“Comparing and Combining Lexicase Selection and Novelty Search,” co-authored by Lia Jundt ’18, explored an innovative method for selecting parent programs in genetic programming.
The work, which grew out of Jundt’s honors thesis in computer science, combined two methods to create a new parent selection method called novelty-lexicase selection. Analogous to natural selection in biological evolution, parent selection determines which programs in the population get to pass their genetics to future generations.
Helmuth also presented “Lexicase Selection of Specialists,” which was named Best Paper for the Genetic Programming track of the conference.
Written with Edward Pantridge of Swoop, Inc., and Lee Spector of Hampshire College, the paper explained how selecting specialist programs — those good at some skills and poor at others — sets lexicase selection apart from other parent selection methods.