Comparison of Neural Network Training Algorithms for the prediction of the patient’s post-operative recovery area

Abstract

An Artificial Neural Network(ANN) is a well known universal approximator to model smooth and continuous functions. ANNs operate in two stages: learning and generalization. Learning of a neural network is to approximate the behavior of the training data while generalization is the ability to predict well beyond the training data. In order to have a good learning and generalization ability , a good training algorithm is needed. Training a neural network can be treated as a nonlinear mathematical optimization problem and different algorithms can have quite different effects on the training result. As a result, training with different algorithms and repeating with multiple random initial weights can be helpful in getting a better solution to the neural network training problem. In addition to the popular basic back propagation training algorithm, many other algorithms are available. These include conjugate gradient descent, quasi-Newton, and Levenberg-Marquardt etc. This paper presents a novel comparison of different algorithms for the prediction of the patient's postoperative recovery area.

Artificial neural network modeling of the water quality index for Kinta River (Malaysia) using water quality variables as predictors

“…The former algorithm frequently produced lower testing set errors and higher R 2 values than the latter. This finding agrees with empirical comparisons in the literature which evidenced that the QP training algorithm is one of the fastest, most reliable algorithms and that it outperforms the majority of the rest heuristic variants of the BP algorithm on a broad range of modeling problems (Fahlman, 1988;Shanthi et al, 2009). Consequently, all networks were trained with the standard QP algorithm due to its fast convergence and stability besides its better performance on the studied data than the BP algorithm.…”

Performance comparison of second order conjugate algorithms in neural networks for predictive data mining

“…Newton's method computes the second order derivatives of Taylor's approximation as hessian matrix and therefore finds out the point of minimum much faster than first order methods (Shanthi et al, 2009;Yu and Wilamowski, 2012). This utilizes the curvature information to search a more direct route towards the point of minima.…”

Development and Comparative Study of Effects of Training Algorithms on Performance of Artificial Neural Network Based Analog and Digital Automatic Modulation Recognition

“…For instance, SCG was chosen because of its modest memory requirements with high accuracy and speed due to inexpensive calculation of the gradient information [33]. Similarly, the CONJGRAD training algorithm was chosen because it is also known to be a fast training algorithm with numerical efficiency and very low memory requirement [34].…”