• Sorted by Date • Classified by Publication Type • Classified by Research Category •
Ben Torben-Nielsen, Karl Tuyls, and Eric Postma. Shaping Realistic Neuronal Morphologies. A Genetic Programming Approach. In Proceedings of the International Joint Conference on Neural Networks IJCNN'2006, Vancouver, Canada, 2006.
Neuronal morphology plays a crucial role in theinformation processing capabilities of neurons. Despite theimportance of morphology for neural functionality, biologicaldata is hard to obtain and scarce. Therefore, virtual neuronsare devised to allow extensice modelling and experimenting. Themain problem with current virtual-neuron generation methodsis that they impose severe a priori constraints on the virtualmorphologies. These constraints are based on widespread assumptionsand beliefs about the morphology of real neurons.To overcome this problem, we present a new method based onL-Systems and Evolutionary Computation that imposes a posterioriconstraints on candidate virtual neuron morphologies.As a proof of principle, our experiments show the power ofthe new method. Moreover, our method revealed a limitationin the description of neural morphology in the literature.We empirically show that Hillman’s fundamental parametersof neuron morphology are satisfactory but not sufficient todescribe neuronal morphology. The results are discussed andan outline for future research is given. We conclude thatwe succeeded in devising a new method for virtual-neurongeneration that does not impose a priori limitations on thevirtual-neuron morphology.
@INPROCEEDINGS{Tbnijcnn06,
author = {Ben Torben-Nielsen and Karl Tuyls and Eric Postma},
title = {Shaping Realistic Neuronal Morphologies. A Genetic Programming Approach},
booktitle = {Proceedings of the International Joint Conference on Neural Networks
IJCNN'2006},
year = {2006},
abstract = {Neuronal morphology plays a crucial role in the
information processing capabilities of neurons. Despite the
importance of morphology for neural functionality, biological
data is hard to obtain and scarce. Therefore, virtual neurons
are devised to allow extensice modelling and experimenting. The
main problem with current virtual-neuron generation methods
is that they impose severe a priori constraints on the virtual
morphologies. These constraints are based on widespread assumptions
and beliefs about the morphology of real neurons.
To overcome this problem, we present a new method based on
L-Systems and Evolutionary Computation that imposes a posteriori
constraints on candidate virtual neuron morphologies.
As a proof of principle, our experiments show the power of
the new method. Moreover, our method revealed a limitation
in the description of neural morphology in the literature.
We empirically show that Hillman’s fundamental parameters
of neuron morphology are satisfactory but not sufficient to
describe neuronal morphology. The results are discussed and
an outline for future research is given. We conclude that
we succeeded in devising a new method for virtual-neuron
generation that does not impose a priori limitations on the
virtual-neuron morphology.},
address = {Vancouver, Canada},
bib2html_pubtype = {Refereed Conference},
bib2html_rescat = {Machine Learning - Datamining},
owner = {K.Tuyls},
timestamp = {2006.11.23},
}
Generated by bib2html.pl (written by Patrick Riley ) on Mon Jan 18, 2010 09:51:52