ConvnetJS demo: Generative Adversarial Networks on toy 2d data

The simulation below shows a GAN trained on 2d toy dataset with a few data points of fake data generated by GAN (red) and 1 (green) being the real data. The network is set up as:

Feel free to change this, the text area above gets eval()'d when you hit the button and the network gets reloaded. Every 10th of a second, all points are fed to the network multiple times through the trainer class to train the network. The resulting predictions of the network are then "painted" under the data points to show you the generalization. The background color represents predictions from the discriminator, with red being regions of fake data and green being regions of real data.

Go back to ConvNetJS