Related Experiment Videos
Code growth, explicitly defined introns, and alternative selection schemes
1Computer Science Department, City University, London, UK. peters@soi.city.ac.uk
Evolutionary Computation
|February 25, 1999
Summary
Genetic programming introns contribute to code growth but are not the sole cause. New methods minimize bloat by understanding and controlling intron behavior, improving evolutionary algorithm performance and parsimony.
Area of Science:
- Computer Science
- Evolutionary Computation
- Artificial Intelligence
Background:
- Genetic programming (GP) often suffers from code bloat, leading to inefficient program representations.
- Introns, non-coding sequences in genetic material, have been hypothesized to influence code growth in GP.
- Previous research suggests introns play a role in code growth, but their exact mechanisms and impact require further investigation.
Purpose of the Study:
- To experimentally investigate the role of explicitly defined introns in code growth within tree-based genetic programming.
- To understand the specific behaviors of introns that contribute to code bloat.
- To develop and evaluate methods for minimizing code bloat by managing intron activity and employing alternative evolutionary strategies.
Main Methods:
- Introduction of explicitly defined introns into tree-based genetic programming representations.
- Systematic negation of various intron behaviors to isolate their effects on code growth.
- Examination of alternative selection schemes and recombination operators.
- Comparative analysis of new methods against standard selection techniques.
Main Results:
- Introns were confirmed to contribute to code growth, but they are not the exclusive cause.
- Removal of introns reduced the rate of code growth but did not eliminate it entirely.
- A novel system was developed that effectively minimizes unnecessary code bloat.
- Alternative selection and recombination methods demonstrated improvements in both performance and parsimony compared to standard methods.
Conclusions:
- Intron behavior is a significant factor in genetic programming code growth, necessitating targeted management.
- Understanding and controlling introns, alongside optimizing selection and recombination, is crucial for developing parsimonious and efficient genetic programs.
- The developed system offers a viable approach to mitigating code bloat in genetic programming, enhancing its practical applicability.