The performance comparisons of backpropagation algorithm's family on a set of logical functions
Akaraphunt Vongkunghae, Anuchit Chumthong, Non-members · 2007
This paper presents the performance comparisons between the training algorithms, Gradient Descent Backpropagation (GD), Gradient Descent Backpropagation with Momentum (GDM), Resilient Backpropagation (RP), and Levenberg-Marquardt Backpropagation (LM). The algorithms are used to train feedforward artiflcial neural networks. The training data are logic functions, or, exclusive or, 3 bit parity, and 5 bit counting. The converged training number, converged time, network size, and ratio of output squared-error sum to input squared-error sum (OSISE) are the results that are arranged in tables to demonstrate the performance of each backpropagation algorithm.