Genetic Programming with Local Scoring

TL;DR

This paper introduces novel genetic programming techniques involving local scoring, suppose-expressions, and cyclic evolution, enabling more precise program synthesis and handling complex functions beyond existing methods.

Contribution

It presents new mutation-based methods, including local scoring and cyclic evolution, improving the ability to synthesize correct programs for complex functions.

Findings

Successfully evolved correct code for integer and list functions

Outperformed some existing genetic programming techniques

Demonstrated effectiveness on intractable functions

Abstract

We present new techniques for synthesizing programs through sequences of mutations. Among these are (1) a method of local scoring assigning a score to each expression in a program, allowing us to more precisely identify buggy code, (2) suppose-expressions which act as an intermediate step to evolving if-conditionals, and (3) cyclic evolution in which we evolve programs through phases of expansion and reduction. To demonstrate their merits, we provide a basic proof-of-concept implementation which we show evolves correct code for several functions manipulating integers and lists, including some that are intractable by means of existing Genetic Programming techniques.