Generalization Characteristics of Complex-Valued Feedforward Neural Networks in Relation to Signal Coherence

被引:239
作者
Hirose, Akira [1 ]
Yoshida, Shotaro [1 ]
机构
[1] Univ Tokyo, Dept Elect Engn & Informat Syst, Tokyo 1138656, Japan
关键词
Complex-valued neural network; function approximation; generalization; supervised learning; CARRIER-FREQUENCY MODULATION; BACKPROPAGATION ALGORITHM; DOMAIN;
D O I
10.1109/TNNLS.2012.2183613
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Applications of complex-valued neural networks (CVNNs) have expanded widely in recent years-in particular in radar and coherent imaging systems. In general, the most important merit of neural networks lies in their generalization ability. This paper compares the generalization characteristics of complex-valued and real-valued feedforward neural networks in terms of the coherence of the signals to be dealt with. We assume a task of function approximation such as interpolation of temporal signals. Simulation and real-world experiments demonstrate that CVNNs with amplitude-phase-type activation function show smaller generalization error than real-valued networks, such as bivariate and dual-univariate real-valued neural networks. Based on the results, we discuss how the generalization characteristics are influenced by the coherence of the signals depending on the degree of freedom in the learning and on the circularity in neural dynamics.
引用
收藏
页码:541 / 551
页数:11
相关论文
共 26 条
[1]   Blur identification by multilayer neural network based on multivalued neurons [J].
Aizenberg, Igor ;
Paliy, Dmitriy V. ;
Zurada, Jacek M. ;
Astola, Jaakko T. .
IEEE TRANSACTIONS ON NEURAL NETWORKS, 2008, 19 (05) :883-898
[2]   ON THE COMPLEX BACKPROPAGATION ALGORITHM [J].
BENVENUTO, N ;
PIAZZA, F .
IEEE TRANSACTIONS ON SIGNAL PROCESSING, 1992, 40 (04) :967-968
[3]   A COMPLEX MAPPING NETWORK FOR PHASE SENSITIVE CLASSIFICATION [J].
BIRX, DL ;
PIPENBERG, SJ .
IEEE TRANSACTIONS ON NEURAL NETWORKS, 1993, 4 (01) :127-135
[4]   COMPLEX-DOMAIN BACKPROPAGATION [J].
GEORGIOU, GM ;
KOUTSOUGERAS, C .
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-ANALOG AND DIGITAL SIGNAL PROCESSING, 1992, 39 (05) :330-334
[5]   Nonlinear adaptive prediction of complex-valued signals by complex-valued PRNN [J].
Goh, SL ;
Mandic, DP .
IEEE TRANSACTIONS ON SIGNAL PROCESSING, 2005, 53 (05) :1827-1836
[6]  
Hara T, 2004, NEURAL NETWORKS, V17, P1201, DOI 10.1016/j.nuenet.2004.07.012
[7]   CONTINUOUS COMPLEX-VALUED BACKPROPAGATION LEARNING [J].
HIROSE, A .
ELECTRONICS LETTERS, 1992, 28 (20) :1854-1855
[8]   APPLICATIONS OF COMPLEX-VALUED NEURAL NETWORKS TO COHERENT OPTICAL COMPUTING USING PHASE-SENSITIVE DETECTION SCHEME [J].
HIROSE, A .
INFORMATION SCIENCES-APPLICATIONS, 1994, 2 (02) :103-117
[9]  
Hirose Akira, 2011, Frontiers of Electrical and Electronic Engineering in China, V6, P171, DOI 10.1007/s11460-011-0125-3
[10]   Behavior control of coherent-type neural networks by carrier-frequency modulation [J].
Hirose, A ;
Eckmiller, R .
IEEE TRANSACTIONS ON NEURAL NETWORKS, 1996, 7 (04) :1032-1034