Recurrent Neural Networks algorithms and applications
Yuexing Chen, Jiarun Li
Abstract
Since recurrent neural networks (RNNs) were firstly proposed, it is widely used, and many extended RNNs algorithms have been developed, which achieve good results in many application fields. To report the latest research results of RNNs and help researchers quickly understand the latest progress of RNNs algorithm. In this research, we briefly introduce the basic principle of RNN, review more than 20 papers about RNNs and summarize the previous work. Then, we make qualitative and quantitative analyses on four representative algorithms (CNN-RNN, ResNet-110, ResNet-164, and ResNet-1001). Among them, the model of CNN-RNN obtains the best performance on the CIFAR-100 dataset with standard data augmentation. Finally, we discuss the future development trend of RNNs.