Biomedical Engineering Reference
In-Depth Information
generates a real number between 0 and 1, from a
uniform random distribution, and then it performs
the pick operation if the generated number is lower
than the value of the
Ppick
function. This func-
tion is the product of two factors, which take into
account, respectively, the
absolute
accumulation
of descriptors of a given class and their
relative
accumulation (i.e., with respect to other classes).
While the absolute factor is inspired by the pick
probability defined in (Bonabeau & al., 1999), the
relative factor was introduced here to achieve the
spatial separation of descriptors.
local region accumulates more descriptors of a
class, with respect to other classes,
fr
increases
and accordingly the value of the pick probability
for this class decreases, and vice versa. This fa-
cilitates the spatial separation of descriptors of
different classes. The constants
k1
,
k2
are set to
the value 0.1, as in analogous formulas defined
in (Bonabeau et al., 1999).
The pick operation can be performed with two
different
modes
. If the
copy
mode is used, the
agent, when executing a pick operation, leaves
the descriptors on the current host, generates a
replica of them, and carries such replicas until it
will drop them in another host. Conversely, with
the
move
mode, as an agent picks the descriptors,
it removes them from the current host (except for
the
local
descriptors, which cannot be removed),
thus preventing an excessive proliferation of
replicas.
2
2
f
k
a
2
P
=
⋅
pick
(1)
k
+
f
k
+
f
1
a
2
r
The
fa
fraction is computed as the number of
lo-
cal
descriptors of the class of interest, maintained
by the hosts located in the
visibility region
, out of
the overall number of descriptors of the class of
interest (including both
local
and
remote
) that are
maintained by the same hosts. The value of
fa
is
comprised between 0 and 1, as well as the value
of
fr
, defined below. The
visibility region
includes
all the hosts that are reachable from the current
host with a given number of hops, i.e. within the
visibility radius
, which is an algorithm parameter.
As more
remote
descriptors of a class are accu-
mulated in the visibility region,
fa
decreases, the
first factor of the pick function decreases as well,
and the probability that the agent will picks the
descriptors becomes lower. This facilitates the
formation of accumulation regions. Conversely,
the pick probability is large if a small number of
remote
descriptors are located in the local region
of the Grid.
The
fr
fraction is computed as the number of
descriptors of the class of interest, accumulated in
the hosts located in the visibility region, divided
by the overall number of descriptors, of all classes,
that are accumulated in the same region. As the
Drop Operation
As well as the pick function, the drop function is
first used to break the initial equilibrium and then
to strengthen the mapping of descriptors of dif-
ferent classes in different Grid regions. Whenever
an agent gets to a new Grid host, it must decide,
if it is carrying some descriptors of a given class,
whether or not to
drop
such descriptors in the
current host. As opposed to the pick operation,
the drop probability function
Pdrop
, shown in
formula (2), is directly proportional to the relative
accumulation of descriptors of the class of interest
in the visibility region. Therefore, as accumulation
regions begin to emerge, the drop operations are
and more and more favored in these regions. In
(2) the threshold constant
k3
is set to 0.3, as in
(Bonabeau & al., 1999).
2
f
r
P
=
drop
(2)
k
+
f
3
r
Search WWH ::
Custom Search