A parallel learning method of neural networks with feature extraction mechanism by autoencoder

摘要

In general, practical problems are complicated. Deep learning is a methodology of machine learning using neuralnetworks with a large number of layers. The superior capability of deep learning for practical problems is shownbecause of its higher level representation. However, learning problems cause a learning plateau. Previous researcheshave indicated the effectiveness of pre-training. Pre-training obtains a good feature that is emblematic of a problemand good initial weight value. This study proposes a pre-training method in parallel with fine-tuning using denoisingautoencoder that is used corrupted input. The effectiveness of the proposed method is confirmed through computationexperiments in which learning problems with corrupted data are used.