Research On Classification Operation Stability Of LSTM Recurrent Neural Network Based On Variable Value Theory

With the rapid development of the Internet era,artificial intelligence has penetrated into all aspects of daily life,machine learning has achieved relatively remarkable achievements in many fields,and the development of recurrent neural network RNN is more obvious to all,and its development also promotes For its application,the two complement each other to make it successful in many fields such as text classification,sentiment analysis,time series prediction,machine translation,image description generation,speech recognition and text generation.Many different structures have emerged in the development process of RNN,but the LSTM among them is a more widely used and more basic one,and based on the LSTM structure,some variant structures for solving different practical problems have also appeared.Based on LSTM,the corresponding structural components were modified by adding and discarding.In this paper,based on the relevant theoretical knowledge in the variable value system,through the variable value measurement method and different variable value visualization types,the stability of the LSTM recurrent neural network classification operation is studied.By shifting the sequence and then changing the local positional relationship among the sequence elements,we obtain The test data set composed of different sequences,after the trained classifier performs the classification operation,the corresponding classification result data set can be obtained,and the visualization and quantization operations of different comparison units are performed on it,and the obtained graphs and values are further synthesized Analyze and study the stability of LSTM recurrent neural network classification operation.At the same time,this paper selects four different fungal gene sequences as test data,and displays the model test results.The gene sequence is used as the sequence type data input by the model.Due to its sequence distribution characteristics,it has a certain representativeness,and then The test results are interpretable.This article introduces the significance of the study in conjunction with the corresponding background and current situation,and then introduces the relevant basic theories and terminology in the article;then systematically summarizes the overall framework of the model from the aspects of architecture and workflow,and then expands it.According to the general before and after process of experimental data processing,each module in the model and its contained sub-modules are described in more detail.From the input,output,and operation aspects of the module,the work content of the module is specifically explained.Rich graphical results and quantified data table form are used for final output,and a comprehensive comparison and analysis are made.Finally,the research work done is summarized and prospected accordingly.