Introduction

This chapter reviews the two artificial neural network models of multilayer perceptron and probabilistic neural networks. Multilayer perceptron neural networks have been the most widely used artificial neural network model to date and the core principle for the currently dominating deep learning approaches. The neural network models, however, inherently suffer from various numerically oriented problems, such as many hyperparameters to tune, being stuck in local minima during the iterative network parameter optimization (i.e., training), a non-straightforward manner of performing incremental training, or difficulty in the network analysis obtained after the training. These drawbacks are primarily ascribed to the distributed manner of data representation. In contrast, the data representation taken by a probabilistic neural network is local, thus being essentially free from these issues. While the significance of the probabilistic neural network model has yet to be discovered by the community, the model has great potential to provide robust and straightforward approaches to various problems of interest in the pattern classification domain. This chapter, therefore, reveals the beneficial properties of the probabilistic neural network compared to the conventional multilayer neural network models.