Synthesizing a program with the desired input-output behavior by means of genetic programming is an iterative process that needs appropriate guidance. That guidance is conventionally provided by a fitness function that measures the conformance of program output with the desired output. Contrary to widely adopted stance, there is no evidence that this quality measure is the best choice; alternative search drivers may exist that make search more effective. This study proposes and investigates a new family of behavioral search drivers, which inspect not only final program output, but also program behavior meant as the partial results it arrives at while executed.
@INPROCEEDINGS { krawiec2:2014:EuroGP,
ABSTRACT = { Synthesizing a program with the desired input-output behavior by means of genetic programming is an iterative process that needs appropriate guidance. That guidance is conventionally provided by a fitness function that measures the conformance of program output with the desired output. Contrary to widely adopted stance, there is no evidence that this quality measure is the best choice; alternative search drivers may exist that make search more effective. This study proposes and investigates a new family of behavioral search drivers, which inspect not only final program output, but also program behavior meant as the partial results it arrives at while executed. },
ADDRESS = { Granada, Spain },
AUTHOR = { Krzysztof Krawiec and Una-May O'Reilly },
BOOKTITLE = { Genetic Programming },
EDITOR = { Miguel Nicolau and Krzysztof Krawiec and Malcolm Heywood },
KEYWORDS = { Genetic programming, Program synthesis, Semantics, Program behavior },
LOCATION = { Heidelberg },
MONTH = { 23-25 April },
NOTES = { Proceedings of the 17th European Conference on Genetic Programming, EuroGP 2014 },
ORGANISATION = { EvoStar },
PAGES = { 210--221 },
PUBLISHER = { Springer },
SERIES = { Lecture Notes in Computer Science },
TITLE = { Behavioral Search Drivers for Genetic Programing },
VOLUME = { 8599 },
YEAR = { 2014 },
}