demogng.tex

关于自组织神经网络的一种新结构程序,并包含了其它几种神经网络的程序比较

\documentclass{article}
\usepackage{a4}
%\usepackage{/u/user/loos/TeX/latex2html/texinputs/html2e}
\usepackage{html}
\usepackage{epsfig}
\usepackage{changebar}
\usepackage{apalike}

%commands for epsf-files with a width which is a fraction of \textwidth
%
% my epsf (not centered, width set to given fraction of textwidth)
\newcommand{\myepsf}[2]{\epsfxsize#2\textwidth \epsffile{#1}}
% my centered epsf (centered, width set to given fraction of textwidth)
\newcommand{\mycepsf}[2]{\epsfxsize#2\textwidth \centerline{\epsffile{#1}}}
% my centered and boxed epsf
\newcommand{\mycbepsf}[2]{\epsfxsize#2\textwidth \centerline{\fbox{\epsffile{#1}}}}
% my boxed epsf
\newcommand{\mybepsf}[2]{\epsfxsize#2\textwidth \fbox{\epsffile{#1}}}

% my standard epsf (centered, full width)
\newcommand{\mysepsf}[1]{\mycepsf{#1}{1}}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

% my new centered epsf (centered, width set to given fraction of textwidth)
\newcommand{\mynewcepsf}[2]{\centerline{\epsfig{width=#2\textwidth, file=#1}}}
% my new centered epsf (centered, width set to given fraction of textwidth)
\newcommand{\mynewchepsf}[3]{\centerline{\epsfig{width=#2\textwidth, height=#3,file=#1}}}
% my new centered and boxed epsf
\newcommand{\mynewcbepsf}[2]{\centerline{\fbox{\epsfig{width=#2\textwidth, file=#1}}}}
% my new standard epsf (centered, full width)
\newcommand{\mynewsepsf}[1]{\mynewcepsf{#1}{1}}
% my new standard height epsf (centered, full width, specified height)
\newcommand{\mynewshepsf}[2]{\mynewchepsf{#1}{1}{#2}}


\newcommand{\A}{{\cal A}}
\newcommand{\bw}{\mbox{\bf w}}
\newcommand{\bxi}{\mbox{\boldmath$\xi$}}

\newcommand{\simImDGNG}[2]{\htmladdnormallink{#1}{http://www.neuroinformatik.ruhr-uni-bochum.de/ini/VDM/research/gsn/DemoGNG/#2.html}\htmladdimg{../smallDuke.gif}}
\newcommand{\simDGNG}[2]{\htmladdnormallink{#1}{http://www.neuroinformatik.ruhr-uni-bochum.de/ini/VDM/research/gsn/DemoGNG/#2.html}}

\title{\htmladdnormallink{DemoGNG v1.5}{../../GNG.html}}
\author{\\\\\htmladdnormallinkfoot{\Large{\emph{Hartmut S. Loos$^1$}}}{http://www.neuroinformatik.ruhr-uni-bochum.de/ini/PEOPLE/loos}\\
\Large{\emph{and}} \\
\htmladdnormallinkfoot{\Large{\emph{Bernd Fritzke$^2$}}}{http://pikas.inf.tu-dresden.de/\~{}fritzke}\\ \\ \\ \\
$^1$\htmladdnormallink{Systems Biophysics}{http://www.neuroinformatik.ruhr-uni-bochum.de/ini/VDM/top.html}\\Institute for Neural
  Computation\\Ruhr-Universit\"at Bochum\\~\\
$^2$\htmladdnormallink{Neural Computation Group}{http://pikas.inf.tu-dresden.de/NN/}\\Artifical Intelligence Institute\\Computer Science
Department\\Technical University Dresden\\~\\~\htmladdimg{../java_powered.gif}}
\begin{document}

\maketitle 
\begin{abstract}
  DemoGNG, a Java applet, implements several methods related to
  competitive learning. It is possible to experiment with the
  methods using various data distributions and observe the learning process.
  A common terminology is used to make it easy to compare one method
  to the other. 
  Thanks to the Java programming language the implementations run on a
  very large number of platforms without the need of compilation or
  local adaptation. Hopefully, the experimentation with the models
  will increase the intuitive understanding and make it easier to
  judge their particular strengths and weaknesses.

%  \begin{rawhtml}
%    This document is available in <a href="../../README"> ASCII</a>-, <a
%    href="../DemoGNG.dvi"> DVI</a>-, and <a href="../DemoGNG.ps"> Postscript</a>-Format.
%  \end{rawhtml}
\end{abstract}

%
% Keywords: neural network, competitive learning, unsupervised learning,
%           LBG, Hard Competitive Learning, Neural Gas,
%           Competitive Hebbian Learning, Growing Neural Gas,
%           Self-Organizing Map, Kohonen Map, Growing Grid, simulation,
%           Voronoi diagram, Delaunay triangulation, probability distribution
%           


\newpage

\tableofcontents
\vspace{2cm}
\listoffigures
\newpage

\section{Introduction}

In the area of competitive learning a rather large number of models exist
which have similar goals but differ considerably  in the way they work. A
common goal of those algorithms is to distribute a certain number of vectors
in a possibly high-dimensional space. The distribution of these vectors should
reflect (in one of several possible ways) the distribution of input signals
which in general is not given explicitly but only through sample vectors.


\section{System Requirements}
To run DemoGNG, a Web browser with Java enhancements
(e.g. \htmladdnormallink{Netscape}{http://home.netscape.com} 2.0 or newer, 
\htmladdnormallink{Microsofts Internet Explorer}{http://www.microsoft.com},
...) is needed or, e.g. the \htmladdnormallink{Sun
  Microsystems}{http://java.sun.com} appletviewer from the Java
Developers Kit (JDK). 

%The JDK is only needed for developement. All necessary files
%are included in the tar archive. The JDK can be obtained for Linux
%from \newline
%\htmladdnormallink{http://www.blackdown.org/java-linux/Mirrors.html}{http://www.blackdown.org/java-linux/Mirrors.html}\newline
%and for other operating systems from \newline
%\htmladdnormallink{http://java.sun.com/java.sun.com/products/JDK/index.html}{http://java.sun.com/java.sun.com/products/JDK/index.html}.

\section{Installation}
\htmladdnormallink{Download}{../../DemoGNGcode.html#download} the
\htmladdnormallinkfoot{compressed tar version}{ftp://ftp.neuroinformatik.ruhr-uni-bochum.de/pub/software/NN/DemoGNG/DemoGNG-1.5.tar.gz},
save it to a file, say DemoGNG-1.5.tar.gz, and then extract the files with 

\begin{verbatim}
% gunzip DemoGNG-1.5.tar.gz
% tar xvf DemoGNG-1.5.tar
\end{verbatim}

Instead, you can \htmladdnormallink{download}{../../DemoGNGcode.html#download}
DemoGNG in \htmladdnormallinkfoot{zip-format}{ftp://ftp.neuroinformatik.ruhr-uni-bochum.de/pub/software/NN/DemoGNG/DemoGNG-1.5.zip}.

After uncompressing and unpacking you should have the following: 

\begin{tabular}{lp{9cm}}
\texttt{COPYING}                &     copy of the GNU General Public License \\
\texttt{Changes}                &       change history for DemoGNG \\
\texttt{ComputeGNG.java}        &       Java source for class ComputeGNG \\
\texttt{DemoGNG.java}           &       Java source for class DemoGNG \\
\texttt{EdgeGNG.java}           &       Java source for class EdgeGNG \\
\texttt{EdgeVoronoi.java}       &       Java source for class EdgeVoronoi \\
\texttt{FPoint.java}            &       Java source for class FPoint \\
\texttt{GraphGNG.java}          &       Java source for class GraphGNG \\
\texttt{GridNodeGNG.java}       &       Java source for class GridNodeGNG \\
\texttt{HalfEdgeVoronoi.java}   &      Java source for class HalfEdgeVoronoi \\
\texttt{LineGNG.java}           &       Java source for class LineGNG \\
\texttt{ListElem.java}          &       Java source for class ListElem \\
\texttt{ListGNG.java}           &       Java source for class ListGNG \\
\texttt{NodeGNG.java}           &       Java source for class NodeGNG \\
\texttt{SelGraphics.java}       &       Java source for class SelGraphics \\
\texttt{SiteVoronoi.java}       &       Java source for class SiteVoronoi \\

\texttt{DemoGNGcode.html}        &    starting page for source documentation \\
\texttt{GenerateHTML.tcl}        &    a tcl-script to generate some
                                      HTML-pages with different parameters \\
\texttt{Makefile}                &       Makefile for Unix \texttt{make} \\
\texttt{README}                  &       the ASCII-version of this document \\
\texttt{SwitchGNG.html}          &       the main HTML-page \\
\texttt{audio/}                  &       subdirectory containing audio
                                         files \\
\texttt{doc/}                    &       subdirectory containing the javadoc
                                         generated documentation of DemoGNG
                                         with the needed images \\
\texttt{tex/}                    &       subdirectory containing the LaTeX 
                                         source of the manual with the
                                         DVI-, PS- and HTML-format
\end{tabular}


Type \texttt{make world} to generate some necessary files (all these files are
included in the archive).

\begin{tabular}{lp{9cm}}
 \texttt{*.html}          &       needed for SwitchGNG.html \\
 \texttt{*.class}         &       compiled classes \\
 \texttt{doc/*.html}      &       javadoc generated HTML-files
\end{tabular}


\section{Starting the Java Applet}

Start the Web browser with the main HTML-page.\newline
Example:
\begin{verbatim}
% netscape SwitchGNG.html
\end{verbatim}
Important: Make sure you have Java support enabled.

Instead, you can also use an appletviewer with one of the generated
HTML-files.\newline
Example:
\begin{verbatim}
% appletviewer GNG.html
\end{verbatim}
Important: The appletviewer requires an applet included in the Web
           page, so you must choose one of the generated HTML-files.


\section{Manual}

\begin{figure}[htb]
  \begin{center}
    \epsfig{file=eps/all_plus.ps, width=\textwidth}
    \caption{Overview}
  \end{center}
\end{figure}

After the Java applet has been launched, the DemoGNG main window
appears. This window can be divided into four regions:

\begin{enumerate}
\item Network Model
\item Drawing Area
\item General Options
\item Model Specific Options
\end{enumerate}

\subsection{Network Model} \label{Network_Model}

\begin{figure}[htb]
  \begin{center}
    \epsfig{file=eps/model-menu.ps, width=3cm}
    \caption{Pull-down menu with all available models}
  \end{center}
\end{figure}

The first part shows the actual algorithm. To select another algorithm
or restart the current, click on the pull-down menu and choose the
desired algorithm. 
The following algorithms are available:
\begin{itemize}
\item \simDGNG{LBG}{LBG_7}~\cite{Linde80}
\item \simDGNG{LBG-U}{LBG-U_7}~\cite{Fritzke97a}
\item \simDGNG{Hard Competitive Learning}{HCL_5} (standard algorithm)
  \begin{itemize}
  \item with constant learning rate
  \item with decreasing learning rate
  \end{itemize}
\item \simDGNG{Neural Gas}{NG_3}~\cite{Martinetz91}
\item \simDGNG{Competitive Hebbian Learning}{CHL_0}~\cite{Martinetz91,Martinetz93}
\item \simDGNG{Neural Gas with Competitive Hebbian Learning}{NGwCHL_2}~\cite{Martinetz91,Martinetz94}
\item \simDGNG{Growing Neural Gas}{GNG_3}~\cite{Fritzke94a,Fritzke95}
\item \simDGNG{Growing Neural Gas with Utility}{GNG-U_10}~\cite{Fritzke97b}
\item \simDGNG{Self-Organizing Map}{SOM_8}~\cite{Kohonen82b}
\item \simDGNG{Growing Grid}{GG_8}~\cite{Fritzke95b}
\end{itemize}

\newcommand{\pwida}{0.48}
\begin{figure}[htb]
\begin{minipage}[b]{\pwida\textwidth}
  \mysepsf{eps/LBG_Delaunay.ps}
  \centerline{LBG, LBG-U}
  \medskip
\end{minipage}
\hfill
\begin{minipage}[b]{\pwida\textwidth}
  \mysepsf{eps/HCL_Voronoi.ps}
  \centerline{Hard Competitive Learning}
  \medskip
\end{minipage}
\hfill
\begin{minipage}[b]{\pwida\textwidth}
  \mysepsf{eps/NG.ps}
  \centerline{Neural Gas}
  \medskip
\end{minipage}
\hfill
\begin{minipage}[b]{\pwida\textwidth}
  \mysepsf{eps/CHL_Signals.ps}
  \centerline{Competitive Hebbian Learning}
  \medskip
\end{minipage}
\hfill
\begin{minipage}[b]{\pwida\textwidth}
  \mysepsf{eps/NGwCHL.ps}
  \centerline{Neural Gas with CHL}
  \medskip
\end{minipage}
\hfill
\begin{minipage}[b]{\pwida\textwidth}
  \mysepsf{eps/GNG_Teach.ps}
  \centerline{Growing Neural Gas, GNG-U}
  \medskip
\end{minipage}
\hfill
\begin{minipage}[b]{\pwida\textwidth}
  \mysepsf{eps/SOM.ps}
  \centerline{Self-Organizing Map}
  \medskip
\end{minipage}
\hfill
\begin{minipage}[b]{\pwida\textwidth}
  \mysepsf{eps/GG.ps}
  \centerline{Growing Grid}
  \medskip
\end{minipage}
\caption{Example screenshots of the available models} 
\end{figure}

%\begin{figure}[htb]
%\begin{center}
%  \epsfig{width=5.5cm,file=eps/GNG_Teach.ps}