Artificial Neuron




February 8, 2017

About artificial neuron.

Artificial neuron is simply a computational unit, which makes a particular computation based on other units it is connected to.

Pre-activation (or input activation): a(x) = b + ∑wi + xi

Neuron (output) activation: h(x) = g(a(x))

where, g(.) is called activation function, b is called neuron bias and w is the connection weights.

Activation functions:

Capacity of a single neuron

Capacity essentially means, the complexity of the computation that a neuron can perform.

If the input is transformed in a better representation, we can actually have a linear seperable.

Thereby, complicated computations cannot be modelled by single neuron. This is the main reason behind why we go for multilayer neural network.