Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Dynamic Graphs via Tensorized and Lightweight Graph Convolutional Networks

Minglian Han

TL;DR

This paper addresses the challenge of learning representations on dynamic graphs by proposing TLGCN, a tensorized, memory-efficient graph convolutional network. It introduces a Spatio-Temporal Information Propagation (STIP) module based on the tensor M-product to jointly propagate spatio-temporal information, and a Tensorized Lightweight Graph Convolution (TLGC) module that omits feature transformation and nonlinear activation for efficiency. The approach leverages a banded lower-triangular transformation matrix $M$ to control temporal influence and achieves superior weight estimation performance on four real-world dynamic graphs, with ablation studies confirming the importance of STIP and the lightweight design. Overall, TLGCN demonstrates improved accuracy and reduced memory footprint, offering practical benefits for real-time dynamic graph analysis.

Abstract

A dynamic graph (DG) is frequently encountered in numerous real-world scenarios. Consequently, A dynamic graph convolutional network (DGCN) has been successfully applied to perform precise representation learning on a DG. However, conventional DGCNs typically consist of a static GCN coupled with a sequence neural network (SNN) to model spatial and temporal patterns separately. This decoupled modeling mechanism inherently disrupts the intricate spatio-temporal dependencies. To address the issue, this study proposes a novel Tensorized Lightweight Graph Convolutional Network (TLGCN) for accurate dynamic graph learning. It mainly contains the following two key concepts: a) designing a novel spatio-temporal information propagation method for joint propagation of spatio-temporal information based on the tensor M-product framework; b) proposing a tensorized lightweight graph convolutional network based on the above method, which significantly reduces the memory occupation of the model by omitting complex feature transformation and nonlinear activation. Numerical experiments on four real-world datasets demonstrate that the proposed TLGCN outperforms the state-of-the-art models in the weight estimation task on DGs.

Learning Dynamic Graphs via Tensorized and Lightweight Graph Convolutional Networks

TL;DR

This paper addresses the challenge of learning representations on dynamic graphs by proposing TLGCN, a tensorized, memory-efficient graph convolutional network. It introduces a Spatio-Temporal Information Propagation (STIP) module based on the tensor M-product to jointly propagate spatio-temporal information, and a Tensorized Lightweight Graph Convolution (TLGC) module that omits feature transformation and nonlinear activation for efficiency. The approach leverages a banded lower-triangular transformation matrix to control temporal influence and achieves superior weight estimation performance on four real-world dynamic graphs, with ablation studies confirming the importance of STIP and the lightweight design. Overall, TLGCN demonstrates improved accuracy and reduced memory footprint, offering practical benefits for real-time dynamic graph analysis.

Abstract

A dynamic graph (DG) is frequently encountered in numerous real-world scenarios. Consequently, A dynamic graph convolutional network (DGCN) has been successfully applied to perform precise representation learning on a DG. However, conventional DGCNs typically consist of a static GCN coupled with a sequence neural network (SNN) to model spatial and temporal patterns separately. This decoupled modeling mechanism inherently disrupts the intricate spatio-temporal dependencies. To address the issue, this study proposes a novel Tensorized Lightweight Graph Convolutional Network (TLGCN) for accurate dynamic graph learning. It mainly contains the following two key concepts: a) designing a novel spatio-temporal information propagation method for joint propagation of spatio-temporal information based on the tensor M-product framework; b) proposing a tensorized lightweight graph convolutional network based on the above method, which significantly reduces the memory occupation of the model by omitting complex feature transformation and nonlinear activation. Numerical experiments on four real-world datasets demonstrate that the proposed TLGCN outperforms the state-of-the-art models in the weight estimation task on DGs.

Paper Structure

This paper contains 21 sections, 19 equations, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Static Graph Neural Networks
  4. Dynamic Graph Neural Networks
  5. Preliminaries
  6. Problem Formulation
  7. Tensor M-Product Framework
  8. Graph Convolutional Network
  9. Methodology
  10. Spatio-Temporal Information Propagation Module
  11. Tensorized Lightweight Graph Convolution Module
  12. Weight Estimation Module
  13. Experiments
  14. Evaluation metrics
  15. Datasets
  16. ...and 6 more sections

Figures (7)

  • Figure 1: Third-order tensor form of a dynamic graph.
  • Figure 2: The overall framework of the proposed TLGCN.
  • Figure 3: The temporal information propagation.
  • Figure 4: The spatial information propagation.
  • Figure 5: Ablation study.
  • ...and 2 more figures