Mengjie Zhang, Intelligent Crossover in Genetic Programming
Intelligent Crossover in Genetic Programming
- Overview:
This project concerns genetic operator, building blocks and their relationship.
Crossover was considered the most important genetic operator in genetic programming (GP) and was claimed the "central storm" of GP. However, recent research shows that this operator can be very powerful at the beginning of evolution, but seems less powerful in later stage.
In the standard crossover operator, two parent programs are selected from the population based on their fitness. Then a crossover point was randomly chosen in each of the program and swap the two programs at the two crossover points. While this operator is easy to implement, it might be possible to destroy some good portion of programs (sub programs), which are potential building blocks. It might be also possible for this operator to combine two small building blocks (subprograms) together to form a large building blocks.
- Tasks and Goals of this Project:
The goal of this project is to investigate intelligent crossover operators in genetic programming to extend the power of the crossover operator and to preserve good building blocks. Specifically, this project will investigate:
- Whether the standard crossover operator preserve building blocks or disrupt good building blocks to discover the relationship between building blocks and the crossover operator.
- How to introduce intelligence to the standard crossover operator to develop new intelligent crossover operators which can preserve building blocks and improve the effectiveness of GP.


