부산대학교 도서관

Recently, parallel and distributed computing has become indispensable in various application domains, such as cloud computing and extensive data analysis, to meet the growing demands for efficiency and reliability. In this study, we compared sequential and parallel computing techniques applied to a deep learning model for object detection. By leveraging multiple evaluation metrics, such as F 1 score, execution time, and accuracy, we meticulously evaluated the model's performance. Our analysis revealed that parallel execution considerably decreased the training time of model, resulting in an average time savings of 40 % while maintaining the training process's accuracy and quality. These finding have significant implications for real world applications, as they demonstrate tremendously the potential for expedited deep learning model training without compromising accuracy. This can tremendously benefit medical imaging and realtime video analysis, where accuracy and efficiency are paramount.