Databases Reference
In-Depth Information
9.3.5.
Bare-bones PSO
Kennedy
113
has proposed bare-bones PSO with an objective to move the
particles according to its probability distribution rather than by addition of
velocity. In other words bare-bones PSO can be termed as velocity-free PSO.
Bare-bones PSO throw light on the relative importance of particle motion
and the neighborhood topology. The particle update rule was replaced with
a Gaussian distribution of mean (
p
i
+
p
g
)
/
2 and standard deviation
.
This empirical distribution resembled a bell curve centred at (
p
i
+
p
g
)
/
2.
This method works as well as the PSO on some problems, but less effective
on other problems (Richer and Blackwell).
114
Richer and Blackwell
114
have replaced the Gaussian distribution on
bare-bones with a Levy distribution. The Levy distribution is bell-shaped
like the Gaussian but with fatter tails. The Levy has a tunable parameter,
α
,
which interpolates between the Cauchy distribution (
α
= 1) and Gaussian
(
α
= 2). This parameter can be used to control the fatness of the tails.
In a series of trials, Richer and Blackwell found that Levy bare-bones at
α
=1
.
4 reproduces canonical PSO behavior, a result which supports the
above conjecture. Levy spring constants PSO produced excellent results
(Richer and Blackwell
114
). The explanation might lie at the tails again,
where large spring constants induce big accelerations and move particles
away from local optima.
|
p
i
−
p
g
|
9.4. Fuzzy Swarm Net Classifier
To implement the fuzzy net with swarm intelligence, initially we take a set
of fuzzy nets. Each net is treated as a particle and the set of fuzzy nets
are treated as swarm. Each net shares the same memory in a distributed
environment.
7
At any instance of time all the nets are supplied with one input record
and the respective target. All the nets in the distributed environment are
initialized to random weights
w
ij
in the range [0, 1].
Let us consider that the input-output data are given by:
(
X
i
,y
i
)=(
x
1
,i
,x
2
,i
,...,x
n,i
,y
i
)
,
where
i
=1
,
2
,...,N.
The input-output relationship of the above data can
be described in the following manner:
y
=
f
(
x
1
,x
2
,...,x
N
)
