Biomedical Engineering Reference
In-Depth Information
Figure 17. Comparison between real values of the time series and the RANN predictions (1 hidden
neuron)
Figure 18. RANN architecture (1 hidden neuron)
1 ,8 0 0
2 .0 4 1
-1 .0 1 5
O u tp u t
In p u t
1
3
8 .4 0 4
1 .1 7 8
-0 .2 1 6
-9 .3 4 3
-2 .7 1 9
2
2 .6 4 6
FUTURE RESEARCH DIRECTIONS
With regard to the internal network function-
ing, as it can be studied in the result tables, in
the time series forecast problems the use of time
decreased activation improve the training and
the mean squared error level. This is caused by
the persistence of the internal activation states
of the ANN. Then, these ANN are able to adapt
themselves in a more flexible way to the resolu-
tion of this kind of problems.
In this chapter a real life performance is simulated
into an ANN and training it through the evolu-
tionary computation techniques: the activation
of a neuron by means of the generation of an ac-
tion potential. This idea of modelling the action
potential for emulating real life performances
(functioning of the biological neuron) could be
used either for simulating the functioning of other
Search WWH ::




Custom Search