How to interpret "soft" and "max" in SoftMax regression?

I know the softmax regression form, but I wonder why it has such a name? Or just for some historical reasons?

+4
source share
2 answers

The maximum number of two numbers max(x,y)can have sharp corners / steep edges, which are sometimes an undesirable property (for example, if you want to calculate gradients).

To soften the edges max(x,y), you can use the option with soft edges: the softmax function. It still performs the max function in its core (well, more precisely, its approximation), but is smoothed out.

, .

+4

, xi, , wi xi, 1 (, ), - wi=exp(a*xi) a, . a=0 . , a max, xi. , a, "" . , , SoftMax.

+1
source

All Articles