Information Technology Reference
In-Depth Information
maturation
, the mutated clones and edited cells are reselected to ensure that only
those cells with a higher affinity than a certain threshold survive. The whole process
is performed iteratively until a certain stable state is achieved. In PAIA, the principle
is used to provide a selection pressure to effectively drive the population towards the
Pareto front over many iteration steps.
Immune Network Theory.
According to this theory, Abs not only have paratopes
but also epitopes. This results in the fact that Abs can be stimulated by recognizing
other Abs, and for the same reason can be suppressed by being recognized. Conse-
quently, the immunological memory can be acquired by this self-regulation and
mutual reinforcement learning of B-cells. In [19], Farmer
et al.
created an immune
network model defined by a differential equation which demonstrates that Abs' con-
centration is determined by two activations-Abs' activation and Ags' activation, one
suppression-Abs' suppression, and
Apoptosis
. The suppression function is a mecha-
nism that allows to regulate the over-stimulated B-cells to maintain a stable memory.
This metaphor is used in PAIA to regulate the dynamics of the population.
Abs' Concentration.
Initially, only a small number of B-cells cruise in the body. If
they encounter foreign Ags, some of them are activated and then they proliferate. The
immune system should maintain a specific Abs concentration. This process is adap-
tive, i.e. the number of clones that are proliferated during the activation process and
how many of them are maintained at each iteration step and at the end in order to neu-
tralize Ags is adaptive. This makes sense since if a large number of initial B-cells is
available then undoubtedly it can kill any Ags at the cost of spending more energy to
activate B-cells and secrete Abs. However, only an optimal number of B-cells during
each step is necessary (less means more time is needed to reach the required concen-
tration; more means redundant B-cells are introduced). This is the main inspiration for
us to design PAIA's structure so that the population is adaptive at each iteration step.
3 The Algorithm
The synthesis of the above three immune metaphors generates the new algorithm-
Population Adaptive Based Immune Algorithm (PAIA), which aims to:
1.
provide a generic AIS framework for MOP solving;
2.
make the population size adaptive to the problem;
3.
reduce the evaluation times so that only the necessary evaluations are carried out;
Here, we mainly discuss the last two aims, and leave the first one until after pre-
senting the whole algorithm. The last two aims are related to the last problem raised
in Section 2.1 which needs detailing. To preserve the search capability, all population-
based GAs require a sufficiently large population, and such a population is fixed
during the search mechanism. This makes the initial population size crucial to the
success of such algorithms. Deb pointed out in [16] that NSGA II failed to converge
to the true Pareto front for ZDT4 using a population of 100. He suggested (but not
proved) that 500 may be needed for a successful outcome. Hence, the population size
