Generative Adversarial Networks Playground

In this demo you can experiment with training a Generative Adversarial Network (GAN) to generate samples from a simple 2D distribution. The GAN is trained to generate samples that are indistinguishable from the real data distribution. The key component is that you don't have to have access to the real data distribution, you only need to be able to sample from it. Go on the right hand side canvas to put down samples for your distribution, and scroll down to see how different variations of GANs learn to generate samples from it. By default, this top section will remain at the top of your screen which allows you to change the data distribution even during training.

Controls:

• Click on the canvas to put down samples for the real data distribution.
• Click on the Clear button to clear the canvas.
• Click on the Toggle Visor button (or press `) to track the training process.
• Click on the Toggle Top Section button (or press t) to keep the top section visible while scrolling.

Clear Toggle Visor Toggle Top Section