Learning Spiking Neural P System with Error Propagation
Dorian Lazăr
Artificial neural networks of first generation have dominated the area of learning systems with many state-of-the-art methods on tasks that were previously thought to be solvable only by humans. Among these types of neural networks, one of the simplest type is that of a Multi-Layer Perceptron (MLP). Spiking Neural P systems (SN P systems) are another class of artificial neural networks of third generation that are inspired by the cell membrane mechanisms and how they communicate.
We propose in this paper a new type of SN P system, called Learning SN P system with Error Propagation (LSNPEP system), with an architecture similar to that of a Layered Spiking Neural P system (LSN P system). The LSNPEP system uses the membrane computing mechanisms and their rules to learn in a way which is similar to MLPs and therefore, showing that LSNPEP systems can learn anything MLPs can do.
In this paper we describe the LSNPEP system showing its similarities, but also its main extensions making it slightly more general than LSN P system, proving its superiority in terms of accuracy (on the testing side) not only on LSN P system models, but on five more other approaches on eight different classification datasets including tabular data and images. It surpasses by a large margin the LSN P system on the digit recognition dataset.
Keywords: Machine learning, data classification, spiking neural networks, spiking neural P systems, layered spiking neural P systems, supervised learning
DOI: 10.32908/ijuc.v21.zhang07