A Relative Comparison of Training Algorithms in Artificial Neural Network

Abstract

Artificial Neural Networks (ANNs) give a practical, general method for learning discrete-valued, real-valued, and vector-valued functions from examples. The algorithms used for training models aim to construct an optimization framework and apprehend the parameters in the target function from the provided dataset. ANN learning is vigorous to errors in the training data which has been successfully applied to scenarios like speech recognition, interpreting visual scenes and robotics. This paper basically aims to provide an experimental study to compare various optimization or training algorithms and determines the best suited optimization method corresponding to a particular dataset in terms of accuracy and loss. To the best of our knowledge, this paper is the first to consider different learning rate values to study the comparison between different training or optimization algorithms.