Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.


Gan Inspired Innovations

  • Be the first to comment

  • Be the first to like this


  1. 1. GENERATIVE ADVERSARIAL NETWORKS The most interesting idea in the last 10 years in ML. - Yann LeCun, Facebook’s AI Research Director
  2. 2. BRIEF OVERVIEW • Generative vs Discriminative models: • Discriminative – Learn boundary between classes, P(y|x) • Generative – Learn distribution of all classes, P(x,y) • Advantages of generative model: • Bayes rule for discrimination. • Underlying structure of data. • Two networks in GANs: Generator and Discriminator • Evolution of GANs • Applications
  3. 3. EVOLUTION OF GAN • DCGAN • Improved DCGAN • Conditional GAN • Info GAN • StackGAN • Others
  5. 5. DISCRIMINATOR • Our discriminator is a convolutional neural network that takes in an image of size 28 x 28 x 1 as input and returns a single scalar number that describes whether or not the input image is "real" or "fake"
  6. 6. GENERATOR • You can think of the generator as a kind of reverse convolutional neural network. A typical CNN like our discriminator network transforms a 2- or 3-dimensional matrix of pixel values into a single probability. A generator, however, takes a d-dimensional vector of noise and upsamples it to become a 28 x 28 image.
  7. 7. EXAMPLE SAMPLES Generated samples after training a DCGAN on celebA dataset with 202K images Generated bedroom samples
  9. 9. OTHER INTERESTING APPLICATION • Caption to image generation. • Results: • the flower shown has yellow anther red pistil and bright red petals. • this flower has petals that are yellow, white and purple and has dark lines • the petals on this flower are white with a yellow center • this flower has a lot of small round pink petals. • this flower is orange in color, and has petals that are ruffled and rounded. • the flower has yellow petals and the center of it is brown • this flower has petals that are blue and white.
  10. 10. • Image generation from text • High resolution caption to image generation. • Uses two stacked GANS. StackGAN