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

Node Dependent Local Smoothing for Scalable Graph Learning

About

Recent works reveal that feature or label smoothing lies at the core of Graph Neural Networks (GNNs). Concretely, they show feature smoothing combined with simple linear regression achieves comparable performance with the carefully designed GNNs, and a simple MLP model with label smoothing of its prediction can outperform the vanilla GCN. Though an interesting finding, smoothing has not been well understood, especially regarding how to control the extent of smoothness. Intuitively, too small or too large smoothing iterations may cause under-smoothing or over-smoothing and can lead to sub-optimal performance. Moreover, the extent of smoothness is node-specific, depending on its degree and local structure. To this end, we propose a novel algorithm called node-dependent local smoothing (NDLS), which aims to control the smoothness of every node by setting a node-specific smoothing iteration. Specifically, NDLS computes influence scores based on the adjacency matrix and selects the iteration number by setting a threshold on the scores. Once selected, the iteration number can be applied to both feature smoothing and label smoothing. Experimental results demonstrate that NDLS enjoys high accuracy -- state-of-the-art performance on node classifications tasks, flexibility -- can be incorporated with any models, scalability and efficiency -- can support large scale graphs with fast training.

Wentao Zhang, Mingyu Yang, Zeang Sheng, Yang Li, Wen Ouyang, Yangyu Tao, Zhi Yang, Bin Cui• 2021

Related benchmarks

TaskDatasetResultRank
Node ClassificationChameleon
Accuracy41.9
867
Node ClassificationWisconsin
Accuracy58
864
Node ClassificationCornell
Accuracy55
851
Node ClassificationTexas
Accuracy0.63
801
Node ClassificationSquirrel
Accuracy24.6
786
Node ClassificationPubmed
Accuracy79.2
627
Node Classificationogbn-arxiv (test)
Accuracy72.24
497
Node ClassificationReddit (test)--
201
Node ClusteringCora
NMI53.7
168
Node ClusteringCiteseer
NMI41.7
140
Showing 10 of 27 rows

Other info

Follow for update