Dr. Ting Hu
Computational Genetics Laboratory
Geisel School of Medicine, Dartmouth College

Robustness and Evolvability in Evolutionary Computation

Candidate for Faculty Position in Computer Science

Department of Computer Science
Tuesday, October 28, 2014, 2:30 p.m., Room EN-2022

---

 

Abstract

Robustness, the maintenance of a phenotype character in the presence of genotypic changes, is the result of a redundant mapping between genotype and phenotype, where many mutational variants of a genotype produce an identical phenotype. Such robustness, at first glance, seems to hinder the capability of innovating, i.e., evolvability. However, empirical evidences have been reported on living organisms that in robustness can facilitate evolvability since it allows genetic variants to expand in neutral spaces that provide a staging ground for future adaptive innovations. Redundant genotype to phenotype mapping is common in Evolutionary Computation. We use genotype networks to characterize the mutational connectivity of an Evolutionary Computation system, and to further quantify robustness and evolvability at the genotype, phenotype, and fitness levels. We show that robustness and evolvability correlate very differently at these three levels, and their interplay crucially depends on how the mutational connections are distributed among phenotypes.