Binary classification

In binary classification, we would produce a single value p as output, with p between 0 and 1.

• This value p would correspond to the probability of being of class 1.
• The probability of being of class 0 would then simply be 1 − p.
• We would then use a threshold 0.5 to decide if the sample is predicted of class 0 or 1.

They often use Activation functions

Multi-class classification is needed when we have more than 2 classes as we can no longer rely on a single output value $p$.