Recent Methodology in Connectionist Systems - Advancing Artificial Intelligence Through Biological Process Applications - page 104

capabilities which would allow us to deal with a
wider range of problems. Moreover, this has indi-
rectly benefited Neuroscience since experiments
with computational models that simulate brain
circuits pave the way for difficult experiments
carried out in laboratories, as well as providing
new ideas for research.

For the same reason, we intend to analyse
how the new CSs solve complex problems, for
instance time processing ones where totally or
partially recurrent networks would play a role.
These networks could combine their functioning
with this new behaviour.

ACkNOWLEDGMENT

FUTURE RESEARCH DIRECTIONS

This work was partially supported by grants from
the Spanish Ministry of Education and Culture
(TIN2006-13274), and grants from A Coruña
University, Spain (Proyectos para equipos en
formación-UDC2005 and UDC2006).

Some algorithms that might cover all the astrocyte
functioning items based on empirical data from
physiological experimentation should be defined;
this definition will enable the continuation of the
research in this field, especially when we face with
the way in which a given astrocyte modifies the
connections of the artificial neurons.
The presence of such a large amount of pos-
sibilities (Porto, 2004; Porto & Pazos, 2006) is
an advantage that allows the experimentation of
various possibilities when facing a real problem,
but it presents at least one disadvantage. How
do we know which is the best option to solve a
determined problem? Mathematically speaking,
it is impossible to know that the final choice is
indeed the best. The use of Evolutionary Com-
putation techniques is proposed in order to avoid
such disadvantage. It will enable the acquisition
of the most suitable artificial glia parameters and
ANGNs-fitted architectures. An experimentation
that might gradually use the different proposed
options is therefore expected by means of the
definition of optimisation protocols and the estab-
lishment of artificial glia behaviour protocols.
Moreover, given that it has been proved that
the glia acts upon complex brain circuits, and that
the more an individual's brain has developed, the
more glia he has in his nervous system (Ramón y
Cajal, 1911), we want to apply the observed brain
behaviour to more complex network architectures.
Particularly after having checked that a more
complex network architecture achieved better
results in the problem presented here.

REFERENCES

Araque, A., Púrpura, V., Sanzgiri, R., & Haydon,
P. G. (1999). Tripartite synapses: glia, the unac-
knowledged partner. Trends in Neuroscience ,
22(5).

Araque, A., Carmignoto, G., & Haydon, P. G.
(2001). Dynamic Signaling Between Astrocytes
and Neurons. Annu. Rev. Physiol , 63, 795-813.

Dorado, J. (1999). Modelo de un Sistema para la
Selección Automática en Dominios Complejos,
con una Estrategia Cooperativa, de Conjuntos de
Entrenamiento y Arquitecturas Ideales de Redes
de Neuronas Artificiales Utilizando Algoritmos
Genéticos . Phd Thesis. Facultad de Informática.
University of A Coruña, Spain.

Dorado, J., Santos, A., Pazos, A., Rabuñal, J.R.,
& Pedreira, N. (2000). Automatic selection of the
training set with Genetic Algorithm for training
Artificial Neural Networks. In: Proc. Genetic
and Evolutionary Computation Conference GEC-
CO'2000 , (pp. 64-67) Las Vegas, USA.

Erkmen, I., & Ozdogan, A. (1997) Short term
load forecasting using genetically optimized
neural network cascaded with a modified Koho-