A Comparison of RBF Neural Network Training Algorithms for Inertial Sensor Based Terrain Classification

Abstract

This paper introduces a comparison of training algorithms of radial basis function (RBF) neural networks for classification purposes. RBF networks provide effective solutions in many science and engineering fields. They are especially popular in the pattern classification and signal processing areas. Several algorithms have been proposed for training RBF networks. The Artificial Bee Colony (ABC) algorithm is a new, very simple and robust population based optimization algorithm that is inspired by the intelligent behavior of honey bee swarms. The training performance of the ABC algorithm is compared with the Genetic algorithm, Kalman filtering algorithm and gradient descent algorithm. In the experiments, not only well known classification problems from the UCI repository such as the Iris, Wine and Glass datasets have been used, but also an experimental setup is designed and inertial sensor based terrain classification for autonomous ground vehicles was also achieved. Experimental results show that the use of the ABC algorithm results in better learning than those of others.

Artificial Bee Colony Training of Neural Networks

“…Similar success with ABC trained neural networks 4 has also been claimed by numerous other authors [16,17,18,22,23], and further improved results have been obtained with hybrid learning algorithms involving the ABC combined with more traditional neural network training algorithms [9,19,21]. The key question to be addressed in this paper is: how can these good ABC results be reconciled with the earlier negative results that CantuPaz and Kamath obtained for the closely related population-based EAs [4]?…”

“…This is clearly not preventing the ABC algorithm from finding good solutions, but, together with the finding that the scout bees are not making any useful contribution, it does mean that the ABC is actually performing little more than stochastic hill climbing, which one would expect to end up with similar results to an informed hill climbing algorithm like BP, albeit more slowly. It also means that previous claims that the ABC can avoid becoming stuck in local optima better than BP [15,16,17,22] could well prove unfounded too.…”

Learning of RBF network by using Teaching-Learning Based Optimization (TLBO) Algorithm

“…Here the nature inspired, population based algorithms like evolutionary and swarm intelligence algorithms are coming into picture. Tuba Kurban et al [14], observed in their comparison of RBF neural network training algorithm for inertial sensor based terrain classification, conventional algorithms like gradient descent and Kalman filtering (both are derivative based) have some weakness such as converging to a local minima and time-consuming process of finding the optimal gradient. Salman Mohaghehi et al [15], compared PSO and back propagation algorithm for training RBF neural networks for identification of a power system with statcom, concluded PSO algorithm has shown to have several advantages, both in terms of robustness and the efficiency in finding the optimal weights for the RBFN neuroidentifier.…”

Radial basis function neural networks: a topical state-of-the-art survey

“…During implementation, the nal weights are used to classify the presence of hidden information. Kurban and Besdok [162], have made a comparison of training algorithms of RBFNs for classi cation purposes. For training they have used Arti cial Bee Colony (ABC) algorithm, Genetic algorithm, Kalman ltering algorithm, and gradient descent algorithm.…”