Softmax
A function turning a vector of arbitrary scores into positive numbers that sum to one. It exponentiates first, so it is far sharper than simply dividing by the total.
What people get wrong
Only differences between scores matter. Adding the same constant to every input leaves the output identical, which is why implementations subtract the maximum for numerical stability.
A definition is not an understanding. This is taught properly in What a neural network is, part of Neural networks.
Learn this properlyTerms from the same topic