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.
hub Mixed citations
Pitfalls of Graph Neural Network Evaluation
Mixed citation behavior. Most common role is background (50%).
25
Pith papers citing it
Background
50% of classified citations
abstract
Semi-supervised node classification in graphs is a fundamental problem in graph mining, and the recently proposed graph neural networks (GNNs) have achieved unparalleled results on this task. Due to their massive success, GNNs have attracted a lot of attention, and many novel architectures have been put forward. In this paper we show that existing evaluation strategies for GNN models have serious shortcomings. We show that using the same train/validation/test splits of the same datasets, as well as making significant changes to the training procedure (e.g. early stopping criteria) precludes a fair comparison of different architectures. We perform a thorough empirical evaluation of four prominent GNN models and show that considering different splits of the data leads to dramatically different rankings of models. Even more importantly, our findings suggest that simpler GNN architectures are able to outperform the more sophisticated ones if the hyperparameters and the training procedure are tuned fairly for all models.
hub tools
citation-role summary
background 5
dataset 3
citation-polarity summary
representative citing papers
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.
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.
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.
Invariant-Stratified Propagation (ISP) enhances GNN expressivity beyond 1-WL by stratifying nodes according to graph invariants and encoding structural heterogeneity in hierarchical strata.
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.
C3E estimates hidden dimensions and depths for GNNs by treating them as communication channels to reduce over-squashing and improve representation learning.
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.
RS-GNNs predict random sets over classes using belief functions to jointly produce class probabilities and epistemic uncertainty estimates for graph nodes.
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.
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.
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.
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.
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.
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.
ADR achieves theoretically zero-forgetting class-incremental graph learning by combining backpropagation adaptation with ridge-regression-based layer-wise merging of GNN linear transformations.
Transductive Sharpening adds an entropy-minimization term on unlabeled-node predictions to the training objective for graph node classification.
GNN generalization depends explicitly on graph structural complexity measured by effective edges, with a new regularization method shown to balance underfitting and overfitting.
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.
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.
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.
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.
LR-GMP unifies graph prompting via a low-rank Graph Message Prompt paradigm to achieve better generalization than component-specific methods.
A survey classifying machine unlearning into centralized (exact and approximate), distributed/irregular data, verification, and privacy/security categories with technique overviews.
citing papers explorer
-
Energy-Balanced Hyperspherical Graph Representation Learning via Structural Binding and Entropic Dispersion
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.
-
HSG-12M: A Large-Scale Benchmark of Spatial Multigraphs from the Energy Spectra of Non-Hermitian Crystals
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.
-
How Wide and How Deep? Mitigating Over-Squashing of GNNs via Channel Capacity Constrained Estimation
C3E estimates hidden dimensions and depths for GNNs by treating them as communication channels to reduce over-squashing and improve representation learning.