softmax function, also known as the normalized exponential function. It is a binary sigmoid function on the promotion of multi-classification is the result of a multi-purpose classification to show up in the form of probability. The following figure shows the calculation method of softmax:
We know that the range of the exponential function range of zero to positive infinity. And the probability value similarities is that they are non-negative real numbers. Then we can take advantage of the exponential function to map the multi-classification results to zero to positive infinity. Then normalized, we will have a probability approximation.
How to sum up softmax output into multi-classification probability:
1) Molecular: by an exponential function, the output is mapped to a real number from zero to positive infinity.
2) denominator: all results are summed, normalized.