Learning of Phase-Amplitude-Type Complex-Valued Neural Networks with Application to Signal Coherence Conference Paper

abstract

• Springer International Publishing AG 2017. This paper presents a limited-memory BFGS (L-BFGS) based learning algorithm for complex-valued neural networks (CVNNs) with phase-amplitude-type activation functions, which can be applied to deal with coherent signals effectively. The performance of the proposed L-BFGS algorithm is compared with traditional complex-valued stochastic gradient descent method on the tasks of wave-related signal processing with various degrees of coherence. The experimental results demonstrate that both faster convergence speed and smaller training errors are achieved by our algorithm. Furthermore, the phase outputs of the CVNNs trained by this algorithm are more stable when white Gaussian noises are added to the input signals.

authors

• Huang, Tingwen

published proceedings

• NEURAL INFORMATION PROCESSING, ICONIP 2017, PT I

author list (cited authors)

• Wu, R., Huang, H. e., & Huang, T.

citation count

• 4

complete list of authors

• Wu, Rongrong||Huang, He||Huang, Tingwen

publication date

• October 2017

publisher

• Springer Nature  Publisher

published in

• Lecture Notes in Artificial Intelligence  Journal

keywords

• Coherent Signals
• Complex-valued Neural Networks
• Limited-memory Bfgs Algorithm
• Phase-amplitude-type

Digital Object Identifier (DOI)

• 10.1007/978-3-319-70087-8_10

International Standard Book Number (ISBN) 13

• 978-3-319-70086-1

start page

• 91

end page

• 99

volume

• 10634