1. receives a vector of input signals $z=(z_{1},z_{2},…,z_{I})$
   • usually real numbers, denoted by $R^I$
2. each associated with a weight $v_{i}$ ($w_{i}$ in diagram) (aka parameters) that determines how much importance or influence that particular signal has
3. calculates the net input signal, which is the weighted sum of the inputs
4. uses an activation function $f_{AN}$
   • to transform this net input into the output signal ($o$)
     • transforms them into an output between $[0, 1]$ or $[-1, 1]$ (depending on the chosen $f_{AN}$)
   • $f_{AN}$ introduces nonlinearity to the model, enabling it to learn complex patterns
5. also a threshold value/bias ($\theta$)
   • influences output signal's strength
   • If the output signal surpasses the threshold, the neuron "fires" (outputs 1), otherwise, it remains inactive (outputs 0)

Calculating net input signal (3)

Two types of ANs that compute the net input signal

1. Summation Units (SU):

   • weighted sum of all input signals $n e t=\sum_{i=1}^I z_i v_i$

2. Product Units (PU):

   • weighted product of all input signals $n e t=\prod_{i=1}^I z_i^{v_i}$
   • network can represent interactions involving more than two input signals. e.g. $x_{1}⋅x_{2}⋅x_{3}​$ would be a 3rd-order interaction
   • allows to model more complex relationships in the data

Activation Functions $f_{AN}$ (4)

• receives net input signal and bias
• determines the output (or firing strength) of the neuron
• outputs $[0, 1]$ or $[-1, 1]$
  • in fancy terms: [$f_{A N}(-\infty)=0$ or $f_{A N}(-\infty)=-1$] and $f_{A N}(\infty)=1$
• Types of activation functions