Skip to content

Approximating a 2D distribution (a subset of R^2) using a simple GAN

License

Notifications You must be signed in to change notification settings

Gholamrezadar/GAN-2D-data

Repository files navigation

GAN-2D-data

Approximating a 2D distribution (a subset of R^2) using a simple GAN

Original Distribution

The original data distribution is a subset of R^2 (a skewed 2D Gaussian)

original

Generated Using GAN

The model has figured out what the safe subset of R^2 is where the generator can fool the discriminator, without ever knowing how the original points where generated. We can make the generated distribution (q) more similar to the real distribution (p) using various techniques (e.g. Earth-Movers Loss) but that's not the point of this experiment. they are left for future more realistic projects.

generated

Loss History

loss_history

About

Approximating a 2D distribution (a subset of R^2) using a simple GAN

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published