self-checking
pertaining to a circuit, with
respect to a set of faults, if and only if it is
fault-secure and self-testing.
self-commutated
See
natural commutation
.
self-demagnetizing field
a field inside of
a permanent magnet that is opposed to its own
magnetization, which is due to internal cou-
pling of its poles following the introduction
of an air gap in the magnetic circuit.
self-electro-optic device (SEED)
a bi-
stable device that is a PIN photodetector and
also an optical modulator; the intrinsic region
is generally constructed as a quantum-well
stack. Detection of light alters the electri-
cal bias on the PIN, which in turn alters the
transmission through the device; the optical
transmission change exhibits hysteresis and
a two-state transmission character.
self-focusing
focusing of an electromag-
netic beam in a nonlinear medium by the gain
or index profile resulting from the action of
the beam on the medium.
self-generating neural network (SGNN)
networks of self-organizing networks, each
node network of which is an incomplete self-
organizing network. For this kind of network
of neural networks, not only the weights of
the neurons but also the structure of the net-
work of neural networks are learned from the
training examples.
SGNN can be as complex as acyclic di-
rected graph, but the most frequently used
SGNN takes a tree structure and is called a
self-generating neural tree (SGNT), which
is very similar to self-organizing tree but
with much higher ratio of neuron utilization.
Since many fewer neurons participate in the
competition during the training and classifi-
cation, the speed of SGNT is much faster.
SGNN has found applications in diagno-
sis of communication networks, image/video
coding, large-scale Internet information ser-
vices, and speech recognition.
See also
self-generating neural tree
,
self-organizing
neural tree
.
self-generating neural tree (SGNT)
a
simplified version of self-generating neural
network with a tree structure. SGNT is nor-
mally much faster in training and classifica-
tion, but with less descriptive power com-
pared with the corresponding SGNN because
of its simple topological structure.
How-
ever, if the number of network nodes is the
same, SGNT has the same descriptive power,
higher ratio of neuron utilization, higher
speed, and may end up with higher accuracy,
since large-scale networks can be generated
and trained quickly. See also
self-generating
neural network
,
self-organizing neural tree
.
self-modifying code
a program using a
machine instruction that changes the stored
binary pattern of (usually) another machine
instruction in order to create a different
instruction which will be executed subse-
quently. Definitely not a recommended prac-
tice and not supported on all processors.
self-organizing algorithm
a training al-
gorithm for a self-organizing system consist-
ing of the following main steps:
1. Calculate the similarities of the training
vector to all the neurons in the system and
compare them to find the neuron closest to
the training vector, i.e., the winner.
2. Update the weights of the winner and
its neighborhood according to
w
i
(t + 1) = w
i
(t) + α (x
i
(t) − w
i
(t)) ,
where
w
i
(t) is the ith weight of the neuron
at time
t, x
i
(t) is the ith component of the
training vector at
t, and α is a training rate.
The neighborhood of the winner starts from a
bigger area and reduces gradually during the
training period.
self-organizing neural tree (SONT)
a
tree-like network of self-organizing neural
networks, each node of which is a Koho-
nen network.
Each of the neurons in the
higher level networks has its child network
c
2000 by CRC Press LLC