New papers on neural blind signal processing.

[Simone G.O. Fiori (Pg)]

Dear Colleagues,
I would like to announce three new papers on blind signal processing
by neural networks, which might be of interest to people working on
principal/independent component analysis, blind system deconvolution,
learning on Stiefel-manifold, and probability structure identification.

Best wishes for the incoming new year!
Kind regards, 
Simon.

===========================================================================
"Blind Deconvolution by Simple Adaptive Activation Function Neuron",
S. Fiori, Neurocomputing (full paper)

The `Bussgang' algorithm is one among the most known blind deconvolution 
techniques in the adaptive signal processing literature. It relies on a 
Bayesian estimator of the source signal that requires the prior 
knowledge of the source statistics as well as the deconvolution noise 
characteristics. In this paper we propose to implement the estimator 
with a simple adaptive activation function neuron, whose activation function
is endowed with one learnable parameter; in this way the algorithm does not 
require to hypothesize deconvolution noise level. Neuron's weights adapt 
through an unsupervised learning rule that closely recalls non-linear minor 
component analysis. In order to assess the effectiveness of the proposed 
method, computer simulations are presented and discussed.

Downloadable at: http://www.unipg.it/~sfr/publications/NEUCOM2001.ps

============================================================================
"Probability Density Function Learning by Unsupervised Neurons",
S. Fiori, Int. Journal of Neural Systems (full paper)

In a recent work we introduced the concept of pseudo-polynomial adaptive 
activation function neuron (FAN) and presented an unsupervised information-
theoretic learning theory for such structure. The learning model is based 
on entropy optimization and provides a way of learning probability 
distributions from incomplete data. The aim of the present paper is to 
illustrate some theoretical features of the FAN neuron, to extend its 
learning theory to asymmetrical density function approximation, and to 
provide an analytical and numerical comparison with other known density 
function estimation methods, with special emphasis to the universal 
approximation ability. The paper also provides a large survey of PDF 
approximation via functional expansion on orthogonal bases wrt weighting 
kernels, which lead to eg the Gram-Charlier expansion, and of PDF learning 
from incomplete data, as well as results of several experiments performed 
on real-world problems and signals. One of the experiment presents 
preliminary results about statistical characterization of the macro-
mechanical properties of polypropylene composites reinforced with natural 
fibers.

Downloadable at: http://www.unipg.it/~sfr/publications/IJNS2001.zip

============================================================================
"A Theory for Learning Based on Rigid-Bodies Dynamics",
S. Fiori, IEEE Transactions on Neural Networks (full paper)

A new learning theory derived from the study of the dynamics of an abstract 
system of masses, rigidly constrained over mutually-orthogonal immaterial
axes and moving in a multidimensional space under an external force field, is 
presented. The set of rational-kinematic equations describing system's 
dynamics may be directly interpreted as a learning algorithm for 
neural layers which preserve the ortho-normality of the connection matrices;
as a consequence, the proposed learning theory belongs to the class of 
strongly-constrained learning paradigms that allow a neural network to learn
connection patterns over orthogonal group. Relevant properties of the 
proposed learning theory are discussed within the paper, along with results 
of computer simulations performed in order to assess its effectiveness in 
applied fields. The connections with the general theory of Stiefel-flow 
learning and the Riemannian gradient theory are also discussed, and the 
experiments concern optimal data compression by the PCA and signal separation 
by the ICA. This paper summarizes the work done by the Author on this topic
during the last five years.

Downloadable at: http://www.unipg.it/~sfr/publications/TNN2001.zip

===================================================
Dr Simone Fiori (EE, PhD)- Assistant Professor
Dept. of Industrial Engineering (IED)
University of Perugia (UNIPG)
Via Pentima bassa, 21 - 05100 TERNI (Italy)
eMail: sfr at unipg.it - Fax: +39 0744 492925
Web: http://www.unipg.it/~sfr/
===================================================