|
|
Radial network
is based on functions which have non-zero value in given
space around
centers and executes local aproximation, which range of
working is usually
much more limited than in sigmoidal networks. |
|
In radial networks
hided neuron executes radial changing function only in
surroundings chosen centrum - c and only there taking
non-zero value. The role of
hidden neuron is to translate in radial way of the space
surrouding one point or group
of pionts exemplifing cluster. Superposition of signals
coming from all hidden neurons, making by output neuron,
let us obtain translation whole multidimensional space.
That kind of neuron role, lets to connect in easy way,
base function parameters with phisical
deployment of learned data in multidimensional space.
Thanks to that it is
possible relative easy obtaining of start parameters in
supervisioned learning process. Applying similar
learning algorithms with start values close to optimal
multiplies
probability of success obtain for radial networks. |
|
|
Next - Page 3/3
|
|
|
|
| |
|