DOIONLINE 
DOIONLINE Publish In |
International Journal of Management and Applied Science (IJMAS)-IJMAS |
 Journal Home Volume Issue |
||||||||
Issue |
Volume-9,Issue-2 ( Feb, 2023 ) | |||||||||
Paper Title |
Proper Batch Size for Deep Learning with Small Datasets | |||||||||
Author Name |
Zhang Meng, Fuminori Kimura, Osamu Honda | |||||||||
Affilition |
||||||||||
Pages |
5-9 | |||||||||
Abstract |
Generally, deep learning needs significant number of training data in order to achieve sufficient efficiency. However, there are not a few cases that sufficient number of training data are not collected. In this paper, the authors investigated that proper batch size for deep learning with small datasets. The authors try to discriminate small foreign materials using You Only Look Once (YOLO). The effect of batch size and data size on the effect of the model was explored by experiments varying the number of training data images and batch data size. The experimental results demonstrated that there is an appropriate batch size for that is neither too large nor too small. Besides, the recall was found to be strongly affected by batch size, on the other hand, the precision was largely unaffected by it.. Keywords - YOLO, Batch Size, Object Detection | |||||||||
| View Paper | ||||||||||
