home · mobile · calendar · colloquia · 1995-1996 · 

Colloquium - Whitley

Shaking the Foundations: Genetic Algorithm and Recent Theory
Computer Science Department, Colorado State University

Genetic algorithms are a search process based on simulated evolution. Much of Holland's original theory about the computational behavior of genetic algorithms relies on the assumption that a genetic algorithm intelligently processes samples from different hyperplane subpartitions of the space. The schema theorem, for example, explicitly calculates lower bounds on the expected sampling rate of different hyperplane subpartitions. Furthermore, the 2-armed bandit analogy has been used to argue that genetic algorithms sample hyperplanes in an optimal fashion by allocating exponentially more trials over time to those hyperplanes that are observed to have higher fitness than average. Holland argues that this also implies the genetic algorithm should rank hyperplanes according to fitness.

Recently, many of Holland's ideas have been under attack. We now have better theoretical tools for looking at genetic algorithms, including Markov models. So what have we learned? The evidence for and against the hyperplane processing abilities of genetic algorithms will be reviewed.

Refreshments will be served immediately before the talk at 3:30pm.
Hosted by Andreas Weigend.

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:13)