With the increasing number of layers of deep neural network, the performance of the model tends to be saturated gradually. As a very deep network architecture, the deep residual network shows beneficial characteristics in precision and convergence. This paper deeply studied the essence of deep residual network and intended to design an optimal model for classification in Fashion-MNIST dataset. Due to the significant effects of the hyper-parameters in neural network models, the method of designing model structure and optimizing training process were investigated. Next, the performances of the training models under different conditions were compared, and finally an optimal training model whose accuracy was up to 96.21% was obtained. The experimental result shows that the performance of the model can be improved by partly widening the network and selecting more advanced training process.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.