Author: Zen and the Art of Computer Programming
1 Introduction
GAN (Generative Adversarial Network) is a deep learning method that has become popular all over the world in recent years. It can generate high-quality images, such as hand-painted pictures or anime character skins. The basic idea of GAN is to use a generator (Generator) and a discriminator (Discriminator), and the two networks learn competing strategies respectively, so that the generator can continuously improve its self-generation ability, while the discriminator needs to deceive the generator to the greatest extent. . Therefore, a game relationship is formed between the two, and finally a balance point is reached.
What is a generator? It is a machine learning model composed of a network structure, parameters, and loss function. The input of this model is a random vector, and the output is a real picture or other high-dimensional data. According to different data distributions and task requirements, different generator network structures can produce different results.
What is a discriminator? It is also a machine learning model consisting of a network structure, parameters, and loss function. Its input is a picture or other high-dimensional data, and the output is a probability value, indicating the probability that the picture belongs to a real sample. The goal of the network is to distinguish real samples from fake samples as much as possible.
2. Explanation of basic concepts and terms
2.1 Introduction to GAN
GAN (Generative Adversarial Networks) is a deep learning model consisting of a generator network and a discriminator network. It can generate high-quality images, such as hand-drawn style pictures or anime character skins.
The following content is excerpted from Baidu Encyclopedia:
Generating Adversarial Networks (GANs) is a deep learning model consisting of a generation network and a discriminative network.
Generator refers to the neural network that generates the data required by the model from random input conditions. For example,