Long-Term Evolution of Genetic Programming Populations

TL;DR

This paper reports on a long-term genetic programming experiment evolving large binary trees over 100,000 generations, revealing insights into bloat, convergence, and size distribution that challenge existing theories.

Contribution

It presents the first extensive long-term evolution experiment in genetic programming, highlighting limits to bloat and deviations from theoretical size distributions.

Findings

No evidence of building block evolution

Identifies a limit to program bloat

Periods of complete functional convergence observed

Abstract

We evolve binary mux-6 trees for up to 100000 generations evolving some programs with more than a hundred million nodes. Our unbounded Long-Term Evolution Experiment LTEE GP appears not to evolve building blocks but does suggests a limit to bloat. We do see periods of tens even hundreds of generations where the population is 100 percent functionally converged. The distribution of tree sizes is not as predicted by theory.