@techreport{oai:ipsj.ixsq.nii.ac.jp:02000696,
 author = {上野,里奈 and 西島,直 and Rina Ueno and Nao Nishijima},
 issue = {10},
 month = {Feb},
 note = {モデル学習などのML（Machine Learning）処理は，大量データを扱うため，近年では複数ノードで構成される分散コンピューティング環境が広く利用されている．ML技術の急速な進化に伴い，アプリケーションやライブラリ，ハードウェアなどの更新頻度は極めて高い．これらの更新は，モデルの精度向上や新機能の実現に不可欠である．しかし，分散環境を更新するたびにML処理は停止・中断され，処理完了までの時間が大幅に延びてしまう．また，更新時に互換性の問題が生じるとシステム全体の動作が不安定になるリスクがある．本研究は，MLアプリケーションの開発の加速，分散コンピューティング環境の運用効率向上を目的とし，互換性を確保しつつ効率的に環境を更新する手法を提案する．, Machine learning (ML) processes, such as model training, involve handling large-scale data, and in recent years, distributed computing environments comprising multiple nodes have been widely utilized. With the rapid evolution of ML technologies, the update frequency of applications, libraries, and hardware is exceedingly high. These updates are essential for improving model accuracy and enabling new functionalities. However, every update to the distributed environment results in the suspension or interruption of ML processes, significantly extending the time required for task completion. Furthermore, compatibility issues during updates pose risks of instability in the overall system operation. This study aims to accelerate the development of ML applications and enhance the operational efficiency of distributed computing environments by proposing a method for efficiently updating the environment while ensuring compatibility.},
 title = {分散コンピューティング環境の機能更新スケジューリングの提案},
 year = {2025}
}