Scaled Conjugate Gradient Learning for Complex-Valued Neural Networks

Abstract

In this paper, we present the full deduction of the scaled conjugate gradient method for training complex-valued feedforward neural networks. Because this algorithm had better training results for the real-valued case, an extension to the complex-valued case is a natural way to enhance the performance of the complex backpropagation algorithm. The proposed method is exemplified on well-known synthetic and real-world applications, and experimental results show an improvement over the complex gradient descent algorithm.