Contact Information

Computer Science Department
Colgate University
McGregory Hall, 3rd Floor
13 Oak Drive
Hamilton, NY 13346
(tel) 315.228.7719
Charlotte Jablonski, Administrative Assistant
cjablonski@colgate.edu

Event Detail

Dept Tea: Evolving Software that Generalizes

Start: Thursday, March 22, 2018, 11:30 a.m.
End: Thursday, March 22, 2018, 12:20 p.m.
Location: 319 McGregory Hall

Speaker: Tom Helmuth, Assistant Professor, Hamilton College

Abstract: Genetic programming can be used to automatically synthesize programs from scratch. Inspired by biological evolution, genetic programming uses a population of programs that undergo generations of selection and reproduction to search for a program that passes the given specifications. In particular, we use specifications based on input/output example test cases showing the desired behavior of the program. Unfortunately, programs often evolve that pass all given test cases, but fail to generalize to unseen data. I will show how we can use an automatic simplification technique to remove unuseful code from evolved programs. While this technique was originally designed to make programs easier to understand, I have found that it has the added benefit of often creating programs that generalize better than the original programs. I will present results showing significant improvements in generalization made by automatic simplification and will explain why this technique should be adopted more widely in genetic programming.