What does an artificial neuron do? Simply put, it calculates the weighted sum of inputs, adds the bias, and decides whether to exclude this value or use it further. The artificial neuron doesn't know of a threshold that can be used to figure out whether the output value switches neurons to the activated state. For this purpose, we add an activation function. It checks the value that's produced by the neuron for whether external connections should recognize that this neuron as activated or whether it can be ignored. It determines the output value of a neuron, depending on the result of a weighted sum of inputs and a threshold value.

Let's consider some examples of activation functions and their properties.