Research And Imporovement Of Recurrent Neural Network

As a typical technology in artificial intelligence,deep learning has attracted more and more attention in recent years,and has been applied in image processing and natural language understanding.Recurrent neural network is a type of neural network in deep learning.The unique design of recurrent neural network makes it very effective on in the processing sequence data and modeling changes in the time series.It has become a popular research area in deep learning.Under this background,this paper research on the improvement of recurrent neural network technology.At present,study on improving recurrent neural network mainly concentrated on two aspects,one is its ability to learn information in long time serial data is insufficient,the other is some network structures of recurrent neural network are very complexed.Based on a typical recurrent neural network——simple RNN,this paper implements an improved simple RNN by weight initialization without changing its structure deeply and then combines it with convolutional neural network to propose an improved model.Experiment results show that the improved network structure improves the training performance as well as training speed,and achieves a good balance between training effect and training speed.Secondly,in order to deal with more complex application scenarios,this paper makes further research based on the previous framework.This paper realizes a seq2 seq framework by using bidirectional recurrent neural network,and combines it with convolution neural network to propose an improved model of Bidirectional RNN based on seq2 seq.The extended model has the ability to deal with more complex tasks.The experimental results show that the extended network structure can deal with more complex tasks and has good performance.