preview

Single Neuron Character Recognition Essay

Better Essays

Nelson Mandela Metropolitan University

Character
Recognition – Single
Neuron
WRCI 411 – Assignment 1

RXXXXX XXXXXX - 2100XXXX
August 2013

RXXXXX XXXXXX – s2100XXXX
WRCI 411 – Assignment 1

August 2013

Contents
List of Figures and Tables ........................................................................................................................ 1
Figures ................................................................................................................................................. 1
Tables .................................................................................................................................................. 1
Theory …show more content…

The output is based on a function, usually a step function or sigmoid function (basically a rounded or differentiable step function). The neuron should output a 0 until it ‘fires’ when it should output a 1. This allows it to be used as a logic function. For this case the neuron should fire for only one letter/character (and an interfered equivalent) and for all other possible cases it should not fire.
The strength of a neural network is that it can ‘learn’ or be ‘taught’ a pattern and recognise this pattern. From this it should be able to make decisions for cases it has not seen before. There are many methods for teaching a neuron, most relying on derivation to find the slope of the so-called weight-space (some rely on brute-force, but these are seldom used). The basic Gradient Decent rule acts on the slope of the weight-space and will be used for this report.

Method
Teaching
The code was written in m-code and implemented by MathWorks Matlab R2011a.
The artificial neuron is taught by manipulating the values of the elements in the weight matrix (W in the code) until the error in the output matrix (from the dot product of the teaching set (X) and W) is lower than a set number (0.005). This is achieved with a while loop.
Within the while loop is another while loop which loops another error checking loop though each of the vectors of the X matrix. The code edits each of the elements of the W matrix by using the equation: =
− η


Represented in

Get Access