I’m really stuck creating a neural network that tells you that he saw this state before, or he never saw it, if not tells not. Like numbers from 0 to 10, I’ve thought of one output neuron that has 0 if number not shown or 1 so, but I’m stuck, maybe 2 outputs (0,1) without softmax or argmax ?
ahem ! well don’t let this topic die
I am not sure how I would do it your way with a neuron that directly classifies seen or unseen.
However, an unsupervised way would be to overfit an autoencoder to some dataset. The idea is that given a sample from the training set, it will achieve a very low loss. Given an unseen sample it would get a very high loss.