Short Definition
Softmax converts a vector of scores into a normalized distribution whose values are positive and sum to one.
Intuition
Softmax turns raw model scores into relative probabilities by making larger scores dominate while still keeping every class comparable.
Technical Definition
Softmax exponentiates each score and divides by the sum of all exponentiated scores, producing a categorical probability distribution.
Example
A classifier uses softmax over class logits before computing cross-entropy loss.
Common Misunderstandings
Softmax probabilities can be overconfident.
Softmax normalizes scores; it does not prove the model is correct.