Adrian Silvescu – Fourier Neural Networks (Article)

Description

Fourier Neural Networks (Article), Adrian Silvescu – Fourier Neural Networks (Article), Fourier Neural Networks (Article) download, Adrian Silvescu – Fourier Neural Networks (Article) review, Fourier Neural Networks (Article) free torent
Adrian Silvescu – Fourier Neural Networks (Article)
size 587 KB
The first mathematical model of a neuron was proposed by McCulloch & Pitts1943.
The underlying idea that this model tries to capture is that the response function of a neuron is a weighted sum of its inputs filtered through a nonlinear function:
y=h(Pwixi í).
Much progress has been done in the field of neural networks since that time
but this idea still remained a very fundamental one. Although the model of the
computational unit(neuron)
per se
is simple, neural networks are powerful com-
puters, higher levels of complexity being achieved by connecting many neurons
together.
In this paper we try to propose more general and powerful models for the
neuron as a computational unit. There are may motivations for this investiga-
tion.
One of them is the fact that although the power of computers increased
quite a lot since 1943 we are still not able to simulate and train but toy-size
neural networks. So although from a theoretical point of view creating com-
plexity out of very basic components is desirable, from a practical point of view
more powerful models of the computational units(neurons) are more appealing
because they can help reduce the size of the networks by some orders of magni-
tude and are also more suitable to coarse grained paralelization. More complex
and powerful computational imply also a more compact representation of the
information stored in the network, making it an improvement from an Occam
razor point of view.
Another motivation towards more general and elaborated models neurons
comes from the discoveries in neurobiology that show more that more complex
phenomena take place at the neuron level.
Although apparently different from the early model of McCulloch&Pitts our
model is still based on the same kind of idea (although in a more general way)
”of computing the output of the neuron as weighted sum of the activations
produced by the inputs”.
We will first introduce a general framework and discuss some of the issues
that appear. Then a particular model, the Fourier Neural Networks is intro-
duced and closely examinated. Next some specific theoretical results are pre-
sented followed by experimental results. Finally the conclusions and further de-
velopment are discussed.
Note:
The Fourier Neural Networks were introduced
in Silvescu1997.