Semantic analysis of program initialisation in genetic programming

Abstract

Population initialisation in genetic programming is both easy, because random combinations of syntax can be generated straightforwardly, and hard, because these random combinations of syntax do not always produce random and diverse program behaviours. In this paper we perform analyses of behavioural diversity, the size and shape of starting populations, the effects of purely semantic program initialisation and the importance of tree shape in the context of program initialisation. To achieve this, we create four different algorithms, in addition to using the traditional ramped half and half technique, applied to seven genetic programming problems. We present results to show that varying the choice and design of program initialisation can dramatically influence the performance of genetic programming. In particular, program behaviour and evolvable tree shape can have dramatic effects on the performance of genetic programming. The four algorithms we present have different rates of success on different problems.

The Effect of Distinct Geometric Semantic Crossover Operators in Regression Problems

“…Although methods aiming at GP diversity are not new, they usually considered only syntactic diversity. In [4], in turn, the authors studied the impact of semantic diversity during population initialization, showing that greater diversity leads to improved results. Indirect semantic methods, on the other hand, use regular GP operator, but only accept individuals if they respect some semantic-related criteria, such as their semantic difference to their parents [3] or to a geometric (semantically intermediate) individual of their parents [6].…”

Semantic-Based Subtree Crossover Applied to Dynamic Problems

“…The method improved GP performance, presumably because it increased semantic diversity. The method has also been applied to mutation [3] and to the initialisation phase of GP [2].…”

Progress properties and fitness bounds for geometric semantic search operators

“…This simple formalism, known as sampling semantics or simply semantics, provides a relatively detailed account on program execution at no additional computational cost. Previous studies have shown that, even though sampling semantics cannot convey information on all aspects of program execution, it can be leveraged to design efficient semantic-aware search operators [8,13,23,26,27,30,31], population initialization [2,7] and selection techniques [6].…”