Thesis on LVQ and SOM for sequences online available

[Marc Strickert]

Please find the PhD thesis

"Self-Organizing Neural Networks for Sequence Processing"
by Marc Strickert, Department of Mathematics and Computer Science,
University of Osnabrück, Germany, 2004

in the internet:
Thesis.pdf    Size: 1.58 MBytes
http://elib.ub.uni-osnabrueck.de/publications/diss/E-Diss384_thesis.pdf

Supervisors: Prof. Dr. Barbara Hammer, Prof. Dr. Helge Ritter


Abstract:
This work investigates the self-organizing representation of temporal
data in prototype-based neural networks. Extensions of the supervised
learning vector quantization (LVQ) and the unsupervised
self-organizing map (SOM) are considered in detail. For LVQ learning,
adaptive metrics are studied with a particular focus on the built-in
detection of data attributes involved for a given classifcation task;
generalized relevance LVQ (GRLVQ) and supervised relevance neural gas
with general metrics (SRNGGM) are discussed. For unsupervised sequence
processing, two modifcations of SOM are pursued: the SOM for
structured data (SOMSD) realizing an efficient back- reference to the
previous best matching neuron in a triangular low-dimensional neural
lattice, and the merge SOM (MSOM) expressing the temporal context as a
fractal combination of the previously most active neuron and its
context.  The first SOMSD extension tackles data dimension reduction
and planar visualization, the second MSOM is designed for obtaining
higher quantization accuracy. The supplied experiments underline the
high data modeling quality of the presented methods.

Keywords:
vector quantization, self-organization, relevance learning,
classification, clustering, sequence processing, context, fractal
representation, GRLVQ, SRNG, MSOM.


 From the table of contents

1 Introduction 2
2 Data preprocessing for temporal networks

I Supervised LVQ-type learning

3 Learning Vector Quantization (LVQ)
4 LVQ with cost function
   Generalized Relevance LVQ (GRLVQ)
   Supervised Relevance Neural Gas (SRNG)
5 BB-Tree: Rules from trained GRLVQ or SRNG networks
6 Experiments:
   Mushroom data, Hypothyroid data, Speaker identifcation,
   DNA splice site recognition

II Unsupervised SOM-type learning

7 Self-Organizing Maps (SOM)
8 SOMSD with lattice topology and alternative lattices
9 Experiments:
   Mackey-Glass time series, Binary automata, Reber grammar

10 Merge SOM (MSOM) with data topology
11 Experiments:
    Mackey-Glass time series, Binary automata, Reber grammar,
    Speaker identifcation by a posteriori MSOM labeling

III Discussion and Outlook 107



--

____ Marc Strickert
____ http://pgrc-16.ipk-gatersleben.de/~stricker/
____ mail: stricker at ipk-gatersleben.de
____ Institute of Plant Genetics and Crop Plant Research (IPK)
____ Pattern Recognition Group (PRG) / Department of Cytogenetics
____ Corrensstr. 3, D-06466 Gatersleben, Germany.   (Room A-0.05)
____ Tel./Fax: ++49 (0) 39482  - 5 182 / - 5 137.