Research On Data Generation Model Based On Generative Adversarial Network

Data is the foundation of the information society.At present,the production of data is still highly dependent on artificial factors,resulting in high cost and low efficiency.Efficient and independent data production has become a necessary breakthrough to liberate the productivity of the information society.The essence of data production is to obtain expected data results through mathematical operations between data.Artificial intelligence technology related to neural network is expected to become a powerful tool for data production.In recent years,the development of neural network in the field of image is relatively mature,especially the use of traditional supervised learning image recognition technology,at the same time,the field of image generation has become a research hotspot.In particular,the emerging Generative Adversarial Network(GAN),because of its good expansibility and weak supervised learning mode,has played a great role in promoting the development of image generation technology.This paper studies the general theoretical methods of data generation technology and the general data generation model based on deep learning,and finally applies it in the field of image generation,obtaining an image-to-image neural network application.For generation refers to the qualitative characteristics of image data,the purpose of this article to GAN,as the core framework of data generation model,we design a hybrid GAN model,the model compared with the general GAN model improves to do the following three points: In the aspect of network structure,the auto-encoder structure is applied to the generator of GAN to extract the corresponding generative factors from the generative sources with prior information,at the same time,parallel training structure is used to replace the serial training structure of GAN,so as to simplify the complex training process of generative countermeasure network and obtain the experimental results quickly.In terms of training mechanism,the model adopts the method of dynamic prior information in the training process to improve the fitting efficiency of the model generated by the neural network.The experimental results show that compared with the traditional strong prior information model,this model can learn stronger generating ability and generate higher image quality.At the same time,because of the simplification of its training process and network structure,it can get the experimental results faster and is suitable for experimental research.