Bayesian Probabilities from Topological Scenarios of Artificial Neural Networks

Nieto-Chaupis H. · International Conference on Electrical, Computer, and Energy Technologies, ICECET 2025 (2025) · DOI: 10.1109/icecet63943.2025.11472394

Resumen

Conditional probabilities aligned to theorem of Bayes from a family of perceptron topologies are derived. To accomplish this, inputs are considered as random numbers. Also, weights are generalized and have been proposed as negative exponential functions. Topologies have been generated as function of weight. An extra operation was carried out through action of nonlinear operators. Activation function under this context acquires the meaning of confirmation. Since resulting output is paired to them of inputs, then it is feasible to claim that whole procedure becomes a topological mechanism that ends at Bayesian probabilities. The complete theoretical scheme is illustrated by some numerical examples. The case of pandemic is emphasized and discussed.

Instituciones: Universidad Autónoma del Perú