my picture

Anthony Awuley Machine Learning and Software Engineer

12 Mar

Using Computer Vision in Genetic Programming to Compare Linear and Tree Genome Representations


This work uses GP to identify specific features on an image using linear and tree genome representations. Grammatical Evolution (GE) is used to represent linear GP using a BNF (Backus Naur Form) grammar. Tree-genome however, is used to represents GP individuals in the form of binary trees. Panmictic model was used to implement the tree-based GP. A non- parametric analysis is performed on results of the computer vision problem to measure statistical significance between both representations at 95% significance level. Mann-Whitney test performed on the results show that linear-genome representation (using GE) outperformed tree-genome representation and is further supported by other key performance indicators.
Linear Representation: Grammatical Evolution in Java
Tree Representation: Evolutionary Computation in Java
Data Processing: Java + Shell Scripting

Leave a Comment