Research And Application Of Membrane System Based On Hypergraph And Convolutional Neural Network

With the development of society,artificial intelligence(AI)has penetrated into all areas of human life.It is the goal of artificial intelligence to match the logic,thinking,learning,sensing and analysis capabilities of the machine with the human brain.Convolutional Neural Network(CNN)is the current popular direction of artificial intelligence research.Its modeling method is inspired by the human brain nervous system.It is a kind of simulation and reconstruction of the brain nervous system.Because it is a network system constructed based on the function and structure of a biological neural network,it has functions of learning,information processing,and information storage,and has become an important means to realize artificial intelligence.At present,the algorithms of convolutional neural networks are relatively mature,but the membrane system(P System)has great potential.The combination of the two will expand the application field of membrane systems.At the same time,the introduction of Hypergraph can help the membrane system to express more complex and diverse relationships and provide new ideas for the development of the membrane system.This paper combines hypergraph and convolutional neural network with membrane system to construct a new membrane system based on hypergraph and deep neural network.In this new system,the separation of the membrane provides a protective space for the calculation.At the same time,the parallelism of the membrane system allows multiple calculations to be performed simultaneously,thereby improving the calculation efficiency and reducing the time complexity.The addition of hypergraphs enables the system to express complex multivariate relationships well,which provides the possibility for the improvement of subsequent algorithms and the innovation of calculation methods;The forward / backward propagation algorithm of convolution neural network provides the basis for the system to learn,so that the system can self-train parameters and expand its application field.We proved the computational completeness of the new system by simulating a Turing machine,and successfully applied the system to the problem of image classification,proving the feasibility and practicability of the system.Chapters 5 and 6 of this paper focus on the research and application of membrane systems based on hypergraphs and convolutional neural networks.We studied the challenging problem ofclassifying pictures.On the basis of convolution calculation,using the membrane system based on hypergraph and deep neural network to make the following improvements to the picture classification model:First,based on the special inclusion of hyper-membrane,this paper proposes a multi-stream convolution hyper-membrane system architecture.Using maximum parallelism in the same layer to perform convolution operations in multiple membranes can extract features more accurately and quickly.Secondly,this paper designs a new input data processing method.Using this method,we can quickly reduce the dimensionality of the input image while ensuring that the image features are not lost.In this paper,the two feature maps after compression and dimensionality reduction are input into their respective convolution streams for calculation,which reduces the input dimension.Finally,we designed a special nuclear membrane for the convolution kernel:In the multi-stream convolution hyper-membrane system,the distribution method of the convolution kernel is optimized,and the time complexity of the convolution calculation in the optimized system is greatly reduced.We conducted three sets of classification experiments on handwritten digits for three different kinds of convolutional membrane systems.We compared the three sets of experiments with traditional membrane systems and convolutional neural network systems,and obtained good results.From this we can see that the combination of hypergraph and deep neural network has expanded the application field and practicality of membrane system.