In the supervised learning regime, the true values of hue were presented in the form of “correct answers,” allowing the neural network to learn the transformation between RGB and Hue.

In most cases, the neural network was able to learn the representation from its 949 training examples, in some sense “memorizing the correct answers.”

img-0 img-1 img-2 img-3

It is interesting to watch its attempt evolve:

img-4 img-5 img-6 img-7