B Cell–Inspired Algorithms - Immunological Computation: Theory and Applications - page 138

dopt-aiNet()

1. Initialization
Create a random initial population of B cells
(initial immune network)
2. Immune Network Dynamics
while not stopping criterion met do

2.1. compute B cell fitness

for each B cell do
compute the fitness of current B cell
normalize vector of all B cell fitnesses

2.2. clonal expansion and somatic mutation
for each B cell do

generate N c clones of current B cell
add new clones to current B cell population

for each B-cell do

if mutated B-cell c ′ is better than original B cell c then

c . rank = c . rank +1

c=c ′

else

c . rank = c . rank − 1

if c . rank = 0 then

make c part of the B-cell memory pool

for each B-cell do
if c . rank > 0 then

apply one-dimensional mutation to current B-cell
apply gene-duplication mutation to mutated B-cell

if mutated B-cell m is better than the original then

m . rank = m . rank +1
else
m . rank = m . rank − 1
2.3. compute average fitness
compute average fitness of the B-cell population
2.4. network supression and metadynamics
if average error stagnates apply B-cell suppression mechanism using
suppression threshold s and introduce a percentage d of new
randomly generated B-cells into the network

2.5. population control

if average error stagnates and the number of B-cells exceeds

the maximum number of B-cells
then
remove a percentage of less fit B-cells

Figure 5.10 The dopt-aiNet algorithm.