Pith. sign in

REVIEW 19 cited by

Geom-GCN: Geometric Graph Convolutional Networks

Not yet reviewed by Pith; the record is open.

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

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2002.05287 v2 pith:HDZOPA5C submitted 2020-02-13 cs.LG cs.CVstat.ML

Hongbin Pei , Bingzhe Wei , Kevin Chen-Chuan Chang , Yu Lei , Bo Yang This is my paper
classification cs.LGcs.CVstat.ML
keywords graphnetworksaggregationgeom-gcngraphsconvolutionalgeometricneural
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Message-passing neural networks (MPNNs) have been successfully applied to representation learning on graphs in a variety of real-world applications. However, two fundamental weaknesses of MPNNs' aggregators limit their ability to represent graph-structured data: losing the structural information of nodes in neighborhoods and lacking the ability to capture long-range dependencies in disassortative graphs. Few studies have noticed the weaknesses from different perspectives. From the observations on classical neural network and network geometry, we propose a novel geometric aggregation scheme for graph neural networks to overcome the two weaknesses. The behind basic idea is the aggregation on a graph can benefit from a continuous space underlying the graph. The proposed aggregation scheme is permutation-invariant and consists of three modules, node embedding, structural neighborhood, and bi-level aggregation. We also present an implementation of the scheme in graph convolutional networks, termed Geom-GCN (Geometric Graph Convolutional Networks), to perform transductive learning on graphs. Experimental results show the proposed Geom-GCN achieved state-of-the-art performance on a wide range of open datasets of graphs. Code is available at https://github.com/graphdml-uiuc-jlu/geom-gcn.

Discussion (0). Sign in to comment.

Forward citations

Cited by 19 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Delving into Instance-Dependent Label Noise in Graph Data: A Comprehensive Study and Benchmark

    cs.LG 2025-06 conditional novelty 7.0 of 10

    BeGIN provides 10 graph datasets with six label-noise types and a broad evaluation, showing that LLM-simulated noise hurts GNNs more than uniform or pairwise noise.

  2. HeAD-CP: Heterophily-Aware Diffused Conformal Prediction Sets for Graph Neural Networks

    cs.LG 2026-07 conditional novelty 6.0 of 10

    Graph conformal prediction with node-wise softmax-derived diffusion coefficients avoids DAPS's heterophily failure and beats it on 8/10 benchmarks in oracle selection.

  3. NodeImport: Imbalanced Node Classification with Node Importance Assessment

    cs.LG 2026-07 conditional novelty 6.0 of 10

    A closed-form importance score — the gradient alignment between a node and a balanced meta-set — filters labeled, unlabeled, and synthetic nodes, improving GNN balanced accuracy under class imbalance.

  4. Frequency-Corrupt Based Graph Self-Supervised Learning

    cs.LG 2026-04 unverdicted novelty 6.0 of 10

    FC-GSSL is a graph self-supervised method that corrupts nodes/edges with high low-frequency contribution and reconstructs low-frequency/general targets, improving node and graph prediction on most tested benchmarks.

  5. Structural Bias Beyond Homophily: A Study of Fairness in Link Prediction

    cs.LG 2026-02 conditional novelty 6.0 of 10

    Fairness in link prediction is strongly shaped by graph topology, and fairness-aware methods remain sensitive to structural biases beyond homophily.

  6. Sheaves Reloaded: A Directional Awakening

    cs.LG 2025-06 conditional novelty 6.0 of 10

    A directed sheaf Laplacian and the DSNN architecture bring edge-orientation information into sheaf-based graph learning with modest-but-consistent benchmark gains.

  7. Graph Positional Autoencoders as Self-supervised Learners

    cs.LG 2025-05 conditional novelty 6.0 of 10

    A dual-path graph autoencoder that reconstructs node features and Laplacian-eigenvector distances reports strong self-supervised results on heterophilic and molecular benchmarks, with some overstatement in the margins...

  8. Aggregation Buffer: Revisiting DropEdge with a New Parameter Block

    cs.LG 2025-05 conditional novelty 6.0 of 10

    Adding a degree-normalized Aggregation Buffer block to a frozen pretrained GNN and training it with DropEdge improves node-classification accuracy and robustness on 12 benchmarks.

  9. Are Heterogeneous Graph Neural Networks Truly Effective for Node Classification? A Causal Perspective

    cs.LG 2025-10 reject novelty 5.0 of 10

    A large benchmark suggests tuned RGCN matches complex HGNNs and heterogeneous graphs help mainly via homophily and local-global label discrepancy, but the causal analysis is circular.

  10. Dual-Center Graph Clustering with Neighbor Distribution

    cs.LG 2025-07 reject novelty 5.0 of 10

    A dual-center graph clustering method that aligns nodes to both feature centers and neighbor-distribution centers outperforms prior methods in experiments.

  11. OpenGT: A Comprehensive Benchmark For Graph Transformers

    cs.LG 2025-06 conditional novelty 5.0 of 10

    OpenGT benchmarks 16 graph models on 14 datasets, finding graph transformers excel on heterophilous graphs, though several observations are not robustly supported.

  12. The Oversmoothing Fallacy: A Misguided Narrative in GNN Research

    cs.LG 2025-06 conditional novelty 5.0 of 10

    The authors argue GNN oversmoothing is overstated, that zero-collapsing (vanishing gradient) is the actual failure mode, and that residual connections plus batch normalization suffice to train very deep GCNs.

  13. SGS-GNN: A Supervised Graph Sparsification method for Graph Neural Networks

    cs.LG 2025-02 conditional novelty 5.0 of 10

    A supervised edge-sampling sparsifier that keeps only 20% of edges can match or beat the full graph for GNN node classification, with especially large gains on heterophilic graphs.

  14. A Batch-Insensitive Dynamic GNN Approach to Address Temporal Discontinuity in Graph Streams

    cs.LG 2025-06 reject novelty 4.0 of 10

    BADGNN adds a Lipschitz regularizer and an attention-temperature adjustment so that memory-based dynamic GNNs can train with large batches while keeping predictive accuracy.

  15. How do Probabilistic Graphical Models and Graph Neural Networks Look at Network Data?

    stat.ML 2025-06 conditional novelty 4.0 of 10

    SBM-style probabilistic models outperform graph neural networks on link prediction when node features are low-dimensional, noisy, or the graph is heterophilic.

  16. Learn Beneficial Noise as Graph Augmentation

    cs.LG 2025-05 conditional novelty 4.0 of 10

    PiNGDA learns beneficial noise on graph topology and attributes via a trainable generator, framing standard graph contrastive learning as a point estimate of positive-incentive noise.

  17. COMBINEX: A Unified Counterfactual Explainer for Graph Neural Networks via Node Feature and Structural Perturbations

    cs.LG 2025-02 reject novelty 4.0 of 10

    COMBINEX generates counterfactual explanations for GNNs by jointly perturbing node features and edges, but in practice it mostly modifies features and its experimental claims are contradicted by its own appendix.

  18. Unifying Adversarial Perturbation for Graph Neural Networks

    cs.LG 2025-08 reject novelty 3.0 of 10

    Adding perturbations directly to every hidden embedding of a GNN is claimed to subsume existing feature-, edge-, and weight-perturbation defenses, but the claim rests on simplifications that the experiments do not act...

  19. Enhancing Power Flow Estimation with Topology-Aware Gated Graph Neural Networks

    eess.SY 2025-07 reject novelty 3.0 of 10

    A gated graph neural network predicts AC power flow voltages and angles on IEEE 30 to 1354 bus systems with reported gains over GNN baselines, though internal inconsistencies weaken the evidence.

Pith tools