Crossentropy loss and Softmax

[Pomo (Malcolm McLean)]

No takers on this question? Nor on my likely related one

Multiclass vs. multiclass-multilabel

Would someone please clarify my understanding of multiclass vs. multiclass-multilabel problems? Two cases… A multiclass problem, where each training image is labelled as class A, B, or C. The loss function is CrossEntropyLoss. The output is three probabilities, for classes A, B, and C. These probabilities necessarily sum to 1 because the final activations are exponentiated and normalized to sum to 1. A multiclass-multilabel problem where each training image is given exactly one tag, A, B,…

I have not been able to figure out the answers on my own.

If no one replies soon, I’ll take them over to stackoverflow and post the replies back here.

Thanks for reading!