{"total":25,"items":[{"citing_arxiv_id":"2605.21247","ref_index":62,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Graph Navier Stokes Networks","primary_cat":"cs.LG","submitted_at":"2026-05-20T14:36:52+00:00","verdict":null,"verdict_confidence":null,"novelty_score":null,"formal_verification":null,"one_line_summary":null,"context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.20248","ref_index":38,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification","primary_cat":"cs.LG","submitted_at":"2026-05-18T06:47:41+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Transductive Sharpening adds an entropy-minimization term on unlabeled-node predictions to the training objective for graph node classification.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.17854","ref_index":24,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Learning over Positive and Negative Edges with Contrastive Message Passing","primary_cat":"cs.LG","submitted_at":"2026-05-18T04:52:07+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Contrastive Message Passing lets GNNs apply similarity-preserving transforms to positive edges and dissimilarity-inducing transforms to negative edges via soft positive semidefinite constraints on weights, yielding gains in low-label high-homophily regimes.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.13597","ref_index":33,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Rethinking Generalization in Graph Neural Networks: A Structural Complexity Perspective","primary_cat":"cs.LG","submitted_at":"2026-05-13T14:32:46+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"GNN generalization depends explicitly on graph structural complexity measured by effective edges, with a new regularization method shown to balance underfitting and overfitting.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.12061","ref_index":108,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"SAGE: A Self-Evolving Agentic Graph-Memory Engine for Structure-Aware Associative Memory","primary_cat":"cs.AI","submitted_at":"2026-05-12T12:47:43+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"SAGE is a self-evolving agentic graph-memory engine that dynamically constructs and refines structured memory graphs via writer-reader feedback, yielding performance gains on multi-hop QA, open-domain retrieval, and long-term agent benchmarks.","context_count":1,"top_context_role":"background","top_context_polarity":"unclear","context_text":"8 gives h(0) v −h ′(0) v 2 ≤ ηϵη−1 p T0 ∥uq∥2 ∆seed +∥W x∥2 ∆X .(106) Taking the maximum overvgives the conclusion. 29 E.4 Single-layer Stability of Structurally Gated Propagation Lemma E.9(Boundedness and stability of structural gate).The structural gate of layerl, g(l) uv = 1 +δtanh(MLP (l) g (z(l) uv)),(107) satisfies g(l) uv ∞ ≤1 +δ,(108) and g(l) uv −g ′(l) uv ∞ ≤δL g,l z(l) uv −z ′(l) uv 2 .(109) Proof. Since the range of tanh is contained in [−1,1] , the first statement follows immediately. Moreover, because tanh is 1-Lipschitz and MLP(l) g is Lg,l-Lipschitz in the trajectory neighborhood, g(l) uv −g ′(l) uv ∞ ≤δ MLP(l) g (z(l) uv)−MLP (l) g (z′(l) uv )"},{"citing_arxiv_id":"2605.11987","ref_index":24,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Random-Set Graph Neural Networks","primary_cat":"cs.AI","submitted_at":"2026-05-12T11:38:13+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"RS-GNNs predict random sets over classes using belief functions to jointly produce class probabilities and epistemic uncertainty estimates for graph nodes.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"In thevanillamodel, logits are computed via a linear layer and converted to probabilities with softmax: ℓv =W z v, p v(i) = softmax(ℓv)i.(22) InRS-GNN, the head outputs focal-set beliefs via a two-layer MLP with sigmoid outputs over a focal-set budgetF: ˆBelv(Ak) = sigmoid(g(zv))k, A k ∈ F.(23) 10 Masses over focal sets are recovered by a Möbius inversion matrixM(precomputed fromF): ˆmv = ˆBelvM,(24) and pignistic probabilities are computed by multiplying masses with the pignistic matrix P induced byF: BetPv = norm( ˆmvP),(25) where norm(·) denotes row-wise normalisation to ensure a proper probability vector. Final predictions use arg maxi BetPv(i), so RS-GNN differs from vanilla primarily in its uncertainty representation rather than in the decision rule."},{"citing_arxiv_id":"2605.09993","ref_index":21,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Learning Graph Foundation Models on Riemannian Graph-of-Graphs","primary_cat":"cs.LG","submitted_at":"2026-05-11T05:09:16+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"R-GFM constructs multi-scale Riemannian graph-of-graphs to learn geometry-adaptive representations, reducing structural domain generalization error and delivering up to 49% relative gains on downstream graph tasks.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"Accordingly, the fused sampling noise is the random vector ˜ϵv(w) := KX k=1 wk ϵ(k) v .(19) Letσ v(w)denote the coordinate-wise standard-deviation vector of the fused sampling noise: σv(w) := q diag Cov(˜ϵv(w)) \u0001 .(20) The strategy chooses the fusion weights by minimizing the overall noise magnitude measured by theℓ 2-norm: w⋆ ∈arg min w∈∆K σv(w) 2.(21) Definitions of σF and σV.A fixed-hop strategy selects a single hop kfixed and always uses x(kfixed) v . Under Assump- tion G.1, its sampling noise isϵ (kfixed) v . Let σv,k := q diag Cov(ϵ(k) v ) \u0001 = q diag Σv,kk \u0001 ∈R d +.(22) We define σF :=σ v,kfixed ,σ V :=σ v(w⋆).(23) 22 Learning Graph Foundation Models on Riemannian Graph-of-Graphs Proof of Theorem 3."},{"citing_arxiv_id":"2605.09862","ref_index":40,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"UFO: A Unified Flow-Oriented Framework for Robust Continual Graph Learning","primary_cat":"cs.LG","submitted_at":"2026-05-11T01:49:05+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"UFO combines flow-based generative replay with instance-level reliability scoring to handle both catastrophic forgetting and catastrophic remembering from noisy supervision in evolving graphs, outperforming baselines on four datasets.","context_count":1,"top_context_role":"dataset","top_context_polarity":"use_dataset","context_text":"where τ is a temperature parameter, and τ 2 is used to compensate for the 1/τ 2 scaling of the gradi- ents. For the current task, the final objective for graph knowledge preservation is: LKP = αELE + αLLL, (13) where αE and αL are loss weights. The whole training process is summarized in Algorithm 1. 5 Experiments 5.1 Experimental Setup Datasets and Implementation Details. We evaluate our method on four real-world datasets: Cora- Full [ 2], CS [ 40], WikiCS [ 41], and Photo [ 40]. More detailed statistics are provided in Ap- pendix A.1. Following [ 42], we divide the classes into a sequence of tasks for each dataset, and split the nodes in each task into 60%, 20%, and 20% for training, validation, and testing, respec- tively. We add symmetric noise to a fraction of labels in the training set while keeping the validation"},{"citing_arxiv_id":"2605.06814","ref_index":43,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"From Model to Data (M2D): Shifting Complexity from GNNs to Graphs for Transparent Graph Learning","primary_cat":"cs.LG","submitted_at":"2026-05-07T18:16:21+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"M2D distillation augments input graphs with model-derived features and structure, letting simple student GNNs match teacher performance while exposing mechanisms such as attention and fairness directly in the data.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.05534","ref_index":51,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Adversarial Graph Neural Network Benchmarks: Towards Practical and Fair Evaluation","primary_cat":"cs.LG","submitted_at":"2026-05-07T00:27:24+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"A large-scale standardized benchmark of GNN attacks and defenses reveals that target node selection and attacked-model training process can completely distort measured attack effectiveness.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"Hagenbuchner, and G. Monfardini. The graph neural network model.IEEE Trans. Neural Networks, 20(1):61-80, 2009. doi: 10.1109/TNN.2008. 2005605. URLhttps://doi.org/10.1109/TNN.2008.2005605. [50] O. Shchur, M. Mumme, A. Bojchevski, and S. Günnemann. Pitfalls of graph neural network evaluation.CoRR, abs/1811.05868, 2018. URLhttp://arxiv.org/abs/1811.05868. 13 [51] S. Tao, H. Shen, Q. Cao, L. Hou, and X. Cheng. Adversarial immunization for certifi- able robustness on graphs. In L. Lewin-Eytan, D. Carmel, E. Yom-Tov, E. Agichtein, and E. Gabrilovich, editors,WSDM '21, The Fourteenth ACM International Conference on Web Search and Data Mining, Virtual Event, Israel, March 8-12, 2021, pages 698-706. ACM, 2021. doi: 10."},{"citing_arxiv_id":"2604.27462","ref_index":4,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Improving Graph Few-shot Learning with Hyperbolic Space and Denoising Diffusion","primary_cat":"cs.LG","submitted_at":"2026-04-30T06:02:16+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"IMPRESS improves graph few-shot learning by learning representations in hyperbolic space and using denoising diffusion to better approximate target distributions from few support samples.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.23324","ref_index":26,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Layer Embedding Deep Fusion Graph Neural Network","primary_cat":"cs.LG","submitted_at":"2026-04-25T14:25:02+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"LEDF-GNN fuses multi-layer embeddings nonlinearly and runs parallel processing on original and reconstructed topologies to capture long-range dependencies and mitigate heterophily-induced misaggregation in deep GNNs.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.11473","ref_index":17,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Learning How Much to Think: Difficulty-Aware Dynamic MoEs for Graph Node Classification","primary_cat":"cs.LG","submitted_at":"2026-04-13T13:44:27+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"D2MoE dynamically allocates expert resources in graph MoEs via difficulty-driven top-p routing based on predictive entropy, yielding higher accuracy and lower memory/time costs on node classification benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.11257","ref_index":41,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Unified Graph Prompt Learning via Low-Rank Graph Message Prompting","primary_cat":"cs.LG","submitted_at":"2026-04-13T10:07:12+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"LR-GMP unifies graph prompting via a low-rank Graph Message Prompt paradigm to achieve better generalization than component-specific methods.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.08980","ref_index":22,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Neighbourhood Transformer: Switchable Attention for Monophily-Aware Graph Learning","primary_cat":"cs.LG","submitted_at":"2026-04-10T05:35:21+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Neighbourhood Transformers apply local self-attention for monophily-aware graph learning, guarantee expressiveness at least as strong as message-passing GNNs, and outperform prior methods on node classification across ten datasets while cutting memory and time costs substantially.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.06391","ref_index":33,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Toward a universal foundation model for graph-structured data","primary_cat":"cs.LG","submitted_at":"2026-04-07T19:21:21+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"A pretrained graph model using feature-agnostic structural prompts matches or exceeds supervised baselines and shows strong zero-shot and few-shot transfer on held-out biomedical graphs, with a 21.8% ROC-AUC gain on SagePPI.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.02633","ref_index":42,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Analytic Drift Resister for Non-Exemplar Continual Graph Learning","primary_cat":"cs.LG","submitted_at":"2026-04-03T01:58:11+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"ADR achieves theoretically zero-forgetting class-incremental graph learning by combining backpropagation adaptation with ridge-regression-based layer-wise merging of GNN linear transformations.","context_count":1,"top_context_role":"dataset","top_context_polarity":"use_dataset","context_text":"Analogous to HAM, ACR provides privacy guarantees and is formally equivalent to joint learning, theoretically estab- lishing a zero-forgetting paradigm. We present the training pseudocode for the proposed ADR in Algorithm 1. IV. EXPERIMENTS A. Experimental Setup Datasets.We conduct a comprehensive evaluation of the proposed ADR on four established node classification bench- marks: CS-CL [42], CoraFull-CL [43], Arxiv-CL [44], and Reddit-CL [2]. These datasets, spanning diverse scales, are systematically arranged into a chronologically ordered task streamT={T 0,T 1, . . . ,TN−1 }following [13], [45], withT 0 as the base task, containing nearly half of the data, andT 1:N−1 as incremental tasks, each comprising two disjoint classes to emulate the continuous arrival of new tasks."},{"citing_arxiv_id":"2603.01388","ref_index":57,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Invariant-Stratified Propagation for Expressive Graph Neural Networks","primary_cat":"cs.LG","submitted_at":"2026-03-02T02:34:40+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Invariant-Stratified Propagation (ISP) enhances GNN expressivity beyond 1-WL by stratifying nodes according to graph invariants and encoding structural heterogeneity in hierarchical strata.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2602.11629","ref_index":6,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"GP2F: Cross-Domain Graph Prompting with Adaptive Fusion of Pre-trained Graph Neural Networks","primary_cat":"cs.LG","submitted_at":"2026-02-12T06:25:21+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"GP2F is a dual-branch graph prompting framework that fuses frozen pre-trained knowledge with task-specific adaptation to reduce estimation error and outperform baselines in cross-domain few-shot node and graph classification.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2602.00407","ref_index":42,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Fed-Listing: Federated Label Distribution Inference in Graph Neural Networks","primary_cat":"cs.LG","submitted_at":"2026-01-30T23:51:49+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Fed-Listing infers client label proportions in FedGNNs from final-layer gradients, outperforming baselines on four datasets and three architectures even in non-i.i.d. settings.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2512.24062","ref_index":30,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Energy-Balanced Hyperspherical Graph Representation Learning via Structural Binding and Entropic Dispersion","primary_cat":"cs.LG","submitted_at":"2025-12-30T08:11:37+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"HyperGRL places graph nodes on a hypersphere and minimizes Helmholtz free energy with structural binding energy and mean-field repulsive potential, regulated by an adaptive thermostat, to produce discriminative representations.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2511.06443","ref_index":48,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"How Wide and How Deep? Mitigating Over-Squashing of GNNs via Channel Capacity Constrained Estimation","primary_cat":"cs.LG","submitted_at":"2025-11-09T16:25:37+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"C3E estimates hidden dimensions and depths for GNNs by treating them as communication channels to reduce over-squashing and improve representation learning.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2506.08618","ref_index":11,"ref_count":2,"confidence":0.98,"is_internal_anchor":true,"paper_title":"HSG-12M: A Large-Scale Benchmark of Spatial Multigraphs from the Energy Spectra of Non-Hermitian Crystals","primary_cat":"cs.LG","submitted_at":"2025-06-10T09:25:19+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"HSG-12M is a large dataset of spatial multigraphs derived from non-Hermitian crystal energy spectra via the Poly2Graph pipeline, positioned as the first large-scale benchmark of this graph type.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"International Conference on Learning Representations (ICLR), 2019. [9] Federico Errica, Marco Podda, Davide Bacciu, and Alessio Micheli. A fair comparison of graph neural networks for graph classification. arXiv preprint arXiv:1912.09893, 2019. [10] Till Schulz and Pascal Welke. On the necessity of graph kernel baselines. In ECML-PKDD GEM Workshop, 2019. [11] Oleksandr Shchur, Maximilian Mumme, Aleksandar Bojchevski, and Stephan Günnemann. Pitfalls of graph neural network evaluation. arXiv preprint arXiv:1811.05868, 2018. [12] Karsten M. Borgwardt, Cheng Soon Ong, Stefan Schönauer, S.V .N. Vishwanathan, Alexander J. Smola, and Hans-Peter Kriegel. Protein function prediction via graph kernels. Bioinformatics (Oxford, England),"},{"citing_arxiv_id":"2410.15001","ref_index":24,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"FIT-GNN: Faster Inference Time for GNNs that 'FIT' in Memory Using Coarsening","primary_cat":"cs.LG","submitted_at":"2024-10-19T06:27:24+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"FIT-GNN applies graph coarsening during inference to deliver orders-of-magnitude faster single-node inference and lower memory use on node and graph classification/regression tasks while keeping competitive accuracy.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2405.07406","ref_index":151,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Machine Unlearning: A Comprehensive Survey","primary_cat":"cs.CR","submitted_at":"2024-05-13T00:58:34+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":2.0,"formal_verification":"none","one_line_summary":"A survey classifying machine unlearning into centralized (exact and approximate), distributed/irregular data, verification, and privacy/security categories with technique overviews.","context_count":1,"top_context_role":"dataset","top_context_polarity":"use_dataset","context_text":"Covtype 54 581 ,012 Classification [111, 146], ... HIGGS [147] 28 11 ,000,000 Binary classification [111, 148], ... YELP2018 5 1 ,561,406 Recommendation [19] Movielens-1m 4 1 ,000,209 Recommendation [19] Movielens-10m 4 10 ,000,054 Recommendation [19] Text AG News [149] 3 127 ,600 Text classification [23] RCV1 [150] 3 804 ,414 Text classification [23] Graph Amazon Photo [151]Features per Node:745 Nodes:7,650 Edges:119,081 Node classification Link prediction [51, 152], ... Cora [153] Features per Node:1,433 Nodes:2,708 Edges:5,429 Node classification Link prediction [52, 152], ... Citseer [154] Features per Node:3,703 Nodes:3,327 Edges:4,732 Node classification Link prediction [52] Pubmed [155] Features per Node:500 Nodes:19,717"}],"limit":50,"offset":0}