hn2 / synthetic-data-gan

Experimenting with generating synthetic data using ydata-synthetic

Artificial Data Generation using GANs

1. GAN - Generative Adversarial Networks

An excellant and more detailed read on GANs: Google Developers

1.1 GAN Structure

A GAN consists of 2 parts - a generator and a discriminator, both neural networks which try to compete with each other.

• The Generator tries to generate content which is ideally supposed to look like the real content, which can be image, text, a just numerical data in general.

  • The generator is penalised if the discriminator is able to distinguish between real and generated content

• The discriminator tries to tell apart the generated content and the real content

  • The discriminator is penalised if it's not able to distinguish between real and generated content

1.2 GAN Training

• Discriminator
  • Tries to classify real data from generated one
  • Discriminator loss is calculated
  • Gradients of discriminator are computed through backpropagation based on discriminator's loss
  • Computed gradients used to change discriminator's weights

• Generator
  • Produce output based on random input
  • Discriminator tries to classify this output as real or fake
  • Generator loss is calculated based on discriminator's classification efforts
  • Gradients of generator are computed through backpropagation based on generator's loss
  • Computed gradients used to change generator's weights