Scholarlink

The global spread of COVID-19 has emphasized the need for accurate pandemic prediction. While previous studies used spatiotemporal and graph-structured mobility data for outbreak forecasts, these models often suffer from long training times and high computational demands, limiting their effectiveness in dynamic scenarios. Additionally, varying regional mobility patterns add complexity, making manual model adjustments difficult. This paper presents AutoGMN, an automated architecture search framework utilizing bidirectional Graph Mamba Networks. We construct a graph where nodes represent regions, with historical COVID-19 data and human mobility as edge weights. The model forecasts future case numbers, integrating transmission control strategies. To reduce manual intervention, we employ differentiable neural architecture search. Our approach, validated against benchmarks in three European countries, shows superior performance in epidemiological forecasting.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Customizing Spatial-Temporal Graph Mamba Networks for Pandemic Forecasting

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Haowei Xu,
  • Chao Gao,
  • Xianghua Li,
  • Zhen Wang,
  • Tanimoto Jun

摘要

The global spread of COVID-19 has emphasized the need for accurate pandemic prediction. While previous studies used spatiotemporal and graph-structured mobility data for outbreak forecasts, these models often suffer from long training times and high computational demands, limiting their effectiveness in dynamic scenarios. Additionally, varying regional mobility patterns add complexity, making manual model adjustments difficult. This paper presents AutoGMN, an automated architecture search framework utilizing bidirectional Graph Mamba Networks. We construct a graph where nodes represent regions, with historical COVID-19 data and human mobility as edge weights. The model forecasts future case numbers, integrating transmission control strategies. To reduce manual intervention, we employ differentiable neural architecture search. Our approach, validated against benchmarks in three European countries, shows superior performance in epidemiological forecasting.