Generative Adversarial Networks (GANs)

Flashcards about Generative Adversarial Networks (GANs)

Generative Adversarial Networks (GANs)

Generates data (creates fake data). Involves a generator and discriminator competing against each other.

Types of GANs

Vanilla GANs, Cycle-GAN, Pix2pix GAN, Style GAN, Deep Convolutional GAN (DCGAN), Conditional GAN (CGAN), Pixel Recurrent Neural Network (PixelRNN), DiscoGAN , Super Resolution GAN (SRGAN), InfoGAN, StackGAN

Pdata(x)

Distribution of real data.

p(z)

Distribution of the generator.

x

Sample from Pdata(x).

z

Sample from P(z).

D(x)

Discriminator network.

G(z)

Generator network.

Mathematical Representation of GANs

Formula: V(D, G) = Ex-Pdata(x) [logD(x)] + Ez-p(z) [log(1 - D(G(z))].

Vanilla GANs

Simplest form of GANs using multi-layer neural networks.

Cycle-GAN

GAN where the generator and discriminator are repeatedly trained in a cyclic manner.

Pix2pix GAN

Generates images from text descriptions using CNN for the generator and RNN for the discriminator.

Style GAN

Deep learning algorithm to generate realistic images composed of a generator and discriminator network.

Deep Convolutional GAN (DCGAN)

A type of GAN composed of deep convolutional neural networks for both generator and discriminator.

Conditional GAN (CGAN)

GAN where the generator is conditioned on extra information, such as a label.

Pixel Recurrent Neural Network (PixelRNN)

Generative neural networks that sequentially predicts the pixels in an image along the two spatial dimensions, modeling the discrete probability of the raw pixel values and encode the complete set of dependencies in the image.

DiscoGAN

GAN algorithm designed to produce more realistic images.

Super Resolution GAN (SRGAN)

GAN to create high-resolution images from low-resolution images.

InfoGAN

GAN algorithm thats loss function encourages the generator to produce images that are not only realistic but also informative.

StackGAN

GAN that uses multiple generators and discriminators.

Gated-GAN method

Uses a gating mechanism to control information flow between generator and discriminator for multi-style GANs.

CycleGANs

Used for unsupervised image-to-image translation.

HST-GAN method

Historical style transfer GAN for generating historical text images.

Few-shot learning

Ability to generate high-quality images with very few training examples.

Deepfakes

Generate realistic fake photos or replace people’s faces with other ones using GANs.