If we want to forecast the probabilities of n classes (p1, p2..pk), which of the following functions can be utilised as an activation function in the output layer so that the sum of p over all n equals 1?