A Weight Initialization Method Associated with Samples for Deep Feedforward Neural Network
Yanli Yang, Yichuan He · 2020
Artificial neural network is an important force to promote the development of artificial intelligence, but it usually needs to be trained before use. The initial weight given randomly is the most widely used method when training neural networks. However, randomly given initial weights are independent of the samples. A weight initialization method associated with samples for deep feedforward neural network (DFFNN) is proposed. The initial weights set by this method are a combination of the original weights given randomly and the weights obtained by the first epoch training. The initial weights not only have random characteristics, but are also closely related to the samples to be trained. The proposed method is tested with the bearing data provided by the Case Western Reserve University (CWRU) Bearing Data Center. The testing results show that the proposed method can accelerate the training of DFFNN to some extent.