Abstract:Infectious disease forecasting has been a key focus and proved to be crucial in controlling epidemic. A recent trend is to develop forecast-ing models based on graph neural networks (GNNs). However, existing GNN-based methods suffer from two key limitations: (1) Current models broaden receptive fields by scaling the depth of GNNs, which is insuffi-cient to preserve the semantics of long-range connectivity between distant but epidemic related areas. (2) Previous approaches model epidemics within single spatial scale, while ignoring the multi-scale epidemic pat-terns derived from different scales. To address these deficiencies, we devise the Multi-scale Spatio-temporal Graph Neural Network (MSGNN) based on an innovative multi-scale view. To be specific, in the proposed MSGNN model, we first devise a novel graph learning module, which directly captures long-range connectivity from trans-regional epidemic signals and integrates them into a multi-scale graph. Based on the learned multi-scale graph, we utilize a newly designed graph convolution module to exploit multi-scale epidemic patterns. This module allows us to facilitate multi-scale epidemic modeling by mining both scale-shared and scale-specific pat-terns. Experimental results on forecasting new cases of COVID-19 in United State demonstrate the superiority of our method over state-of-arts. Further analyses and visualization also show that MSGNN offers not only accurate, but also robust and interpretable forecasting result.
Machine Learning,Artificial Intelligence,Physics and Society,Populations and Evolution
What problem does this paper attempt to address?
The problems that this paper attempts to solve are two key limitations in infectious disease prediction:
1. **Long - range connectivity problem**: Existing methods based on graph neural networks (GNNs) expand the receptive field by increasing the network depth, but this is not sufficient to maintain the semantic connectivity between distant but epidemic - related regions. Specifically, when the current model is modeled on a single scale, due to the gradual aggregation of information, the long - distance connectivity is diluted, and it is difficult to preserve the high - order relationships between distant regions due to the oversmoothing problem.
2. **Multi - scale epidemic pattern neglect problem**: Existing methods usually model the epidemic on only one spatial scale and ignore multi - scale epidemic patterns from different scales. For example, the fine - grained epidemic evolution describes the local epidemic pattern, while the coarse - grained evolution contains a broader regional epidemic pattern. These multi - scale patterns are very important for accurate epidemic prediction, but previous studies have not fully considered this.
To solve these problems, the authors propose an innovative multi - scale spatio - temporal graph neural network (MSGNN). This model addresses the above challenges through the following two main components:
1. **Graph structure learning module**: In order to capture long - distance connectivity, this module constructs a multi - scale spatio - temporal graph to handle different epidemic relationships. Specifically, short - range dependencies are defined at the micro - scale to express high - resolution spatial topology; while at the macro - scale, cross - regional signals are regarded as long - range connectivity, thus significantly reducing the propagation path length between distant regions.
2. **Multi - scale graph convolution module**: Considering the differences between different spatial scales, direct cross - scale aggregation is not feasible. Therefore, this module designs a multi - scale information aggregation scheme. First, the aggregated features of each scale are obtained through scale - specific message passing, then the common epidemic patterns in the multi - scale spatio - temporal graph are extracted, and finally these features are integrated through a multi - scale fusion block to generate the final prediction result.
Through these improvements, MSGNN can model the epidemic on multiple scales, thereby improving the accuracy, robustness, and interpretability of prediction. The experimental results show that the performance of this model on the US COVID - 19 dataset is better than that of existing methods.
### Mathematical formula summary
- **Acquisition of node representation**:
\[
X_s=\text{FC}([x_s^{t - L_b:t}, d^{t - L_b:t}] \oplus I_s) \in \mathbb{R}^{N\times L_b\times C}
\]
\[
X_c=\text{FC}([x_c^{t - L_b:t}, d^{t - L_b:t}] \oplus I_c) \in \mathbb{R}^{M\times L_b\times C}
\]
- **Temporal convolution**:
\[
H_s = \text{MaxPool}[\text{TC}(X_s)] \in \mathbb{R}^{N\times C'}
\]
\[
H_c=\text{MaxPool}[\text{TC}(X_c)] \in \mathbb{R}^{M\times C'}
\]
- **Long - range connectivity modeling**:
\[
[A]_{l,k}^s = f(\theta_s^T\cdot\text{Concat}(H_l, H_k))
\]
- **Short - range dependency modeling**:
\[
[A]_{i,j}^c = f(\theta_c^T\cdot\text{Concat}(H_i, H_j, I_i, I_j))+\frac{1}{\lambda}\delta_{i,j}
\]
- **Graph convolution**:
\[
H'_c=\text{GCN}(\tilde{A}_c, H_c)=\tilde{A}_c f(\tilde{A}_c H_c U_1) U_2
\]
\[
H'_s=\text{GCN}(\tilde{A}_s