Pith. sign in

REVIEW 3 major objections 6 minor 54 references

Towards Effective Open-set Graph Class-incremental Learning

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read This paper proposes OGCIL, unifying graph class-incremental learning and open-set recognition: a prototypical CVAE replays old-class embeddings while a hypersphere loss rejects unknowns, yielding up to 17.6% higher open-set classification…

desk verdict New problem framing, but a representation-space mismatch in the written method makes the headline OSR numbers hard to trust without clarification. read the letter →

arxiv 2507.17687 v1 pith:ZFHV57XE submitted 2025-07-23 cs.LG

Jiazhen Chen , Zheng Ma , Sichao Fu , Mingbin Feng , Tony S. Wirjanto , Weihua Ou This is my paper · ORCID
classification cs.LG
keywords graphneuralnetworksclass-incrementallearningopen-setrecognitionout-of-distributiondetectioncatastrophicforgettingvariationalautoencoderprototypenodeclassification
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

Graph class-incremental learning has assumed closed worlds where every test node belongs to a known class. This paper argues that real deployments also face unknown classes, and that forgetting and open-set failure reinforce each other: forgotten old classes get mislabeled as unknown, while poorly handled unknowns destabilize what the model remembers. It introduces OGCIL, which tackles both at once by generating pseudo-embeddings: a prototypical conditional variational autoencoder replays old-class embeddings, a mixing strategy synthesizes out-of-distribution samples, and a prototypical hypersphere classification loss keeps each known class inside its own prototype-centered ball while pushing everything else outside. Across five graph benchmarks, OGCIL reports open-set classification rates up to 17.6 percentage points above the strongest competitor, with higher open-set AUC-ROC, suggesting the two goals can indeed be met in one framework.

What carries the argument

The load-bearing object is the prototypical hypersphere classification loss $L_{\mathrm{phsc}}$, built on the radial-basis similarity $\varphi(h(z),p_c)=\exp(-\|h(z)-p_c\|^2)$ between the CVAE-encoded embedding $h(z)=\mu_\phi(z)$ and each learnable class prototype $p_c$. For a sample of class $c$, the loss maximizes similarity to $p_c$ and minimizes it for every other prototype, so mixed pseudo-OOD samples and off-class samples are pushed outside all hyperspheres. The prototypical CVAE supports this by imposing a Gaussian latent prior $\mathcal{N}(p_c,I)$ per class, so decoded pseudo-IDs inherit the same centers, and the mixing step creates OOD samples that lie off all prototypes. Knowledge distillation over real and exemplar embeddings stabilizes the GNN representation so generated pseudo-samples remain valid across tasks.

What would settle it

Remove the space mismatch by evaluating OGCIL with the open-set score computed on the encoded embedding $h(z)=\mu_\phi(z)$ instead of the raw embedding $z$ in Eq. 8 on CoraFull; if open-set AUC-ROC changes materially, the published score is measuring distance in the wrong space and the method's rejection regions are not where the paper says they are. A simpler check: measure the mean distance $\|z-h(z)\|$ for known-class test nodes relative to the learned hypersphere radii; if it is comparable to or larger than the radii, Eq. 8 cannot be thresholding the same geometry the loss trained.

Watch

Extended reading notes

Core claim

The central claim is that a single framework can simultaneously mitigate catastrophic forgetting and detect unknown classes in graph class-incremental learning, a setting the paper says no previous method addresses. OGCIL decouples node representations into a task-stable part, regularized by knowledge distillation, and a task-variant encoder. A prototypical CVAE with latent prior centered on class prototypes generates pseudo in-distribution embeddings for old classes, enabling replay without raw graph storage; mixing these with current embeddings produces pseudo out-of-distribution samples. The prototypical hypersphere classification loss then anchors encoded in-distribution embeddings to their class prototypes via a radial-basis similarity and repels off-class and OOD samples, explicitly modeling unknowns as outliers instead of a single unknown cluster. The paper reports consistent OSCR and AUC-ROC gains over class-incremental and open-set baselines across CoraFull, Computer, Photo, CS, and Arxiv, with ablations indicating each component is necessary.

Load-bearing premise

The method's open-set score compares a raw node embedding to prototypes, while the training loss compares an encoded embedding to those same prototypes, so the whole detection scheme assumes these two spaces are effectively the same.

Editorial extensions

If this is right

  • Old-class knowledge can be replayed without storing raw graphs, so continual learners become lighter and more privacy-friendly.
  • Unknown samples are rejected per prototype rather than forced into one 'unknown' class, which should generalize better to multiple unseen classes.
  • The framework stays effective with as few as one exemplar per class and with different GNN backbones, suggesting it transfers across memory and architecture constraints.
  • Thresholding the open-set score $-\min_c \|z-p_c\|^2$ gives a simple deployment rule for flagging novel nodes at inference time.
  • Because replay happens in embedding space, the approach can be layered on top of existing exemplar-selection and regularization strategies.

Reading between the lines

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

  • A testable extension is to compute the open-set score on the encoded embedding $h(z)=\mu_\phi(z)$ instead of the raw embedding $z$; if results change materially, the paper's stated score is measuring distance in a different space than the one the loss trains.
  • The paper leaves implicit that the same embedding-level generation could transfer to non-graph continual learning or to task-incremental variants, since it never depends on reconstructing the raw input structure.
  • The dataset protocol, where unknowns of one task become known later, invites a follow-up evaluation of whether the model can re-identify previously rejected nodes as known, testing the interplay of rejection and memory.
  • Mixing-based OOD generation with a Beta coefficient could be extended to pair-class or adversarial mixup to better cover boundary regions and further sharpen rejection regions.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes OGCIL, a framework for open-set graph class-incremental learning in which a GNN backbone is trained incrementally over tasks with disjoint classes while test nodes may come from classes never seen during training. To mitigate catastrophic forgetting, the method trains a prototypical conditional variational autoencoder whose class-dependent Gaussian prior is centered on learned class prototypes, and generates pseudo in-distribution embeddings by decoding samples from that prior. To detect unknown classes, the method synthesizes pseudo out-of-distribution embeddings by linearly mixing embeddings from different classes, and trains a prototypical hypersphere classification loss that keeps known-class embeddings close to their prototypes while repelling all other samples. The paper reports experiments on five graph benchmarks, comparing against class-incremental and open-set baselines, with OSCR gains up to 17.6% over the best competitor, and includes ablations, hyperparameter studies, and t-SNE visualizations. The central claim is that OGCIL simultaneously reduces catastrophic forgetting and improves open-set detection in this unified setting.

Significance. If the representation-space issue identified below is resolved, this is a useful contribution: it opens a realistic problem setting that combines graph class-incremental learning with open-set recognition, and it provides a systematic empirical study over five datasets with several baselines. The strengths of the paper are its breadth of experiments, the ablations isolating each component (knowledge distillation, prototypical HSC loss, pseudo ID, and pseudo OOD generation), the sensitivity analyses over sample counts, hyperparameters, exemplar strategies, and backbone architectures, and the provision of pseudocode. The proposed method is also modular, as the knowledge-distillation component is stated to be replaceable by other regularization techniques. The main reservation is the load-bearing ambiguity about which representation is used at inference to compute the open-set score, which directly affects the validity of the reported OSCR and AUC-ROC numbers.

major comments (3)
  1. [Section 3.4.2, Eq. (6), and Section 3.5, Eq. (8)] The prototypical hypersphere classification loss is defined on the CVAE-encoded representation h(z) = mu_phi(z), so the prototypes p_c are trained in the latent h-space, but the open-set score at inference is written as s_open(z) = -min_c ||z - p_c||^2 using the raw GNN embedding z. Unless raw z and h(z) are shown to live in the same metric space, which the paper does not establish, distances from z to p_c are not meaningful and the threshold on s_open cannot separate known from unknown nodes. Since the headline results in Table 1 are computed from this score, the paper's central claim depends on an unstated inference-time representation choice. Please change Eq. (8) to use h(z) and specify how h is computed at inference, or justify that the two spaces coincide; if the experiments already used h(z), the equation and surrounding text must be corrected accordingly.
  2. [Section 3.3.3 and Section 3.4.1] Pseudo ID embeddings are generated by sampling h approximately N(p_c, I) and decoding to z_hat = D_theta(z|h), so replay samples live in the raw GNN-embedding space, and the mixing strategy in Eq. (5) operates on those raw-space embeddings. Equation (6) then applies the CVAE encoder h(.) to these pseudo IDs as if they were original embeddings, meaning the replay distribution is sampled in one space while the classification loss is evaluated in another. This double representation shift is neither stated nor justified; please clarify whether pseudo IDs are re-encoded before entering L_phsc and report the effect of this choice on the ablations shown in Figure 3.
  3. [Section 6.1 and overall reproducibility] The appendix states that the code will be released on GitHub upon publication, and the manuscript does not specify which representation was actually fed to the open-set score in the experiments beyond the formula in Eq. (8). Given the mismatch identified above, the reported OSCR and AUC-ROC improvements cannot be independently verified from the text. Please provide the exact evaluation procedure, including the representation used for the open-set score, the thresholding details, and ideally the code or a precise pseudo-code line for the inference step.
minor comments (6)
  1. [Section 3.3.1, Eq. (1)] The reconstruction term is weighted by lambda_reconst but the KL term is unweighted; please clarify whether the KL term is intentionally unweighted or should also carry a balancing coefficient.
  2. [Section 6.1, Algorithm 1] Line 8 of the pseudocode, 'Generate H ID embeddings via D_theta(.), {p_c}', contains a typo and should read 'Generate N_ID pseudo ID embeddings' using the variable N_ID declared in the input list.
  3. [Table 2] For CoraFull, the per-task known and unknown counts sum to more than the dataset's 45 classes across five tasks; please clarify how many classes are permanently unknown versus unknown in earlier tasks and known later, so that the splits are unambiguous.
  4. [Section 4.2] In the paragraph beginning 'In comparison with other open-set baselines', the phrase 'Our HSC loss' should be lowercase ('our HSC loss') because it appears mid-sentence.
  5. [Section 4.4.1] The sentence 'Whereas excessively large numbers introduce noise' is a fragment; please join it to the preceding sentence for readability.
  6. [Section 1 and Section 2.2] The claim of being the first to unify GCIL and OSR should be sharpened in light of the discussed OpenWRF and LifeLongGNN methods, by stating explicitly how the proposed setting differs from those works (e.g., no full historical data access).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the OGCIL framework is supported by external empirical comparisons and standard supervised training, not by fitting-as-prediction or self-citation.

full rationale

The paper's derivation chain is empirical and self-contained. The OGCIL training objective (Eq. 7) combines the prototypical hypersphere loss L_phsc, the prototypical CVAE loss L_pcvae, and knowledge distillation L_kd; the class prototypes p_c are learned jointly with the CVAE and then reused both for pseudo-ID generation (Eq. 4) and for the inference-time open-set score (Eq. 8). This is ordinary supervised training with a learned prototype representation, not a case where a fitted parameter is later 'predicted' against its own fit. The pseudo-OOD samples are synthesized from mixtures of current and pseudo embeddings (Eq. 5), but the reported OSCR and AUC-ROC are computed on held-out test nodes using threshold-agnostic metrics, so there is no evaluation loop that feeds test outcomes back into training. The citations used for design choices are external (e.g., Deep SVDD [31,32] for hypersphere losses, graph prompt tuning [39] for the prompt-style encoder), and no uniqueness theorem or author self-citation is invoked to force the framework. The representation-space mismatch noted between Eq. (6), which uses h(z)=mu_phi(z), and Eq. (8), which uses raw z, is a consistency/correctness concern rather than circularity: it would invalidate the metric-space interpretation of the score, but it does not make the reported result equal to its input by construction. No fitted value is renamed as a prediction and no derivation reduces to its own assumptions, so the circularity score is 0.

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

The method rests on several domain assumptions that are not independently validated: embeddings preserve enough graph structure for replay; linear mixing of known-class embeddings approximates the geometry of true unknown classes; the unit-variance Gaussian prior matches class-conditional embedding densities; knowledge distillation suffices to prevent representation drift; and the evaluation protocol's unknown classes become known later. The hyperparameters (lambda_kd, lambda_reconst, beta, sample counts) are tuned per dataset and affect the reported margins. No new physical or ontological entities are introduced; pseudo-samples are synthetic data, not entities.

free parameters (5)
  • lambda_kd (knowledge distillation weight) = 1 for CS, 100 for other datasets
    Section 6.4: chosen per dataset based on validation; controls how strongly old-task GNN embeddings are preserved, directly affecting replay quality and the reported OSCR margins.
  • lambda_reconst (CVAE reconstruction weight) = 10
    Section 6.4: balances reconstruction and classification; Figure 4e shows excessively high values overemphasize reconstruction and low values degrade pseudo embedding quality.
  • beta (Beta distribution sharpness for OOD mixing) = 5
    Section 6.4: controls how close mixed OOD samples are to the two source embeddings; Figure 4c shows OSCR is stable across beta from 2 to 7.
  • number of pseudo ID samples per class = 300
    Section 6.4 and Algorithm 1: generated at the beginning of each task; too few gives inadequate replay, too many adds noise (Figure 4a).
  • number of pseudo OOD samples per task = 100, refreshed every 20 epochs
    Section 6.4 and Algorithm 1: OOD generation interval and sample count trade coverage versus noise (Figure 4b).
assumptions (5)
  • domain assumption GNN embeddings z capture the essential graph structure needed for classification and open-set detection
    Section 3.3 states the method generates pseudo-samples in embedding space, 'assuming embeddings effectively capture essential graph structures.' If embeddings lose structural information, replay and OOD generation fail.
  • ad hoc to paper Linear mixing of two known-class embeddings yields OOD samples representative of real unseen classes
    Section 3.4.1: mixed embeddings are assumed 'unlikely to align with any single class prototype'; no evidence is provided that true unknown classes lie near the interpolation path.
  • domain assumption The class-conditional distribution of encoded embeddings is well approximated by N(p_c, I)
    Section 3.3.1: the CVAE uses a unit-variance Gaussian prior centered at each prototype; if the true per-class covariance differs, pseudo-samples will be miscalibrated.
  • domain assumption Knowledge distillation (Eq. 3) is sufficient to prevent representation drift across tasks
    Section 3.3.2: KD is the only mechanism stabilizing the GNN encoder; the ablation shows removing it degrades OSCR, so the claim depends on KD being strong enough.
  • domain assumption Unknown classes at task t become known at task t+1
    Section 3.1: the evaluation protocol converts previous unknowns into future known classes, which is an artificial construction; real-world open-set data may never become labeled.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Effective Open-set Graph Class-incremental Learning." pith.science (2026). https://pith.science/paper/ZFHV57XE

@misc{pith2026250717687,
  author       = {Pith},
  title        = {Pith review of: Towards Effective Open-set Graph Class-incremental Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZFHV57XE}},
  note         = {Machine review of arXiv:2507.17687}
}
read the original abstract

Graph class-incremental learning (GCIL) allows graph neural networks (GNNs) to adapt to evolving graph analytical tasks by incrementally learning new class knowledge while retaining knowledge of old classes. Existing GCIL methods primarily focus on a closed-set assumption, where all test samples are presumed to belong to previously known classes. Such an assumption restricts their applicability in real-world scenarios, where unknown classes naturally emerge during inference, and are absent during training. In this paper, we explore a more challenging open-set graph class-incremental learning scenario with two intertwined challenges: catastrophic forgetting of old classes, which impairs the detection of unknown classes, and inadequate open-set recognition, which destabilizes the retention of learned knowledge. To address the above problems, a novel OGCIL framework is proposed, which utilizes pseudo-sample embedding generation to effectively mitigate catastrophic forgetting and enable robust detection of unknown classes. To be specific, a prototypical conditional variational autoencoder is designed to synthesize node embeddings for old classes, enabling knowledge replay without storing raw graph data. To handle unknown classes, we employ a mixing-based strategy to generate out-of-distribution (OOD) samples from pseudo in-distribution and current node embeddings. A novel prototypical hypersphere classification loss is further proposed, which anchors in-distribution embeddings to their respective class prototypes, while repelling OOD embeddings away. Instead of assigning all unknown samples into one cluster, our proposed objective function explicitly models them as outliers through prototype-aware rejection regions, ensuring a robust open-set recognition. Extensive experiments on five benchmarks demonstrate the effectiveness of OGCIL over existing GCIL and open-set GNN methods.

Figures

Figures reproduced from arXiv: 2507.17687 by the authors.

Figure 1
Figure 1. A diagram illustrating the proposed OGCIL framewo [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Average performance comparison of each dataset ov [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Ablation study in terms of average open-set classi [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Performance comparison in terms of average open-s [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: t-SNE embedding of CS dataset over the last task. Da [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 49 canonical work pages

  1. [1]

    Abhijit Bendale and Terrance E Boult. 2016. Towards open set deep networks. In Proceedings of the IEEE Conference on Computer Vision and Patt ern Recognition. 1563–1572

    work page 2016

  2. [2]

    Aleksandar Bojchevski and Stephan Günnemann. 2018. Dee p Gaussian Embed- ding of Graphs: Unsupervised Inductive Learning via Rankin g. In Proceedings of the International Conference on Learning Representations

    work page 2018

  3. [3]

    Chong Chen, Fei Sun, Min Zhang, and Bolin Ding. 2022. Reco mmendation un- learning. In Proceedings of the ACM Web Conference . 2768–2777

    work page 2022

  4. [4]

    Guangyao Chen, Peixi Peng, Xiangqian Wang, and Yonghong Tian. 2021. Ad- versarial reciprocal points learning for open set recognit ion. IEEE Transactions on Pattern Analysis and Machine Intelligence 44, 11 (2021), 8065–8081

    work page 2021

  5. [5]

    Guangyao Chen, Limeng Qiao, Yemin Shi, Peixi Peng, Jia Li , Tiejun Huang, Shil- iang Pu, and Yonghong Tian. 2020. Learning open set network w ith discrimi- native reciprocal points. In Proceedings of the European Conference on Computer Vision. 507–522

    work page 2020

  6. [6]

    Akshay Raj Dhamija, Manuel Günther, and Terrance Boult. 2018. Reducing network agnostophobia. In Advances in Neural Information Processing Systems , Vol. 31

    work page 2018

  7. [7]

    Taoran Fang, Yunchao Zhang, Yang Yang, Chunping Wang, an d Lei Chen. 2023. Universal prompt tuning for graph neural networks. In Advances in Neural In- formation Processing Systems

    work page 2023

  8. [8]

    Falih Gozi Febrinanto, Feng Xia, Kristen Moore, Chandra Thapa, and Charu Ag- garwal. 2023. Graph lifelong learning: A survey. IEEE Computational Intelligence Magazine 18, 1 (2023), 32–51

    work page 2023

Show all 54 references
  1. [9]

    Lukas Galke, Benedikt Franke, Tobias Zielke, and Ansgar Scherp. 2021. Life- long learning of graph neural networks for open-world node c lassification. In Proceedings of the International Joint Conference on Neural Networks. 1–8

    2021

  2. [10]

    Zongyuan Ge, Sergey Demyanov, Zetao Chen, and Rahil Gar navi. 2017. Genera- tive OpenMax for multi-class open set classification. In Proceedings of the British Machine Vision Conference

    2017

  3. [11]

    Qiang He, Hui Fang, Jie Zhang, and Xingwei Wang. 2021. Dy namic opinion maximization in social networks. IEEE Transactions on Knowledge and Data En- gineering 35, 1 (2021), 350–361

    2021

  4. [12]

    Marcel Hoffmann, Lukas Galke, and Ansgar Scherp. 2023. O pen-World Lifelong Graph Learning. In Proceedings of the International Joint Conference on Neural Networks. 1–9

    2023

  5. [13]

    Tiancheng HUANG, Donglin W ANG, and Yuan FANG. 2022. End -to-end open- set semi-supervised node classification with out-of-distribution detection.(2022). In Proceedings of the International Joint Conference on Artific ial Intelligence. 23– 29

    2022

  6. [14]

    Kipf and Max Welling

    Thomas N. Kipf and Max Welling. 2017. Semi-Supervised C lassification with Graph Convolutional Networks. In Proceedings of the International Conference on Learning Representations

    2017

  7. [15]

    James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, J oel Veness, Guillaume Desjardins, Andrei A Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, et al. 2017. Overcoming catastrophic f orgetting in neural networks. Proceedings of the National Academy ...

    2017

  8. [16]

    Xiaoyu Kou, Yankai Lin, Shaobo Liu, Peng Li, Jie Zhou, an d Yan Zhang. 2020. Disentangle-based Continual Graph Representation Learni ng. In Proceedings of the Conference on Empirical Methods in Natural Language Proc essing

    2020

  9. [17]

    Yongqi Li, Meng Liu, Jianhua Yin, Chaoran Cui, Xin-Shun Xu, and Liqiang Nie

  10. [18]

    Shiyu Liang, Yixuan Li, and R Srikant. 2018. Enhancing t he reliability of out-of- distribution image detection in neural networks. In Proceedings of the Interna- tional Conference on Learning Representations

    2018

  11. [19]

    Huihui Liu, Yiding Yang, and Xinchao Wang. 2021. Overco ming catastrophic forgetting in graph neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 35. 8653–8661

    2021

  12. [20]

    Zemin Liu, Xingtong Yu, Yuan Fang, and Xinming Zhang. 20 23. Graphprompt: Unifying pre-training and downstream tasks for graph neura l networks. In Pro- ceedings of the ACM Web Conference . 417–428

  13. [21]

    Bin Lu, Xiaoying Gan, Lina Yang, Weinan Zhang, Luoyi Fu, and Xinbing Wang

  14. [22]

    David Macêdo and Teresa Bernarda Ludermir. 2021. Impro ving entropic out- of-distribution detection using isometric distances and t he minimum distance score. arXiv preprint arXiv:2105.14399 (2021)

    2021 arXiv

  15. [23]

    David Macêdo, Tsang Ing Ren, Cleber Zanchettin, Adrian o LI Oliveira, and Teresa Ludermir. 2022. Entropic out-of-distribution dete ction: Seamless detec- tion of unknown examples. IEEE Transactions on Neural Networks and Learning Systems 33, 6 (2022), 2350–2364

    2022

  16. [24]

    Atefeh Mahdavi and Marco Carvalho. 2021. A survey on ope n set recognition. In Proceedings of the IEEE International Conference on Artifici al Intelligence and Knowledge Engineering. 37–44

    2021

  17. [25]

    Chaoxi Niu, Guansong Pang, Ling Chen, and Bing Liu. 2024 . Replay-and-Forget- Free Graph Class-Incremental Learning: A Task Profiling and Prompting Ap- proach. In Advances in Neural Information Processing Systems

    2024

  18. [26]

    Jinhui Pang, Changqing Lin, Xiaoshuai Hao, Rong Yin, Zi xuan Wang, Zhihui Zhang, Jinglin He, and Huang Tai Sheng. 2024. FTF-ER: Feature-topology fusion- based experience replay method for continual graph learnin g. In Proceedings of the ACM International Conference on Multimed...

    2024

  19. [27]

    Massimo Perini, Giorgia Ramponi, Paris Carbone, and Va siliki Kalavri. 2022. Learning on streaming graphs with experience replay. In Proceedings of the ACM/SIGAPP Symposium on Applied Computing . 470–478

    2022

  20. [28]

    Daiqing Qi, Handong Zhao, Xiaowei Jia, and Sheng Li. 202 4. Revealing an Over- looked Challenge in Class-Incremental Graph Learning. Transactions on Ma- chine Learning Research (2024)

    2024

  21. [29]

    Jiaqian Ren, Lei Jiang, Hao Peng, Yuwei Cao, Jia Wu, Phil ip S Yu, and Lifang He

  22. [30]

    Yixin Ren, Li Ke, Dong Li, Hui Xue, Zhao Li, and Shuigeng Z hou. 2023. Incre- mental graph classification by class prototype constructio n and augmentation. In Proceedings of the ACM International Conference on Informat ion and Knowl- edge Management. 2136–2145

    2023

  23. [31]

    Lukas Ruff, Robert Vandermeulen, Nico Goernitz, Lucas D eecke, Shoaib Ahmed Siddiqui, Alexander Binder, Emmanuel Müller, and Marius Kl oft. 2018. Deep Towards Effective Open-set Graph Class-incremental Learni ng MM ’25, October 27–31, 2025, Dublin, Ireland. one-class classificati...

    2018

  24. [32]

    In Proceedings of the ACM International Conference on Information and Knowledge Management

    From known to unknown: Quality-aware self-improving graph neural net- work for open set social event detection. In Proceedings of the ACM International Conference on Information and Knowledge Management . 1696–1705

  25. [33]

    Oleksandr Shchur, Maximilian Mumme, Aleksandar Bojch evski, and Stephan Günnemann. 2018. Pitfalls of graph neural network evaluati on. arXiv preprint arXiv:1811.05868 (2018)

    2018 arXiv

  26. [34]

    Jake Snell, Kevin Swersky, and Richard Zemel. 2017. Pro totypical networks for few-shot learning. In Advances in Neural Information Processing Systems, Vol. 30

    2017

  27. [35]

    Lukas Ruff, Robert A Vandermeulen, Billy Joe Franks, Kla us-Robert Müller, and Marius Kloft. 2021. Rethinking Assumptions in Deep Anomaly Detection. In Proceedings of the International Conference on Machine Lear ning Workshop

    2021

  28. [36]

    Junwei Su, Difan Zou, Zijun Zhang, and Chuan Wu. 2023. To wards robust graph incremental learning on evolving graphs. In Proceedings of the International Con- ference on Machine Learning . 32728–32748

    2023

  29. [37]

    Xiangguo Sun, Hong Cheng, Jia Li, Bo Liu, and Jihong Guan . 2023. All in one: Multi-task prompting for graph neural networks. In Proceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 2120–2131

    2023

  30. [38]

    Kihyuk Sohn, Honglak Lee, and Xinchen Yan. 2015. Learni ng structured output representation using deep conditional generative models. In Advances in Neural Information Processing Systems , Vol. 28

    2015

  31. [39]

    Xiangguo Sun, Jiawen Zhang, Xixi Wu, Hong Cheng, Yun Xio ng, and Jia Li. 2023. Graph prompt learning: A comprehensive survey and beyond. arXiv preprint arXiv:2311.16534 (2023)

    2023 arXiv

  32. [40]

    Zonggui Tian, Du Zhang, and Hong-Ning Dai. 2024. Contin ual Learning on Graphs: A Survey. arXiv preprint arXiv:2402.06330 (2024)

    2024 arXiv

  33. [41]

    Xin Sun, Zhenning Yang, Chi Zhang, Keck-Voon Ling, and G uohao Peng. 2020. Conditional gaussian distribution learning for open set re cognition. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Patter n Recognition . 13480–13489

    2020

  34. [42]

    Kuansan Wang, Zhihong Shen, Chiyuan Huang, Chieh-Han W u, Yuxiao Dong, and Anshul Kanakia. 2020. Microsoft academic graph: When ex perts are not enough. Quantitative Science Studies 1, 1 (2020), 396–413

    2020

  35. [43]

    Xin Wang, Benyuan Meng, Hong Chen, Yuan Meng, Ke Lv, and W enwu Zhu

  36. [44]

    Junshan Wang, Wenhao Zhu, Guojie Song, and Liang Wang. 2 022. Stream- ing graph neural networks with generative replay. In Proceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 1878–1888

  37. [45]

    Man Wu, Shirui Pan, and Xingquan Zhu. 2020. Openwgl: Ope n-world graph learning. In Proceedings of the IEEE International Conference on Data Mining. 681– 690

    2020

  38. [46]

    Soh Yoshida, Takahiro Ogawa, and Miki Haseyama. 2015. H eterogeneous Graph- based Video Search Reranking using Web Knowledge via Social Media Network. In Proceedings of the 23rd ACM international conference on Multimedia. 871–874

    2015

  39. [47]

    Ryota Yoshihashi, Wen Shao, Rei Kawakami, Shaodi You, M akoto Iida, and Takeshi Naemura. 2019. Classification-reconstruction lea rning for open-set recognition. In Proceedings of the IEEE/CVF Conference on Computer Vision an d Pattern Recognition. 4016–4025

    2019

  40. [48]

    Yanling Wang, Jing Zhang, Lingxi Zhang, Lixin Liu, Yuxi ao Dong, Cuiping Li, Hong Chen, and Hongzhi Yin. 2024. Open-World Semi-Supervis ed Learning for Node Classification. In Proceedings of the IEEE International Conference on Data Engineering

    2024

  41. [49]

    Xikun Zhang, Dongjin Song, and Dacheng Tao. 2022. Hiera rchical prototype net- works for continual graph representation learning. IEEE Transactions on Pattern Analysis and Machine Intelligence 45, 4 (2022), 4622–4636

    2022

  42. [50]

    Fan Zhou and Chengtai Cao. 2021. Overcoming catastroph ic forgetting in graph neural networks with experience replay. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 35. 4714–4722

    2021

  43. [52]

    Qin Zhang, Zelin Shi, Xiaolin Zhang, Xiaojun Chen, Phil ippe Fournier-Viger, and Shirui Pan. 2023. G2Pxy: generative open-set node class ification on graphs with proxy unknowns. In Proceedings of the International Joint Conference on Artificial Intelligence. 4576–4583

    2023

  44. [2019]

    In Proceedings of the ACM International Conference on Multimed ia

    Routing micro-videos via a temporal graph-guided rec ommendation sys- tem. In Proceedings of the ACM International Conference on Multimed ia. 1464– 1472

  45. [2022]

    In Proceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    Geometer: Graph few-shot class-incremental learning via prototype repre- sentation. In Proceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 1152–1161

  46. [2023]

    In Proceedings of the ACM International Conference on Multimed ia

    TIV A-KG: A multimodal knowledge graph with text, imag e, video and audio. In Proceedings of the ACM International Conference on Multimed ia. 2391– 2399

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.