Chapter 8 - Glossary

42
The purpose of life.
Back propagate
A way of processing data in a network, going from the output to the input.
Delta Value
A local gradient between the actual output and the desired output in a neuron. See Chapter 4.2 and 4.3 for a more detailed description.
Forward propagate
A way of processing data in a network, going from the input to the output.
Gnuplot
A simple console based program for plotting graphs under both Linux and Windows.
Hidden layer
The layer not visible from outside the network.
Input layer
The input layer of neural network, it does not contain any active neurons.
Kohonen's SOFM
A specific type of SOFM. See Chapter 2.1 for a more detailed description.
Lambda
A neighbourhood function, that computes a value proportional to the distance. The value is large at small distances and small at far distances.
Layer
A part of a network of neurons.
Learning rate
A parameter to decide how quick the network should be able to learn.
Momentum
A parameter used to overcome local minima.
Network
In this report a system of neurons.
Neuron
The building blocks containing all intelligent in the brain. See Chapter 4.1 for a more detailed description.
Output layer
The layer of neurons that gives the output data.
Overtraining
A behaviour that accurse if the network is trained to much. See Chapter 6.2 for a more detailed description.
Processor cache
An internal very fast and expensive memory in a processor.
RMS error
A way of calculating a type of mean error.
RISC-computer architecture
A type of processor architecture, the one used in PC’s.
Sigma
A function used in the lambda neighbourhood function and it is an approximation of the normal distribution.
SOFM
Self organization feature maps. See Chapter 2.1 for a more detailed description.
Supervised Learning
A type of training when an input and the correct output are given to a network.

[ toc | up ]