TodyNet: Temporal Dynamic Graph Neural Network for Multivariate Time Series Classification

TL;DR

MTSC requires capturing hidden inter-variable dependencies and temporal evolution in multivariate signals. TodyNet addresses this with a temporal dynamic graph framework that learns per-slot adjacency, applies a dynamic graph neural network (DyGIN), and uses temporal graph pooling and temporal convolution in an end-to-end model. It achieves state-of-the-art results on 26 UEA MTSC benchmarks, with notable gains on EEG-like data and robust performance across diverse datasets. By modeling both cross-variable dependencies and temporal dynamics, this approach offers a principled, scalable method for complex multivariate time-series classification with potential impact across domains requiring reliable MTSC.

Abstract

Multivariate time series classification (MTSC) is an important data mining task, which can be effectively solved by popular deep learning technology. Unfortunately, the existing deep learning-based methods neglect the hidden dependencies in different dimensions and also rarely consider the unique dynamic features of time series, which lack sufficient feature extraction capability to obtain satisfactory classification accuracy. To address this problem, we propose a novel temporal dynamic graph neural network (TodyNet) that can extract hidden spatio-temporal dependencies without undefined graph structure. It enables information flow among isolated but implicit interdependent variables and captures the associations between different time slots by dynamic graph mechanism, which further improves the classification performance of the model. Meanwhile, the hierarchical representations of graphs cannot be learned due to the limitation of GNNs. Thus, we also design a temporal graph pooling layer to obtain a global graph-level representation for graph learning with learnable temporal parameters. The dynamic graph, graph information propagation, and temporal convolution are jointly learned in an end-to-end framework. The experiments on 26 UEA benchmark datasets illustrate that the proposed TodyNet outperforms existing deep learning-based methods in the MTSC tasks.

Figures (8)

• Figure 1: The framework of TodyNet. We first split the input time series into $s$ slices, and generate a dynamic graph for each slice. The dynamic graph neural network modules and temporal convolution modules capture spatial and temporal dependencies separately. Afterward, the temporal graph pooling module clusters nodes together with learnable temporal parameters at each layer. The output layer processes concatenate hidden features for final classification results.
• Figure 2: Dynamic Graph Transform. The latter graph aggregates information from the previous graph for corresponding nodes.
• Figure 3: An abstract illustration of Temporal Graph Pooling. At each layer, we utilize time convolution to cluster nodes and extract the temporal features. Then we reconstruct adjacency matrices through convolution weights.
• Figure 4: Boxplot showing the accuracies on 12 UEA datasets vs. changes in the number of graphs $n_G$, pool ratio $\eta$, and learning rate.
• Figure 5: Critical difference diagram on the 26 UEA datasets with $\alpha = 0.05$.

Theorems & Definitions (3)

• Definition 1: Temporal Graph
• Definition 2: Static Graph
• Definition 3: Dynamic Graph