TurSOM: A Turing Inspired Self-organizing Map

chemistoddAI and Robotics

Nov 6, 2013 (3 years and 11 months ago)

84 views

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

TurSOM
: A Turing Inspired Self
-
organizing Map


Presenter: Tsai
Tzung

Ruei


Authors:
Derek Beaton, Iren Valova, Dan MacLean



IJCNN
2009


國立雲林科技大學

National Yunlin University of Science and Technology

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Outline


Motivation


Objective


Methodology


Experiments


Conclusion


Comments


Reference Data

2

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Motivation


The traditional SOM is
s
lower

than
TurSOM

and need for
post
-
processing

methods for cluster identification.

3
















Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Objective



To present a
new

variant of
the SOM algorithm that
utilizes two forms of selforganization:1) neurons, as in the
classical
Kohonen

algorithm and 2) connections, as
presented in
Turing's model of Unorganized Machines
.

4

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Methodology






TurSOM

5

Neuron

Connection

Turing Unorganized Machines

Competitive Learning Techniques

SOM algorithms

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Methodology


Neuron responsibility


Connection responsibility


The gap junction (GJ) mechanism

6

NeuronA


r

NeuronB


Relative bigness

NeuronC


Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Methodology

7


Algorithmic

Explanation

B

100

C

5

A

80

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Experiments


Early
TurSOM

and double spiral problem


8

Purpose

To test the hypothesis of connection
reorganization being beneficial.

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Experiments


Full
-
featured
TurSOM

in handwriting experiment


TurSOM




9

Purpose

To test the full
-
featured
TurSOM

on a

sample from a handwriting dataset

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Experiments


Full
-
featured
TurSOM

in handwriting experiment


typical one
-
dimensional SOM network

10

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Experiments


TurSOM






1D
standardSOM

11

random

the
Peano
-
Iike


convergence

featuring single


chain

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Conclusion


MAJOR CINTRIBUTION


TurSOM

displays behavior of a highly efficient SOM, in terms of both
time

and
computational expense
.


The
TurSOM

algorithm is applicable in a varying number

of fields, just
like the traditional SOM, but
TurSOM

lends

itself more so to
image
processing

and
segmentation
.


No post
-
processing
methods are required in

addition to
TurSOM

to
detect distinct patterns, unlike other

SOM algorithms, due to
TurSOM‘s

connection

reorganization methods.


FUTURE WORK


To take connection reorganization to scale (n
-
dimensional SOM
networks).

12

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Comment


Advantage


Created a more efficient method


Drawback


……


Application


SOM




13

Intelligent Database Systems Lab

N.Y.U.S.T.

I. M.

Reference Data


http://www.im.isu.edu.tw/faculty/pwu/NN/CH06.pptDrawback


http://zh.wikipedia.org/zh
-
tw/%E5%9B%BE%E7%81%B5%E6%9C%BA



14