The scene shows a blue point cloud forming a fixed real-data ring manifold and an amber point cloud representing generator samples that start scattered and, epoch by epoch, morph toward matching the ring, while a translucent grid surface representing the discriminator's decision boundary reshapes itself around the shifting fake cloud.
Pick a training mode from the dropdown, use the speed slider to control how fast epochs advance, and press Play/Pause or Rebuild to restart the run from a freshly scrambled generator state.
Switch between Training, Mode Collapse and Converged modes, adjust epoch speed, and play/pause or rebuild the simulated GAN run.
The original 2014 GAN paper by Ian Goodfellow was reportedly conceived after a late-night bar conversation about whether a network could learn to generate data purely by competing against another network trying to catch it faking.
The scene shows a blue point cloud forming a fixed real-data ring manifold and an amber point cloud representing generator samples that start scattered and, epoch by epoch, morph toward matching the ring, while a translucent grid surface representing the discriminator's decision boundary reshapes itself around the shifting fake cloud.
The scene shows a blue point cloud forming a fixed real-data ring manifold and an amber point cloud representing generator samples that start scattered and, epoch by epoch, morph toward matching the ring, while a translucent grid surface representing the discriminator's decision boundary reshapes itself around the shifting fake cloud.
Pick a training mode from the dropdown, use the speed slider to control how fast epochs advance, and press Play/Pause or Rebuild to restart the run from a freshly scrambled generator state.
The original 2014 GAN paper by Ian Goodfellow was reportedly conceived after a late-night bar conversation about whether a network could learn to generate data purely by competing against another network trying to catch it faking.