Switch-Time neurons. Definition and types
Pavel Stoynov · 2023
Neural networks are used in forecasting or predicting given output variable based on information provided by a set of input variables. There are also other types of models used for the same purpose: linear regression models, non-linear regression-based models, polynomial approximation etc.While linear and non-linear regression models are parametric models, neural networks, like polynomial approximation models, are semi-parametric. Neural networks models, in contrast with polynomial semi-parametric models, don’t provide closed-form solution, so only approximate solution can be provided by the neural networks.In this article, we consider the Switch-time neuron based on the so called Switch-time (ST) distribution [4]. More specifically, we consider two types of Switch-time neurons: one-side Switch-time neuron (ST neuron) and double-side Switch-time neuron (DST neuron) and present R code for simulating different examples of both types of neurons.As the concept neuron is close to the concept regression, we may by analogy consider Switch-time (stopit) regression [5] of both types.