Mengjie Zhang, Simplification of Genetic Programs During Evolution

Simplification of Genetic Programs During Evolution

• Overview:

  In Genetic Programming (GP), programs are usually expressed as trees. This form of GP has been successfully applied to a number of problems such as image enhancement, image recognition, and regression problems.

  However, a GP evolutionary process will inevitably introduce redundancy and unnecessary complexity into evolved programs. Typically, the programs are simplified at the end of the evolutionary process, but not during it. The redundancies and complexities have the undesirable consequences that the programs are very difficult to understand and interpret, and the search process is very inefficient because it explores unnecessarily complex parts of the search space. However, the redundant components of the evolving programs may play an important role in the evolutionary process by providing a wider variety of possible program fragments for constructing new programs.

• Tasks and Goals of this Project:

  The goal of this project is to investigate the effect of performing simplification of the programs during the evolutionary process, to discover whether the reduction in complexity outweighs the possible benefits of redundancy. Specifically, we will investigate:

  • Whether simple algebraic simplification methods can be used in GP during the evolutionary process;
  • Whether numeric hashing methods can be used in GP during the evolutionary process; and
  • Whether the GP simplification destroy good building blocks or not; and
  • Whether the GP simplification improves system performance or not.
  Some regression and classification tasks will be used as examples.

• More information can be seen from Meng's research projects . Related Publications can be viewed from here: Meng's recent publications. Contact me if you want to get a copy of those papers, or if you want to know more detail about this project.