Linkage, in the context of genetic algorithms, represents the ability of bu
ilding blocks to bind tightly together and thus travel as one under the act
ion of the crossover operator. The goal of learning linkage has been intric
ately tied with defeating many of the bogeymen of GAs - building block disr
uption, inadequate exploration, spurious correlation and any number of othe
r perceived stumbling blocks. Recent studies have shown that linkage can be
learned in some very simple problems by simultaneously evolving problem re
presentations alongside their solutions. This paper extends the applicabili
ty of these approaches by tackling their primary nemesis, the race between
allelic selection and linkage learning. (C) 2000 Published by Elsevier Scie
nce S.A. All rights reserved.