Author: Mugunthan Machine Learning General Adversarial Network is a generative model that uses deep learning neural networks like CNN. Two types of modeling methods exist: i) Discriminative and ii). The most common model used in machine learning for classification is the discriminative one. These models take in features X (image) as input and then predict the outcome Y (probability to the image) for those features. Generic models, on the other hand output features X (image), given a random value. Latent Dirichlet Allocation (or LDA) and Gaussian Mixture Model (or GMM) are two examples of generative models. Variational Autoencoders, also known as VAE or General Adversarial Network or GAN, are two examples of Deep Learning generative models. GAN’s workings: A simple representation GAN is made up of two blocks: a generator and a discriminator. The generator does not have a discriminator. It is a network with training that can distinguish between fake and real images. The generator generates an image using random values. This is then fed to the discriminator. Next, the discriminator will determine if the image is real or fake. The generator can then fine tune the image generation process, resulting in identical fake images. You can clearly see that the discriminator and generator are fighting for each new image. Below are images created by NVIDIA’s GAN. These are not faces of any individual, but were generated by GAN! This blog should at most clarify what GAN is. A short introduction to the Generative Adversarial networks (GANs), was published originally in on Medium. People are responding and highlighting this story. Published via
