Processing Random Data with Bayesian Perceptron with Probabilistic Outputs

Nieto-Chaupis H. · 2025 IEEE 16th Annual Ubiquitous Computing, Electronics and Mobile Communication Conference, UEMCON 2025 (2025) · DOI: 10.1109/uemcon67449.2025.11267577

Resumen

We present a theoretical scheme of artificial neural networks concretely based in conditional probabilities and whose internal processing is supported through stochastic techniques. By using the Bayes's rule a theory is constructed. By harnessing this Bayesian approach some numerical tests are done. Thus, given a set of random points optimized by a Monte Carlo mechanism, polynomial distributions have been constructed. Thus, when them enter in a master Bayesian formulation, various probabilities have been derived as function of a variable that is defined by user and it is attached to input function of perceptron. Finally, perceptron appears to exhibit up to two well-defined phases indicating the fact that although exists random data, artificial neural networks provide capabilities to syntonize the data processing inside the most convenient available numerical ranges.

Instituciones: Universidad Autónoma del Perú