Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Graph Neural Alchemist: An innovative fully modular architecture for time series-to-graph classification

Paulo Coelho, Raul Araju, Luís Ramos, Samir Saliba, Renato Vimieiro

TL;DR

This paper introduces a novel Graph Neural Network architecture for time series classification, based on visibility graph representations, offering a powerful and flexible framework for future research and practical implementations.

Abstract

This paper introduces a novel Graph Neural Network (GNN) architecture for time series classification, based on visibility graph representations. Traditional time series classification methods often struggle with high computational complexity and inadequate capture of spatio-temporal dynamics. By representing time series as visibility graphs, it is possible to encode both spatial and temporal dependencies inherent to time series data, while being computationally efficient. Our architecture is fully modular, enabling flexible experimentation with different models and representations. We employ directed visibility graphs encoded with in-degree and PageRank features to improve the representation of time series, ensuring efficient computation while enhancing the model's ability to capture long-range dependencies in the data. We show the robustness and generalization capability of the proposed architecture across a diverse set of classification tasks and against a traditional model. Our work represents a significant advancement in the application of GNNs for time series analysis, offering a powerful and flexible framework for future research and practical implementations.

Graph Neural Alchemist: An innovative fully modular architecture for time series-to-graph classification

TL;DR

This paper introduces a novel Graph Neural Network architecture for time series classification, based on visibility graph representations, offering a powerful and flexible framework for future research and practical implementations.

Abstract

This paper introduces a novel Graph Neural Network (GNN) architecture for time series classification, based on visibility graph representations. Traditional time series classification methods often struggle with high computational complexity and inadequate capture of spatio-temporal dynamics. By representing time series as visibility graphs, it is possible to encode both spatial and temporal dependencies inherent to time series data, while being computationally efficient. Our architecture is fully modular, enabling flexible experimentation with different models and representations. We employ directed visibility graphs encoded with in-degree and PageRank features to improve the representation of time series, ensuring efficient computation while enhancing the model's ability to capture long-range dependencies in the data. We show the robustness and generalization capability of the proposed architecture across a diverse set of classification tasks and against a traditional model. Our work represents a significant advancement in the application of GNNs for time series analysis, offering a powerful and flexible framework for future research and practical implementations.

Paper Structure

This paper contains 12 sections, 6 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Theoretical Background
  4. Visibility Graph
  5. Graph Neural Networks
  6. The Graph Neural Alchemist architecture
  7. Graph Representation Module
  8. Graph Neural Network (GNN) Module
  9. Readout Layer
  10. Multilayer Pooling Perceptron (MLPP) Module
  11. Experimental Results
  12. Conclusion and future works

Figures (2)

  • Figure 1: The proposed GNA: a time series is fed to the architecture and each box on the right represents a fully customizable module of the architecture. Here we demonstrate the use of VG to represent the time series, followed by a 4-layer GraphSAGE network with a readout layer and a 3-layer modified MLP for classification.
  • Figure 2: Spatial structure for a sample of positive-labeled time series from the Earthquakes dataset. The top left plot shows the time series, the top middle plot shows the Degree Distribution, and the top right plot shows the Degree Distribution on a log-log scale. Note how the degree distribution follows a power-law, and, in the bottom plot, how this translates into a sparse scale-free graph.