Automatic Design of Spiking Neural P Systems Based on Genetic Algorithms
Jianping Dong, Michael Stachowicz, Gexiang Zhang, Matteo Cavaliere, Haina Rong and Prithwineel Paul
At present, all known spiking neural P systems (SN P systems) are established by manual design rather than automatic design. The method of manual design poses two problems: consuming a lot of computing time and making unnecessary mistakes. In this paper, we propose an automatic design approach for SN P systems by genetic algorithms. More specifically, the regular expressions are changed to achieve the automatic design of SN P systems. In this method, the number of neurons in system, the synapse connections between neurons, the number of rules within each neuron and the number of spikes within each neuron are known. A population of SN P systems is created by generating randomly accepted regular expressions. A genetic algorithm is applied to evolve a population of SN P systems toward a successful SN P systems with high accuracy and sensitivity for carrying out specific task. An effective fitness function is designed to evaluate each candidate SN P system. In addition, the elitism, crossover and mutation are also designed. Finally, experimental results show that the approach can successfully accomplish the automatic design of SN P systems for generating natural numbers and even natural numbers by using the .NET framework.
Keywords: Spiking neural P systems, genetic algorithm, fitness function, mutation probability, membrane computing