Information Technology Reference
In-Depth Information
We give the details of our algorithm below.
1.
Initialization:
Create a set of cells called the memory pool (M) and the ARB
pool (P) from randomly selected training data.
2.
Antigenic Presentation:
for each antigenic pattern do:
(a) Clonal Expansion: For each element of M, determine its affinity to the
antigenic pattern, which resides in the same class. Select the highest affin-
ity memory cell (mc) and clone mc in proportion to its antigenic affinity to
add to the set of ARBs (P).
(b) Affinity Maturation: Mutate each ARB descendant of the highest affin-
ity mc. Place each mutated ARB into P.
(c) Metadynamics of ARBs: Process each ARB using the resource alloca-
tion mechanism. This process will result in some ARB death, and ulti-
mately controls the population. Calculate the average stimulation for each
ARB, and check for termination condition.
(d) Clonal Expansion and Affinity Maturation: Clone and mutate the ran-
domly selected subset of the ARBs left in P based on their stimulation
level.
(e) Cycle: While the average stimulation value of each ARB class group is
less than a given stimulation threshold go to step 2.c.
(f) Metadynamics of Memory Cells: Select the highest affinity ARB of the
same class as the antigen from the last antigenic interaction. If the affinity
of this ARB with the antigenic pattern is better than that of the previously
identified best memory cell mc then add the candidate (mc-candidate) to
memory set M. If the affinity of mc and mc-candidate are below the affin-
ity threshold, remove mc from M.
3.
Classify:
Classify data items using the memory set M. Classification is per-
formed in a k-Nearest Neighbor fashion with a vote being made among the k clos-
est memory cells to the given data item being classified.
We can characterize AIRS as follows:
•
Memory: The memory of the AIRS algorithm is in the pool of memory cells
developed through exposure to the training data (experiences);
•
Adaptation: The adaptation occurs primarily in the ARB pool. With each
new experience, AIRS evolves a candidate memory cell in reaction to this
experience. If this memory cell is of sufficient quality, then the memory
structure is adapted to include in it.
•
Decision-making: The initial decision is which memory cell is the most
similar to the incoming training antigen. This cell is used as a progenitor for
a pool of evolving cells. During classification, the primary classification de-
cision is made based on the
k
most similar memory cells to the data item be-
ing classified.
These steps are repeated for each training antigen. After training, test data are
presented only to memory cells. k-NN algorithm is used to determine the classes in
test phase. For more detailed information about AIRS, the reader is referred to
[11, 12].
