@article{oai:ipsj.ixsq.nii.ac.jp:00210639,
 author = {木村, 輔 and 松原, 靖子 and 川畑, 光希 and 櫻井, 保志 and Tasuku, Kimura and Yasuko, Matsubara and Koki, Kawabata and Yasushi, Sakurai},
 issue = {2},
 journal = {情報処理学会論文誌データベース（TOD）},
 month = {Apr},
 note = {本論文では，大規模疫病データのための高速予測手法であるEpiCastについて述べる．EpiCastは，様々な地域の大規模疫病データストリームが与えられたときに，その中から疫病の特徴を表現，要約，共有し，長期的かつ継続的に将来の感染者数予測を行う．提案手法は(a)疫病の複雑な拡散過程を非線形モデルで表現し，(b)それらの中に含まれる重要な特徴を各地域で共有し，適切なモデルを選択することで，感染拡大予測を実現する．ここで，提案手法は(c)データストリームの長さに依存せず，一定の計算時間で感染者数を推定する．COVID-19の実データを用いた実験では，EpiCastが大規模疫病データストリームの中から疫病の重要な特徴を発見，共有することで感染者数を長期的に予測し，さらに，既存手法と比較し大幅な精度，性能向上を達成していることを確認した．, Given a large collection of co-evolving epidemics, how can we forecast their future characteristics? In this paper, we propose a streaming algorithm, EpiCast, which is able to model, understand and forecast future epidemic outbreaks as well as pandemics. Our method has the following features for the effective and efficient modeling of the dynamics of spreading viruses. (a) Non-linear: we incorporate a non-linear equation that is suitable for complex epidemic modeling. (b) Dynamic: it maintains multiple such non-linear models to share important patterns among locations, and chooses the non-linear model for the forecast while monitoring a co-evolving epidemic data stream. (c) Scalable: it can quickly forecast future phenomena at any time in a practically constant time. In extensive experiments using real COVID-19 datasets over major countries, we demonstrate that our proposed method outperforms existing methods for time series in terms of forecasting accuracy, and significantly reduces the required computational time.},
 pages = {10--19},
 title = {大規模疫病データのための将来予測アルゴリズム},
 volume = {14},
 year = {2021}
}