Biomedical Engineering Reference
In-Depth Information
FUTURE TRENDS
ANNs, fully connected, without back propagation
and without lateral connections. We have thought
it would be more adequate to start using the latest
research about information processing in CSs for
classification or recognition, since the control of
the connections in the brain lead to the generation
of activation ways.
We believe that the enhancement achieved
with our results is firmly based on what occurs
in brain circuits, given that it coincides with the
most recent observations (Perea and Araque,
2007). Based on these new findings, glia is now
considered as an active partner of the synapse,
dynamically regulating synaptic information
transfer as well as neural information processing.
These results achieved show that the synaptic
modifications introduced in the CSs, and based
on the modelled brain processes (Porto, 2004,
LeRay et al., 2004), enhance the training of the
analysed multilayer architectures. The unsuper-
vised mechanism permits a local adjustment in
the search of the optimal solution. Moreover,
GAs were found to be very efficient in the train-
ing because they helped to adapt the CSs to the
global optimal solution.
In our opinion, CSs are still in a development
phase, possibly even the initial one. Their real
potential has not been reached, or even hinted at
yet. We must remember that the innovation of the
existing ANNs models towards the development
of new architectures is conditioned by the need
to integrate the new parameters into the learning
algorithms so that they can adjust their values.
New parameters, that provide the process element
models of the ANNs with new functionalities,
are harder to come by than optimizations of the
most frequently used algorithms that increase the
calculations and basically work on the computa-
tional side of the algorithm. The ANGNs integrate
new elements and thanks to a hybrid method this
approach did not complicate the training process.
Thus, we believe that the research with these
ANGNs directly benefits Artificial Intelligence
because it can improve information processing
The research in this novel field intends to con-
tinue working, by means of Artificial Intelligence
systems, on the reproduction of the observed bio-
logical processes. More specifically, the present,
as well as future, the main goal of the intended
research works is to increase and to take advan-
tage of the current knowledge regarding nervous
information processing. On the one hand, this
will improve the computer information process-
ing by means of CSs and, on the other hand, it
will enable the consolidation of the hypothesis
about the complex functioning of the biological
neuroglial networks.
It has already been mentioned that, as the way
in which the astrocytes modulate the physiology
of a neural network depending on neural activ-
ity has not been quantified yet, there are many
possibilities of influence over connection weights
(Porto & Pazos, 2006). We have started to study
what happens modyfied the connection weights
according to the neural activity. We keep on
analysing other synaptic modification possibili-
ties based on brain behaviour to apply them to
ANGNs which can solve real problems.
CONCLUSION
Considering the behaviour recently observed in
glial cells (Perea & Araque, 2005; Perea & Araque,
2007) we have decided to create CSs that integrate
some phenomena which are believed to be based
on this behaviour. We have incorporated to the
CSs new elements which will participate together
with the artificial neurons in the information
processing. The initial essential objective was
to check if these new CSs were more efficient
at solving problems currently solved by means
of ANNs.
We have oriented the design of these ANGNs
toward classification problems that are usu-
ally solved by means of feed-forward multilayer
Search WWH ::
Custom Search