Transcript a new framework using ARTMAP neural networks Presenter

國立雲林科技大學
National Yunlin University of Science and Technology
Self-organizing information fusion and
hierarchical knowledge discovery :
a new framework using ARTMAP neural networks
Presenter : Shu-Ya Li
Authors : Gail A. Carpenter*, Siegfried Martens, Ogi J. Ogas
© NN 2005
Intelligent Database Systems Lab
Outline

Motivation

Objective

Methodology

Conclusion

Personal Comments
N.Y.U.S.T.
I. M.
Intelligent Database Systems Lab
Motivation

N.Y.U.S.T.
I. M.
Classifying novel terrain or objects from sparse,
complex data may require the resolution of
conflicting information from sensors working.
Puppy
Dog
Animal
Spot
Intelligent Database Systems Lab
Objective
N.Y.U.S.T.
I. M.

Deriving consistent knowledge from inconsistent information.

An ARTMAP neural network can act as a self-organizing expert
system to derive hierarchical knowledge structures from inconsistent
training data.
Hierarchical
knowledge
Inconsistent
training data
ARTMAP
Intelligent Database Systems Lab
Methodology

Monterey

Ground truth pixels are labeled red
car, other car, roof, road, foot path,
grass, tree, other.
N.Y.U.S.T.
I. M.

Boston

Ground truth pixels are labeled ocean,
ice, river, beach, park, road, residential,
industrial, water, open space, built-up,
natural, man-made.
If need
training
Example
• Red car and other
car pixels are
labeled vehicle.
• Road and foot path
pixels are labeled
pavement.
Example
• The class natural
includes water
(ocean, ice, and
river)
• The class Open
space (includes
beach and park)
testing validation
1 2 3 4
Intelligent Database Systems Lab
Methodology

N.Y.U.S.T.
I. M.
ARTMAP fusion system training
protocol
75 pixels labeled road
No pixels labeled ocean
ocean, ice,
river, beach,
park, road…
4 pixels labeled road
19,919 pixels labeled ocean
maximum number
of labels (here set
equal to 250)
pixels
Intelligent Database Systems Lab
Methodology

N.Y.U.S.T.
I. M.
Rules


equivalence parameter e = 90%
minimum confidence parameter c = 50%
Remove C<c
Intelligent Database Systems Lab
Methodology

N.Y.U.S.T.
I. M.
Graphical representations of knowledge hierarchies
C < 50%
Remove
50% < C < 90%
C > 90%
Intelligent Database Systems Lab
Conclusion
N.Y.U.S.T.
I. M.

The ARTMAP neural network produces one-to-many
mappings from input vectors to output classes, as well as
the more traditional many-to-one mappings.

The procedure is not limited to the image domain.


drug resistance

improve marketing suggestions to individual consumers
Deriving consistent knowledge from inconsistent
information.
Intelligent Database Systems Lab
Personal Comments

Advantage


discover hierarchical knowledge structures
Drawback


N.Y.U.S.T.
I. M.
…
Application

drug resistance

improve marketing suggestions to individual consumers
Intelligent Database Systems Lab