FaceNet/NeuralStyleTransfer Conceptual Doubt

For any model, given a training set with labels,
The model tries to fit the data by working on an optimization problem on the loss function.

For some models, given training set data only and not the labels*[128D vector in face net, Generated image in Neural Style Ttansfer],
if the loss is calculated in some other way for a training example,
Again the model tries to get the best possible labels* by working on an optimization problem on the loss function.

If the first set of models comes under a supervised learning problem, where do the second set of models fall, as they don’t have labels? Are they also called supervised learning problems or just optimization problems?

Hi sgurajap,

Maybe this explanation clarifies?