Research on Compound Numerical Spiking Neural P System
Meng Hu, Xiantai Gou, Fang Deng, Qifen Liu and Haina Rong
The Spiking Neural P System (SNPS) is a neural computing model inspired by the mechanism of biological neurons transmitting information. It has received extensive attention from scholars due to its powerful computing power and brain-like information transmission mode. Its application potential is huge, but in the field of pattern recognition it has not been completely solved. This paper introduces the Markov decision-making process and draws on the Q-Learning in reinforcement learning to design a compound Numerical Spiking Neural P System (NSNPS) with multiple production functions. This learning model has the ability to learn and select the execution of the production function. Through the XOR problem and the IRIS dataset species identification, the effectiveness of the algorithm in the discrete and the continuous state is verified respectively. The experimental results show that the recognition accuracy of NP problem is 100%, and the recognition accuracy of IRIS dataset reaches 98.6%. The model has the ability to solve nonlinear recognition problems, and the compound neuron in NSNPS has a strong computing power. Compared with the traditional CNN network, it uses fewer neurons and has a certain development potential.
Keywords: Membrane computing, Numerical Spiking Neural P system, pattern recognition, compound neuron, reinforcement learning