Information Technology Reference
In-Depth Information
other popular local minima were at (2
,
74), (9
,
16) and their images under the
reflection
x
y
. However, this version of opt-aiNet was using (approximations
to) real numbers respresented in floating-point, which were then rounded to the
nearest integer; thus the comparison is not a fair one. It would be worth making
a proper comparison between these two algorithms in future, also comparing
with the results of [Timmis and Edmonds 2004].
In its most basic form (once the search space has been defined) the BCA
requires only one parameter to be specified, namely the mutation rate (i.e. the
probability that each bit in the hotpsot on a string will flip). Including the
megamutation strategy requires at least one additional parameter to be set,
namely the number of steps to wait before megamutation is applied. So far, we
have not explored the effect of scaling on the performance of the algorithm; the
benchmark problem #2 should be good for large-scale numerical tests, because
the asymptotic growth of the number of solutions is known in advance. Finally,
it would be good to develop other practical and theoretical methods for the
analysis of convergence of the BCA and similar algorithms.
↔
Acknowledgements.
This work was carried out at the Easter School on Arti-
ficial Immune Systems at the University of Aberystwyth in April 2006. We are
grateful to the ARTIST network for organization and financial support. Andy
Hone also thanks the EPSRC for funding the project
Nonlinear dynamics of
artificial immune systems
with a Springboard Fellowship.
References
[Andrews 2006] Andrews, P. Private communication (2006); opt-aiNet code available
at
http://www.elec.york.ac.uk/ARTIST/code.php
[Burger 2000] Burger, E. Exploring the Number Jungle: a Journey into Diophantine
Analysis. Providence, RI, American Mathematical Society (2000).
[de Castro and Timmis 2002a] de Castro, L and Timmis, J. An Artificial Immune Net-
work for Multimodal Function Optimisation, in
Proceedings of IEEE World
Congress on Evolutionary Computation
(2002) 669-674.
[de Castro and Timmis 2002b] de Castro, L and Timmis, J. Artificial Immune Sys-
tems: A New Computational Intelligence Approach. Springer-Verlag (2002).
[Clark et al. 2005] Clark, E., Hone, A., and Timmis, J. A Markov Chain Model of the
B-Cell Algorithm, in
Proceedings of ICARIS 2005
, C. Jacob et al. (Eds.).
Springer LNCS
3627
Springer-Verlag (2005) 318-330.
[Dasgupta and McGregor 1992] Dasgupta, D. and McGregor, D.R. Nonstationary
Function Optimization using the Structured Genetic Algorithm, Parallel
Problem Solving from Nature 2 (Proc. 2nd Int. Conf. on Parallel Problem
Solving from Nature, Brussels 1992), pages 145-154, 1992, Amsterdam, El-
sevier
[Dasgupta 1994] Dasgupta, D. Handling Deceptive Problems Using a different Genetic
Search, in
Proceedings of the First IEEE Conference on Evolutionary Com-
putation 1994, IEEE World Congress on Computational Intelligence
(1994)
807-811.
