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

LanczosNet: Multi-Scale Deep Graph Convolutional Networks

About

We propose the Lanczos network (LanczosNet), which uses the Lanczos algorithm to construct low rank approximations of the graph Laplacian for graph convolution. Relying on the tridiagonal decomposition of the Lanczos algorithm, we not only efficiently exploit multi-scale information via fast approximated computation of matrix power but also design learnable spectral filters. Being fully differentiable, LanczosNet facilitates both graph kernel learning as well as learning node embeddings. We show the connection between our LanczosNet and graph based manifold learning methods, especially the diffusion maps. We benchmark our model against several recent deep graph networks on citation networks and QM8 quantum chemistry dataset. Experimental results show that our model achieves the state-of-the-art performance in most tasks. Code is released at: \url{https://github.com/lrjconan/LanczosNetwork}.

Renjie Liao, Zhizhen Zhao, Raquel Urtasun, Richard S. Zemel• 2019

Related benchmarks

TaskDatasetResultRank
Node ClassificationCora (test)
Mean Accuracy81.9
861
Node ClassificationCiteseer (test)
Accuracy0.706
824
Node ClassificationPubmed
Accuracy78.3
819
Node ClassificationPubMed (test)
Accuracy78.3
546
Node ClassificationCora standard (test)
Accuracy80.4
130
Node ClassificationCiteseer standard (test)
Accuracy68.7
121
Node ClassificationPubmed standard (test)
Accuracy78.3
92
Node ClassificationCora 3% label rate
Accuracy77.7
66
Node ClassificationCora 1% label rate
Accuracy67.5
56
Node ClassificationCora (0.5% label rate)
Accuracy0.608
56
Showing 10 of 24 rows

Other info

Code

Follow for update