抄録
Nowadays, a convolutional neural network (CNN) is considered as a deep learning method for image and voice recognition. A CNN can achieve higher recognition accuracy than other approaches since it can automatically extract features by its learning procedure. However, the training procedure of a CNN is time-consuming. Since the functions of a CNN are close to those of a human brain, when a CNN is applied to a complex application, it must be trained by a large amount of training data, resulting in the size of the CNN becoming huge. To train such a huge neural network by computers, a tremendous amount of training time is required. In this paper, an efficient approach is proposed that can markedly reduce the training time while only slightly sacrificing the recognition accuracy of the training procedure.
