ABSTRACT

QUANTIFYING THE ADAPTIVE CAPACITY OF GENETIC SYSTEMS

Dr. Chris Watkins Department of Computer Science, Royal Holloway, University of London

Could evolution produce organisms of arbitrary complexity? If not, what factors limit the maximal possible complexity of organisms? How much information can be maintained in genomes by selection? This is a rather basic theoretical question in biology, genetic algorithms, and, perhaps unexpectedly, in other areas of machine learning.

The seminar will describe a (hopefully novel) rigorous definition of adaptive capacity in terms of the maximal possible number of distinct varieties of organism that can be reliably produced in an evolutionary system. This number can be approximately calculated for various genetic systems, and its dependence on genome size, mutation rate, and method of encoding will be explored.

An unexpected and quite strong conclusion is that, in principle, information can be stored most efficiently in genomes the more diffusely it is represented. This finding suggests that some genetic information may be stored in ways that are undetectable by current biological sequence comparison methods.


Maintained by rbennett@cs.ucl.ac.uk