Abstract
Genetic algorithms (GAs) have been used to find efficient solutions to
numerous fundamental and applied problems. While GAs are a robust and
flexible approach to solve complex problems, there are some situations
under which they perform poorly. Here, we introduce a genetic algorithm
approach that is able to solve complex tasks plagued by so-called
`'golf-course''-like fitness landscapes. Our approach, which we denote
variable environment genetic algorithms (VEGAs), is able to find highly
efficient solutions by inducing environmental changes that require more
complex solutions and thus creating an evolutionary drive. Using the
density classification task, a paradigmatic computer science problem, as
a case study, we show that more complex rules that preserve information
about the solution to simpler tasks can adapt to more challenging
environments. Interestingly, we find that conservative strategies, which
have a bias toward the current state, evolve naturally as a highly
efficient solution to the density classification task under noisy
conditions.
Users
Please
log in to take part in the discussion (add own reviews or comments).