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

Rethinking Graph Transformers with Spectral Attention

About

In recent years, the Transformer architecture has proven to be very successful in sequence processing, but its application to other data structures, such as graphs, has remained limited due to the difficulty of properly defining positions. Here, we present the $\textit{Spectral Attention Network}$ (SAN), which uses a learned positional encoding (LPE) that can take advantage of the full Laplacian spectrum to learn the position of each node in a given graph. This LPE is then added to the node features of the graph and passed to a fully-connected Transformer. By leveraging the full spectrum of the Laplacian, our model is theoretically powerful in distinguishing graphs, and can better detect similar sub-structures from their resonance. Further, by fully connecting the graph, the Transformer does not suffer from over-squashing, an information bottleneck of most GNNs, and enables better modeling of physical phenomenons such as heat transfer and electric interaction. When tested empirically on a set of 4 standard datasets, our model performs on par or better than state-of-the-art GNNs, and outperforms any attention-based model by a wide margin, becoming the first fully-connected architecture to perform well on graph benchmarks.

Devin Kreuzer, Dominique Beaini, William L. Hamilton, Vincent L\'etourneau, Prudencio Tossou• 2021

Related benchmarks

TaskDatasetResultRank
Graph ClassificationPROTEINS
Accuracy74.11
1383
Graph ClassificationMUTAG
Accuracy78.8
1229
Node ClassificationCora
Accuracy84.81
1225
Node ClassificationCiteseer
Accuracy73.99
1037
Node ClassificationCiteseer (test)
Accuracy0.7337
1013
Node ClassificationChameleon
Accuracy64.02
936
Node ClassificationPubmed
Accuracy88.22
902
Node ClassificationCornell
Accuracy79.62
900
Node ClassificationWisconsin
Accuracy82.66
898
Node ClassificationTexas
Accuracy0.8518
859
Showing 10 of 158 rows
...

Other info

Code

Follow for update