Parallel Genetic Algorithms
(K. S. Leung)

Genetic Algorithms (GAs) originated from the theory of evolution, the way through which all lives on earth are believed to come into existence by some scientists. GAs learn or discover novel things by recursively selecting better "offspring" formed from "parents" by processes such as crossover or mutation. GAs are general search and optimisation methods particularly suited for hard problems with unknown search contours. Its major advantages over other search methods are its inherent parallel and global search characteristics. In order to exploit its capability to the extreme, we have to make full use of its inherent parallelism and map it onto a super parallel/distributed architecture. This project will investigate and improve on some existing parallel GAs such as island and co-evolution models. New migration and generation controls will be developed so that we could optimally map the PGAs on a loosely coupled super computing architecture, which may include multi-processor as well as single-processor workstations. One example of such architecture is the super computing network we are constructing in our CS Department. The dependency among genes in chromosome will have to be tackled to maximize the parallelism in the co-evolution paradigm.