Information Technology Reference
In-Depth Information
The recognition time per record is 15.7 ms for both tests of File 1.1 and File 2. This
time includes not only computations but mainly reading the record from test file,
visualization of the recognition result (projection of the pattern to 3D SFIN) in both
screens of the emulator, and writing the result into output file.
6 Comparison with Neural Network
There is no possibility of direct comparison between immune and neural networks on
the same data of File 1 and File 2, since none publication has been found on the
training and testing any neural network on these data. Nevertheless, a comparison
between SFIN and neural network has been performed using the sonar benchmark
data available in the same KDD archive [1]. This is the data set used by [9] in their
study of the classification of sonar signals using a neural network. The task is to train
a network to discriminate between sonar signals bounced off a metal cylinder (i.e.
submarine) and those bounced off a roughly cylindrical rock.
The KDD file "sonar.mines" contains 111 patterns obtained by bouncing sonar
signals off a metal cylinder at various angles and under various conditions. The file
"sonar.rocks" contains 97 patterns obtained from rocks under similar conditions. The
transmitted sonar signal is a frequency-modulated chirp, rising in frequency. The data
set contains signals obtained from a variety of different aspect angles, spanning 90
degrees for the cylinder and 180 degrees for the rock.
Each pattern is a set of 60 numbers in the range 0.0 to 1.0. Each number represents
the energy within a particular frequency band, integrated over a certain period of time.
The integration aperture for higher frequencies occurs later in time, since these
frequencies are transmitted later during the chirp.
The label associated with each record contains the letter "R" if the object is a rock
and "M" if it is a mine (metal cylinder). The numbers in the labels are in increasing
order of aspect angle, but they do not encode the angle directly.
Two series of experiments have been reported in [9]: 1) an "aspect-angle
independent" series, in which the whole data set is used without controlling for aspect
angle, and 2) an "aspect-angle dependent" series in which the training and testing sets
were carefully controlled to ensure that each set contained cases from each aspect
angle in appropriate proportions.
A standard back-propagation network was used for all experiments in [9]. The
network had 60 inputs and 2 output units, one indicating a cylinder and the other a
rock. Experiments were run with no hidden units (direct connections from each input
to each output) and with a single hidden layer with 2, 3, 6, 12, or 24 units. Each
network was trained by 300 epochs over the entire training set.
Not surprisingly, the neural network's performance on the test set is somewhat
better when the aspect angles in the training and test sets are balanced. These
classification results of the neural network for "aspect-angle dependent" series are
shown in Tab. 3.
It has been also reported that three trained human subjects were each tested on 100
signals, chosen at random from the set of 208 returns used to create this data set.
Their responses ranged between 88% and 97% correct. However, they may have been
using information from the raw sonar signal that is not preserved in the processed data
sets presented here (according to [9]).
