Conjugate gradient (CG) method has been verified to be one effective strategy for training neural networks in terms of its low memory requirements and fast convergence. In this paper, an efficient CG method is proposed to train fully complex neural networks based on Wirtinger calculus. We adopt two ways to enhance the training performance. One is to construct a sufficient descent direction during training by designing a fine tuning conjugate coefficient. Another technique is to pursue the optimal learning rate instead of a fixed constance in each iteration which is determined by employing a generalized Armijo search. To verify the effectiveness and the convergent behavior of the proposed algorithm, the illustrated simulation has been performed on the complex benchmark noncircular signal.
CITATION STYLE
Zhang, B., Wang, J., Wu, S., Wang, J., & Zhang, H. (2018). Fully Complex-Valued Wirtinger Conjugate Neural Networks with Generalized Armijo Search. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 10956 LNAI, pp. 123–133). Springer Verlag. https://doi.org/10.1007/978-3-319-95957-3_14
Mendeley helps you to discover research relevant for your work.