SPEC CPU2000 Benchmark Description File

Benchmark Name: 179.art

Benchmark Author: Charles Roberson & Max Domeika

Benchmark Program General Category: Image Recognition/Neural networks

Benchmark Description:
The Adaptive Resonance Theory 2 (ART 2) neural network is used to recognize
objects in a thermal image.  The objects are a helicopter and an airplane.
The neural network is first trained on the objects.  After training is
complete, the learned images are found in the scanfield image.  A window
corresponding to the size of the learned objects is scanned across the
scanfield image and serves as input for the neural network.  The neural
network attempts to match the windowed image with one of the images it has
learned.

Description of ART 2: The ART 2 neural network models several characteristics
of organic neural processing that is not modelled in more traditional Feed
Forward Neural Networks(FFNN).  In brief, ART 2 neural networks offer the
following advantages over traditional FFNN:

Expectation influences inputs - The past learnings of an ART 2 neural
network influence the matching process.

Creates own classifications - During training, the ART 2 neural network does
not need explicit output information; it creates its own classification
groups.

Learns on-the-fly - ART 2 neural networks are capable of learning and
classifying at the same time.  The benchmark does not use this feature of
ART 2 neural networks.

Contrast enhancement - ART 2 neural networks perform constrast enhancement
through a series of normalizations in the dynamical system.

Input Description: The training files consist of a thermal image of a
helicopter and an airplane.  The scanfile is a field of view containing other
thermal views of the helicopter and airplane.

Output Description: The output data consists of the confidence of a match
between the learned image and the windowed field of view.  In addition, each
F2 neuron's output is printed.  After the entire field of view is scanned
the field of view with the highest confidence of being a match is output.

Programming Language: ANSI C

Known portability issues: None

Reference:

C. W. Roberson, "Design Extensions To Adaptive Resonance Theory Neural
Networks," Master's Project, Clemson University(1994).

M.J. Domeika, C.W. Roberson, E.W. Page, and G.A. Tagliarini, "Adaptive
Resonance Theory 2 Neural Network Approach To Star Field Recognition," in
Applications and Science of Artificial Neural Networks II, Steven K. Rogers,
Dennis W. Ruck, Editors, Proc. SPIE 2760, pp. 589-596(1996).

G.A. Carpenter and S. Grossberg, "ART 2: Stable self-organization of pattern
recognition codes for analog input patterns," Applied Optics, 26,
pp. 4919-4930(1987).
