Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Gated Graph Attention Networks with Learnable Temperature

About

Graph attention networks learn neighbor importance through data-dependent coefficients, but standard layers lack explicit control over unreliable feature dimensions and use fixed sharpness of attention coefficient distributions. This paper proposes gated graph attention and learnable temperature for common graph attention mechanisms. Gated graph attention filters feature or message responses to reduce the influence of unreliable dimensions, while learnable temperature dynamically adjusts the sharpness of the attention coefficient distribution. Experiments on homogeneous and heterophilic heterogeneous benchmarks show that the proposed variants consistently improve the corresponding graph attention backbones, and controlled noise studies further verify their behavior under feature perturbations. Theoretical analysis explains these results by showing that gating improves robustness when only part of the feature coordinates are reliable, while temperature is beneficial when global noise weakens the discriminability of node features.

Zhongtian Ma, Hao Wu, Yexin Zhang, Qiaosheng Zhang, Zhen Wang• 2026

Related benchmarks

TaskDatasetResultRank
Node ClassificationCiteseer (test)
Accuracy0.6943
945
Node ClassificationPubMed (test)
Accuracy78.6
586
Node ClassificationReddit (test)--
201
Node Classificationogbn-products (test)
Test Accuracy80.62
162
Node ClassificationCora (test)
Accuracy81.45
122
Node ClassificationOGBN-ArXiv standard (test)
Accuracy72.04
39
Node ClassificationMag-year H2GB (test)
Accuracy38.66
11
Node ClassificationPokec H2GB (test)
Accuracy76.23
11
Node ClassificationOAG-Chem H2GB (test)
Accuracy17.98
11
Node ClassificationOGBN-MAG H2GB (test)
Accuracy51.99
11
Showing 10 of 11 rows

Other info

Follow for update