摘要 Abstract 

為解決深度學習模型因運算成本過高難以廣泛應用的問題，本研究提出以可解釋神經元進行特徵選擇以降低模型成本。研究聚焦於兩個領域分別為工業應用與時空資料庫，結果驗證本方法在降低成本的同時亦能維持良好效能。

To overcome the high computational cost that limits the wider adoption of deep learning models, this study proposes a novel approach that leverages explainable neurons for feature selection, thereby reducing model complexity and resource demands. The research focuses on two distinct domains—industrial applications and spatiotemporal databases. In both areas, the proposed methods are tailored to the characteristics of the data and are experimentally validated to lower computational cost while maintaining high predictive performance effectively.