An ill-posed problem for Neural Nets

Original article was published by Gaston Mazzei Gador on Artificial Intelligence on Medium


Back to the present: 20 years later…

Suppose it exists an operation that transforms one of this monsters into the other

Do we have more examples of the transformation? No, we only know that it transforms “random monster from Google” into The Cookie Monster from Sesame Street.

We would like to fit and store the transformation inside a Neural Network to, for example, apply it to people’s faces!

UNFORTUNATELY it cannot be done as it is ill-posed.

The reason should be clear from the 1-D equivalent in the graph below

Using a neural network with only one image/example is the equivalent of trying to fit a 1-D function with only one point