A Neuro-genetic approach using simple gene regulatory network to construct a distributed and nested adaptive neural network
Loading...
Date
2008-06
Authors
Rahmat, Romi Fadillah
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
For a long time neural networks have been a popular approach for intelligent
machines development and knowledge discovery. However, problems still exists in
neural networks, such as fixed architecture and excessive training time. One of the
solutions to unravel this problem is by using neuro-genetic approach. A neurogenetic
approach is inspired by a theory in neuroscience which state that the
evolution of human brain structure is significantly affected by its genome structure.
Hence, the structure and performance of a neural network should be based on a gene
created for it. Therefore, to overcome these existing neural network problems and
with the new theory of neuroscience in mind, in this thesis we will attempt to
propose a neuro-genetic approach by using simple Gene Regulatory Network (GRN)
as a more biologically plausible model of neural network.
In realizing our proposed neuro-genetic approach, we ended up with two
methods. Firstly, we proposed GRTE, a gene regulatory training engine to control,
evaluate, mutate and train genes. Secondly, ANNN, a distributed and nested adaptive
neural network that will be constructed based on the genes from GRTE. ANNN can
be applied in two modes, ANNN Correlated Data and ANNN Uncorrelated Data. We
conducted two experiments to evaluate and validate our neuro-genetic approach. In
the first experiment, 8-bits XOR parity problems were used as a dataset, while in the
second experiment, the Proben I 's Gene Benchmark Datasets were used. The results
of the experiments validate the objective of this work.
Description
Keywords
Genetic approach using simple gene regulatory , Nested adaptive neural network