At this point, the data distribution that

the generator can generate is so close to the distribution of real data that the discriminator cannot distinguish it. Generated data and real data. That is to say, for any data x generated by the generator, the discriminator has a probability of judging that it is real and a probability of judging that it was generated. The discriminator determines the probability that any input data, whether it is real or generated, is real or generated. In other words, the discriminator becomes a random guesser in Nash equilibrium. This state reflects that the generator has learned how to simulate the distribution of real data and the discriminator can no longer provide useful feedback to guide the training of the 

generator. As an example, suppose there is a counterfeiter Rich People Phone Number List generator and a police discriminator. They are engaging in a "cat and mouse" game. The counterfeiter's goal is to create counterfeit bills that are as authentic as possible to deceive the police. He may only be able to produce crude counterfeit bills at first but as time goes by his skills gradually improve. Improvement can produce more and more realistic counterfeit banknotes. This is like the generator can only generate data that is quite different from the real data at the beginning, but as the training progresses, the generator's generation ability gradually improves and can generate data that is closer and closer to the real data. The police's goal is to

 distinguish real banknotes from counterfeit banknotes as accurately as possible. He may have a weak ability to identify counterfeit banknotes at the beginning, but as he studies counterfeit banknotes, his recognition ability gradually improves and he can identify counterfeit banknotes more accurately. . This is like the discriminator can only roughly distinguish between real data and generated data at the beginning, but as the training progresses, the discriminant ability of the discriminator gradually improves and can more accurately distinguish between real data and generated data. In this process, the counterfeit banknote maker and The police are constantly improving their skills and eventually reaching a dynamic balance. It's like the generator and the discriminator are constantly improving their capabilities during the training process. In the end, we can harvest very realistic counterfeit banknotes "generators and very capable police officers at the same time." "Discriminator. 2.