Generative Adversarial networks
GANs are algorithmic architectures that are used to generate new synthetic instances of data that can pass for real data. As shown in Figure 5.10, GAN is a generative model that trains the following two models simultaneously:
- A Generative (G) model that captures the data distribution to generate plausible data. The latent space input and random noise can be sampled and fed into the generator network to generate samples that become the negative training examples for the discriminator.
- A Discriminative (D) model that compares the generated image with a real image and tries to identify whether the given image is fake or real. It estimates the probability that a sample came from the training data rather than the real data to distinguish the generator’s fake data from real data. The discriminator penalizes the generator for producing implausible results.
Figure 5.10 – The GAN (source: https://developers...
