A semi-supervised classification method for improving a generative adversarial network based on principal component analysis

Abstract

The invention belongs to the field of generative adversarial network improvement. The invention discloses a semi-supervised classification method for improving a generative adversarial network based on principal component analysis. A principal component analysis (PCA) method is used for carrying out compression dimension reduction on an original picture, vectors obtained through dimension reduction are used for replacing original random noise vectors to serve as input of a generator, the quality of the picture generated by the generative adversarial network is improved, and meanwhile the accuracy of semi-supervised classification is also improved. According to the principal component analysis method, part of features of original data can be reserved while the data dimension is reduced, a random noise input generator is replaced with the data subjected to dimension reduction, data with higher quality (closer to a real picture) can be generated more quickly, the number of network iterations is reduced, and the classification accuracy is improved.

Description

technical field [0001] The invention belongs to the field of improving generative confrontation networks, in particular to a semi-supervised classification method and system based on principal component analysis to improve generative confrontation networks. Background technique [0002] Generative Adversarial Networks (GAN, Generative Adversarial Networks) is a deep learning model, the model through the framework of (at least) two modules: generator (Generative Model) and discriminator (Discriminative Model) mutual game learning to produce quite good output. The generator is given some hidden information to randomly generate observation data. In the original GAN ​​theory, both the generator G and the discriminator D are not required to be neural networks, but only need to be able to fit the corresponding generation and discriminant functions. However, in practice, deep neural networks are generally used as G and D. The generation confrontation network is composed of two p...

AI Technical Summary

Problems solved by technology

[0006] (1) The input of the generator in the generative confrontation network is random noise, and the randomness will make the parameters of the generator difficult to train, and it is difficult to generate high-quality pictures

[0007] (2) In real life, it is a common phenomenon that there are less labeled data and more unlabeled data, and the existing technology cannot make full use of unlabeled samples to train the network model

The cost of labeling large amounts of data is extremely expensive, or even infeasible, so semi-supervised learning is often required

Images