Semantic Search-Based Genetic Programming and the Effect of Intron Deletion

Area of Science:

• Computer Science
• Artificial Intelligence
• Evolutionary Computation

Background:

• Semantics, defined as input-output behavior of solutions, has gained traction in genetic programming (GP).
• Existing GP systems often struggle with complex optimization tasks, especially non-continuous functions.

Purpose of the Study:

• To introduce a new GP system that utilizes semantic information to improve search effectiveness.
• To evaluate the performance of the proposed semantics-based GP system against standard and bacterial GP.

Main Methods:

• Developed a novel GP system that maintains a distribution of semantic behaviors.
• Biased the search towards solutions with semantics similar to the best-found solutions.
• Tested the system on a suite of functions, including non-continuous ones.

Main Results:

• The semantics-based GP system demonstrated superior performance compared to standard GP and bacterial GP.
• Significant improvements were observed on non-continuous test functions.
• Generated solutions were often larger and contained more introns, but intron deletion did not impact performance.

Conclusions:

• The proposed semantics-based GP approach enhances search effectiveness, especially for challenging optimization problems.
• The system's ability to handle non-continuous functions indicates its robustness.
• The presence of introns in generated solutions does not hinder performance, suggesting a tolerance for code redundancy.