high-resolution image synthesis. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 12873–12883

    2021

  8. [8]

    Matthias Fey and Jan Eric Lenssen. 2019. Fast graph representation learning with PyTorch Geometric. arXiv preprint arXiv:1903.02428 (2019)

    arXiv 2019

Show all 54 references
  1. [9]

    Stephen Grossberg. 1987. Competitive learning: From interactive activation to adaptive resonance. Cognitive science 11, 1 (1987), 23–63

    1987

  2. [10]

    Will Hamilton, Zhitao Ying, and Jure Leskovec. 2017. Inductive representation learning on large graphs. Advances in neural information processing systems 30 (2017)

    2017

  3. [11]

    John A Hartigan and Manchek A Wong. 1979. Algorithm AS 136: A k-means clustering algorithm. Journal of the royal statistical society. series c (applied statistics) 28, 1 (1979), 100–108

    1979

  4. [12]

    Arman Hasanzadeh, Ehsan Hajiramezanali, Krishna Narayanan, Nick Duffield, Mingyuan Zhou, and Xiaoning Qian. 2019. Semi-implicit graph variational auto-encoders. Advances in neural information processing systems 32 (2019)

    2019

  5. [13]

    Kaveh Hassani and Amir Hosein Khasahmadi. 2020. Contrastive multi-view representation learning on graphs. InInternational conference on machine learning. PMLR, 4116–4126

    2020

  6. [14]

    Zhenyu Hou, Yufei He, Yukuo Cen, Xiao Liu, Yuxiao Dong, Evgeny Kharlamov, and Jie Tang. 2023. Graphmae2: A decoding-enhanced masked self-supervised graph learner. In Proceedings of the ACM web conference 2023 . 737–746

    2023

  7. [15]

    Zhenyu Hou, Xiao Liu, Yukuo Cen, Yuxiao Dong, Hongxia Yang, Chunjie Wang, and Jie Tang. 2022. Graphmae: Self-supervised masked graph autoencoders. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. 594–604

    2022

  8. [16]

    Weihua Hu, Matthias Fey, Marinka Zitnik, Yuxiao Dong, Hongyu Ren, Bowen Liu, Michele Catasta, and Jure Leskovec. 2020. Open graph benchmark: Datasets for machine learning on graphs. Advances in neural information processing systems 33 (2020), 22118–22133

    2020

  9. [17]

    Weihua Hu, Bowen Liu, Joseph Gomes, Marinka Zitnik, Percy Liang, Vijay Pande, and Jure Leskovec. 2019. Strategies for pre-training graph neural networks.arXiv preprint arXiv:1905.12265 (2019)

    2019 arXiv

  10. [18]

    Eric Jang, Shixiang Gu, and Ben Poole. 2016. Categorical reparameterization with gumbel-softmax. arXiv preprint arXiv:1611.01144 (2016)

    2016 arXiv

  11. [19]

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

    2016 arXiv

  12. [20]

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

    2016 arXiv

  13. [21]

    Scott Kirkpatrick, C Daniel Gelatt Jr, and Mario P Vecchi. 1983. Optimization by simulated annealing. science 220, 4598 (1983), 671–680

    1983

  14. [22]

    Jintang Li, Ruofan Wu, Wangbin Sun, Liang Chen, Sheng Tian, Liang Zhu, Changhua Meng, Zibin Zheng, and Weiqiang Wang. 2023. What’s Behind the Mask: Understanding Masked Graph Modeling for Graph Autoencoders. In Pro- ceedings of the 29th ACM SIGKDD Conference on Knowledge Disco...

    2023

  15. [23]

    Xiang Li, Tiandi Ye, Caihua Shan, Dongsheng Li, and Ming Gao. 2023. Seegera: Self-supervised semi-implicit graph variational auto-encoders with masking. In Proceedings of the ACM web conference 2023 . 143–153

    2023

  16. [24]

    Yixin Liu, Ming Jin, Shirui Pan, Chuan Zhou, Yu Zheng, Feng Xia, and S Yu Philip

  17. [25]

    Costas Mavromatis and George Karypis. 2021. Graph infoclust: Maximizing coarse-grain mutual information in graphs. In Pacific-Asia Conference on Knowl- edge Discovery and Data Mining . Springer, 541–553

    2021

  18. [26]

    Tomas Mikolov. 2013. Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781 (2013)

    2013 arXiv

  19. [27]

    Shirui Pan, Ruiqi Hu, Guodong Long, Jing Jiang, Lina Yao, and Chengqi Zhang

  20. [28]

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. 2019. Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing system...

    2019

  21. [29]

    William M Rand. 1971. Objective criteria for the evaluation of clustering methods. Journal of the American Statistical association 66, 336 (1971), 846–850

    1971

  22. [30]

    Ali Razavi, Aaron Van den Oord, and Oriol Vinyals. 2019. Generating diverse high-fidelity images with vq-vae-2. Advances in neural information processing systems 32 (2019)

    2019

  23. [31]

    Peter J Rousseeuw. 1987. Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. Journal of computational and applied mathematics 20 (1987), 53–65

    1987

  24. [32]

    Hinrich Schütze, Christopher D Manning, and Prabhakar Raghavan. 2008. Intro- duction to information retrieval . Vol. 39. Cambridge University Press Cambridge

    2008

  25. [33]

    Prithviraj Sen, Galileo Namata, Mustafa Bilgic, Lise Getoor, Brian Galligher, and Tina Eliassi-Rad. 2008. Collective classification in network data. AI magazine 29, 3 (2008), 93–93

    2008

  26. [34]

    Oleksandr Shchur, Maximilian Mumme, Aleksandar Bojchevski, and Stephan Günnemann. 2018. Pitfalls of graph neural network evaluation. arXiv preprint arXiv:1811.05868 (2018)

    2018 arXiv

  27. [35]

    Fan-Yun Sun, Jordan Hoffmann, Vikas Verma, and Jian Tang. 2019. Infograph: Un- supervised and semi-supervised graph-level representation learning via mutual information maximization. arXiv preprint arXiv:1908.01000 (2019)

    2019 arXiv

  28. [36]

    Qiaoyu Tan, Ninghao Liu, Xiao Huang, Soo-Hyun Choi, Li Li, Rui Chen, and Xia Hu. 2023. S2gae: Self-supervised graph autoencoders are generalizable learners with graph masking. In Proceedings of the sixteenth ACM international conference on web search and data mining . 787–795

    2023

  29. [37]

    Shantanu Thakoor, Corentin Tallec, Mohammad Gheshlaghi Azar, Mehdi Azabou, Eva L Dyer, Remi Munos, Petar Veličković, and Michal Valko. 2021. Large- scale representation learning on graphs via bootstrapping. arXiv preprint arXiv:2102.06514 (2021)

    2021 arXiv

  30. [38]

    Aaron Van Den Oord, Oriol Vinyals, et al. 2017. Neural discrete representation learning. Advances in neural information processing systems 30 (2017)

    2017

  31. [39]

    Laurens Van der Maaten and Geoffrey Hinton. 2008. Visualizing data using t-SNE. Journal of machine learning research 9, 11 (2008)

    2008

  32. [40]

    Petar Velickovic, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Lio, Yoshua Bengio, et al. 2017. Graph attention networks. stat 1050, 20 (2017), 10–48550

    2017

  33. [41]

    Petar Veličković, William Fedus, William L Hamilton, Pietro Liò, Yoshua Bengio, and R Devon Hjelm. 2018. Deep graph infomax. arXiv preprint arXiv:1809.10341 (2018)

    2018 arXiv

  34. [42]

    Jiancan Wu, Xiang Wang, Fuli Feng, Xiangnan He, Liang Chen, Jianxun Lian, and Xing Xie. 2021. Self-supervised graph learning for recommendation. In Proceed- ings of the 44th international ACM SIGIR conference on research and development in information retrieval. 726–735

    2021

  35. [43]

    Jun Xia, Chengshuai Zhao, Bozhen Hu, Zhangyang Gao, Cheng Tan, Yue Liu, Siyuan Li, and Stan Z Li. 2023. Mole-bert: Rethinking pre-training graph neural networks for molecules. (2023)

    2023

  36. [44]

    Yaochen Xie, Zhao Xu, Jingtun Zhang, Zhengyang Wang, and Shuiwang Ji. 2022. Self-supervised learning of graph neural networks: A unified review. IEEE trans- actions on pattern analysis and machine intelligence 45, 2 (2022), 2412–2429

    2022

  37. [45]

    Ling Yang, Ye Tian, Minkai Xu, Zhongyi Liu, Shenda Hong, Wei Qu, Wentao Zhang, CUI Bin, Muhan Zhang, and Jure Leskovec. 2024. VQGraph: Rethink- ing graph representation space for bridging GNNs and MLPs. In The Twelfth International Conference on Learning Representations

    2024

  38. [46]

    Yuning You, Tianlong Chen, Yang Shen, and Zhangyang Wang. 2021. Graph contrastive learning automated. In International Conference on Machine Learning . PMLR, 12121–12132

    2021

  39. [47]

    Yuning You, Tianlong Chen, Yongduo Sui, Ting Chen, Zhangyang Wang, and Yang Shen. 2020. Graph contrastive learning with augmentations. Advances in neural information processing systems 33 (2020), 5812–5823

    2020

  40. [48]

    Neil Zeghidour, Alejandro Luebs, Ahmed Omran, Jan Skoglund, and Marco Tagliasacchi. 2021. Soundstream: An end-to-end neural audio codec. IEEE/ACM Transactions on Audio, Speech, and Language Processing 30 (2021), 495–507

    2021

  41. [49]

    Ziwen Zhao, Yuhua Li, Yixiong Zou, Jiliang Tang, and Ruixuan Li. 2024. Masked Graph Autoencoder with Non-discrete Bandwidths. In Proceedings of the ACM on Web Conference 2024. 377–388

    2024

  42. [50]

    Yanqiao Zhu, Yichen Xu, Feng Yu, Qiang Liu, Shu Wu, and Liang Wang. 2020. Deep graph contrastive representation learning. arXiv preprint arXiv:2006.04131 (2020)

    2020 arXiv

  43. [51]

    Yanqiao Zhu, Yichen Xu, Feng Yu, Qiang Liu, Shu Wu, and Liang Wang. 2021. Graph contrastive learning with adaptive augmentation. In Proceedings of the web conference 2021. 2069–2080. WWW ’25, April 28-May 2, 2025, Sydney, NSW, Australia Long Zeng, Jianxiang Yu, Jiapeng Zhu, Qi...

    2021

  44. [1990]

    Neural networks 3, 3 (1990), 277–290

    Competitive learning algorithms for vector quantization. Neural networks 3, 3 (1990), 277–290

    1990

  45. [2018]

    arXiv preprint arXiv:1802.04407 (2018)

    Adversarially regularized graph autoencoder for graph embedding. arXiv preprint arXiv:1802.04407 (2018)

    2018 arXiv

  46. [2022]

    IEEE transactions on knowledge and data engineering 35, 6 (2022), 5879–5900

    Graph self-supervised learning: A survey. IEEE transactions on knowledge and data engineering 35, 6 (2022), 5879–5900

    2022

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.