Pick an object and draw it in the training data grid below.
Add and train will train the RBM with the image you have drawn, doing one update of the weights after one reconstruction.
After adding and training more images of the same object, the reconstructions of the image will improve.
Reconstructing an image without training the RBM on it can be done using the test data grid.
Start daydreaming will display multiple reconstructions of the test data grid image.
Training manually using one image at a time takes a long time, so here is
a pre-trained RBM on the MNIST handwritten digit dataset.
Draw a digit in the test data grid and start the daydream to see reconstructions of the digit by the RBM.