Deep Learning - Is There a Convergence?

In the embedded space, you have a structure which is not just like the structure of the cone. That in this case, actually you can find sensuis which lie on a circle around the origin,. Which induce the classes which ave cont so this is a kind of novel equivalent. The idea is that if my training data set is all in one specific region, then it wouldn't make sense to assign very high confidenceIf i get no new data point which is mapped very, very far away from all other clusters. Usually my training data points do not map so far out in the outskirts. And so i would assign a low confidence to it. And this confidence is a problem. For