Ghoulbeigi_Elmira.pdf (615.47 kB)

Indirect Estimation Of Distribution Algorithms For The Evolution Of Tree-Shaped Structures

Download (615.47 kB)
thesis
posted on 22.05.2021, 13:20 by Elmira Ghoulbeigi
This thesis explores indirect estimation of distribution algorithms (IEDAs) for the evolution of tree structured expressions. Unlike conventional estimation of distribution algorithms, IEDAs maintain a distribution of the genotype space and indirectly search the solution space by performing a genotype-to-phenotype mapping. In this work we introduce two IEDAs named PDPE and N-gram GEP. PDPE induces a population of programs, encoded as fixed-length gene expression programming (GEP) chromosomes, by iteratively refining and randomly sampling a probability distribution of program instructions. N-gram GEP attempts to capture regularities in GEP chromosomes by sampling the probability distribution of triplet of instructions (3-grams). We tested the performance of these systems using a variety of non-trivial test problems, such as symbolic regression and the lawn-mower problem. We compared PDPE and N-gram GEP with their predecessors, probabilistic incremental program evolution (PIPE) and N-gram GP, and the canonical GEP algorithm. The results proved that our methodology is more efficient than PIPE and the canonical GEP algorithm.

History

Language

eng

Degree

Master of Science

Program

Computer Science

Granting Institution

Ryerson University

LAC Thesis Type

Thesis