Neural Networks

One of the most successful methods of ML.

Building blocks: Perceptron and Synapse.

Perceptron

Typically a function that maps the entire natural range into a bounded interval ([0,1] or [-1,1]).

Example

• Logistic sigmoid function ReLU function, tanh, softmax, etc.

Synapses

Connections from perceptrons of layer (n-1)th to perceptrons of layer nth, each applying a weight to the transmitted value.

Training Neural Networks is mostly about finding the weights of those synapses.