LanczosNet: Multi-scale deep graph convo-lutional networks

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

Research output: Contribution to conferencePaper

Abstract

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.

Original languageEnglish (US)
StatePublished - Jan 1 2019
Event7th International Conference on Learning Representations, ICLR 2019 - New Orleans, United States
Duration: May 6 2019May 9 2019

Conference

Conference7th International Conference on Learning Representations, ICLR 2019
CountryUnited States
CityNew Orleans
Period5/6/195/9/19

Fingerprint

Quantum chemistry
Convolution
Decomposition
learning method
learning
chemistry
Graph
performance

ASJC Scopus subject areas

  • Education
  • Computer Science Applications
  • Linguistics and Language
  • Language and Linguistics

Cite this

Liao, R., Zhao, Z., Urtasun, R., & Zemel, R. S. (2019). LanczosNet: Multi-scale deep graph convo-lutional networks. Paper presented at 7th International Conference on Learning Representations, ICLR 2019, New Orleans, United States.

LanczosNet : Multi-scale deep graph convo-lutional networks. / Liao, Renjie; Zhao, Zhizhen; Urtasun, Raquel; Zemel, Richard S.

2019. Paper presented at 7th International Conference on Learning Representations, ICLR 2019, New Orleans, United States.

Research output: Contribution to conferencePaper

Liao, R, Zhao, Z, Urtasun, R & Zemel, RS 2019, 'LanczosNet: Multi-scale deep graph convo-lutional networks' Paper presented at 7th International Conference on Learning Representations, ICLR 2019, New Orleans, United States, 5/6/19 - 5/9/19, .
Liao R, Zhao Z, Urtasun R, Zemel RS. LanczosNet: Multi-scale deep graph convo-lutional networks. 2019. Paper presented at 7th International Conference on Learning Representations, ICLR 2019, New Orleans, United States.
Liao, Renjie ; Zhao, Zhizhen ; Urtasun, Raquel ; Zemel, Richard S. / LanczosNet : Multi-scale deep graph convo-lutional networks. Paper presented at 7th International Conference on Learning Representations, ICLR 2019, New Orleans, United States.
@conference{9a719110457040a0ac0bfd40fcb848c7,
title = "LanczosNet: Multi-scale deep graph convo-lutional networks",
abstract = "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.",
author = "Renjie Liao and Zhizhen Zhao and Raquel Urtasun and Zemel, {Richard S.}",
year = "2019",
month = "1",
day = "1",
language = "English (US)",
note = "7th International Conference on Learning Representations, ICLR 2019 ; Conference date: 06-05-2019 Through 09-05-2019",

}

TY - CONF

T1 - LanczosNet

T2 - Multi-scale deep graph convo-lutional networks

AU - Liao, Renjie

AU - Zhao, Zhizhen

AU - Urtasun, Raquel

AU - Zemel, Richard S.

PY - 2019/1/1

Y1 - 2019/1/1

N2 - 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.

AB - 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.

UR - http://www.scopus.com/inward/record.url?scp=85071177476&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85071177476&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:85071177476

ER -