Enhancement of Quantum Particle Swarm Optimization in Elman Recurrent Network With Bounded Vmax Function

Abstract

There are many drawbacks in BP network, such as trap into local minima and may get stuck at regions of a search space. To solve these problems, Particle Swarm Optimization (PSO) has been executed to improve ANN performance. In this study, we exploit errors optimization of Elman Recurrent Neural Network (ERNN) with a new enhance method of Particle Swarm Optimization with an addition of quantum approach to optimize the performance of both networks with bounded Vmax function. Main characteristics of Vmax function are to control the global exploration of particles in Particle Swarm Optimization and Quantum approach is used to improve the searching ability of the individual particle of PSO. The results show that for cancer dataset, Quantum Particle Swarm Optimization in Elman Recurrent Neural Network (QPSOERN) with bounded Vmax of hyperbolic tangent depicted 96.26% and Vmax sigmoid function with 96.35% which both furnishes promising outcomes and better value in terms of classification accuracy and convergence rate compared to bounded standard Vmax function with only 90.98%.

An Improved Back Propagation Leaning Algorithm Using Second Order Methods with Gain Parameter

“…With the added of gain ( n c ) the parameter for CGPRis (19) In terms of minimization search direction, this method minimize along line between current point and the previous point by using line search technique with the added parameter of learning rate. Lastly, the convergence test is evaluated based on some indicators.…”

“…Lastly, the convergence test is evaluated based on some indicators. For the Quasi-Newton to perform reducing in error, the error with respect to weight which is greater than the convergence tolerance will then indicate that the testing is finished as represent in Equation (19).…”

“…Meanwhile, Yu and Wilamowski [17] demonstrated that the used of second order algorithms such as Newton algorithm and Levernberg Marquardt (LM) algorithm also can produced better result which can converge faster than using first order algorithms. Furthermore, Aziz et al [18,19] also produced a very good result of using particle swarm optimization in training Elman neural network.…”