neuron
Table of Contents
A computing structure, specifying:
- input: [0,inf]
- output: [0,1]
- weight: same as input, coupling with input
- activation function: 1
An algorithm using neuron typically would specify:
- kind/behaviour of activation function (sigmoid, threshold)
- output rule (if sigmoid > 1, output 1, otherwise 0)
- learning rule: how to adjust weight.
Backlinks
(setting)
- \(a_0\) - a fixed input unit with weight \(-\theta_j\)
- others usual, like neuron, or if you count input layer, a fully connected neural network between input and output layer.
- types: everything can be real number now(except output)
multilayer perceptron
multiplayer perceptron is a layered architecture of neurons, where:
- neurons are divded into layers
- adjacent layers are interconnected
- the notion of of \(a_0\) in is replaced with
bias
an pattern recognition algorithm with multiple outputs using neurons.
Hebb’s rule
Hebb’s rule is an algorithm of pattern extraction/recognition using neuron.