Neural network crossover in genetic algorithms using genetic programming

Neural network crossover in genetic algorithms using genetic programming

21 February 2024 | Kyle Pretorius, Nelishia Pillay
This paper explores the use of genetic programming (GP) to evolve crossover operators for neural networks (NNs) in genetic algorithms (GAs). The authors address the common belief that crossover operators are destructive and detrimental to GA performance, particularly when used with NN weights. They propose a novel GP to automatically design both reusable and disposable crossover operators, comparing their efficiency. Experiments are conducted to evaluate the performance of GAs using no crossover, a commonly used human-designed crossover operator, and GAs using GP-evolved crossover operators. The results show that GP-evolved disposable crossover operators significantly improve the results obtained from the GA, demonstrating that crossover operators can be effective and beneficial when designed and applied correctly. The study also highlights the importance of including crossover operators in GAs for NN weight optimization, challenging the traditional view of their destructive nature.This paper explores the use of genetic programming (GP) to evolve crossover operators for neural networks (NNs) in genetic algorithms (GAs). The authors address the common belief that crossover operators are destructive and detrimental to GA performance, particularly when used with NN weights. They propose a novel GP to automatically design both reusable and disposable crossover operators, comparing their efficiency. Experiments are conducted to evaluate the performance of GAs using no crossover, a commonly used human-designed crossover operator, and GAs using GP-evolved crossover operators. The results show that GP-evolved disposable crossover operators significantly improve the results obtained from the GA, demonstrating that crossover operators can be effective and beneficial when designed and applied correctly. The study also highlights the importance of including crossover operators in GAs for NN weight optimization, challenging the traditional view of their destructive nature.
Reach us at info@study.space