Efficient design of complex-valued neural networks with application to the classification of transient acoustic signals

被引:1
作者
Paul, Vlad S. [1 ]
Nelson, Philip A. [1 ]
机构
[1] Univ Southampton, Inst Sound & Vibrat Res, Southampton SO17 1BJ, Hants, England
基金
英国工程与自然科学研究理事会;
关键词
ALGORITHM;
D O I
10.1121/10.0028230
中图分类号
O42 [声学];
学科分类号
070206 ; 082403 ;
摘要
A paper by the current authors Paul and Nelson [JASA Express Lett. 3(9), 094802 (2023)] showed how the singular value decomposition (SVD) of the matrix of real weights in a neural network could be used to prune the network during training. The paper presented here shows that a similar approach can be used to reduce the training time and increase the implementation efficiency of complex-valued neural networks. Such networks have potential advantages compared to their real-valued counterparts, especially when the complex representation of the data is important, which is the often case in acoustic signal processing. In comparing the performance of networks having both real and complex elements, it is demonstrated that there are some advantages to the use of complex networks in the cases considered. The paper includes a derivation of the backpropagation algorithm, in matrix form, for training a complex-valued multilayer perceptron with an arbitrary number of layers. The matrix-based analysis enables the application of the SVD to the complex weight matrices in the network. The SVD-based pruning technique is applied to the problem of the classification of transient acoustic signals. It is shown how training times can be reduced, and implementation efficiency increased, while ensuring that such signals can be classified with remarkable accuracy.
引用
收藏
页码:1099 / 1110
页数:12
相关论文
共 45 条
  • [1] Complex-Valued Signal Processing: The Proper Way to Deal With Impropriety
    Adali, Tulay
    Schreier, Peter J.
    Scharf, Louis L.
    [J]. IEEE TRANSACTIONS ON SIGNAL PROCESSING, 2011, 59 (11) : 5101 - 5125
  • [2] Amin MF, 2011, LECT NOTES COMPUT SC, V7062, P550, DOI 10.1007/978-3-642-24955-6_66
  • [3] Pruning algorithms of neural networks - a comparative study
    Augasta, M. Gethsiyal
    Kathirvalavakumar, T.
    [J]. OPEN COMPUTER SCIENCE, 2013, 3 (03) : 105 - 115
  • [4] Bassey J., 2021, arXiv
  • [5] ON THE COMPLEX BACKPROPAGATION ALGORITHM
    BENVENUTO, N
    PIAZZA, F
    [J]. IEEE TRANSACTIONS ON SIGNAL PROCESSING, 1992, 40 (04) : 967 - 968
  • [6] Singular Value Decomposition and Neural Networks
    Bermeitinger, Bernhard
    Hrycej, Tomas
    Handschuh, Siegfried
    [J]. ARTIFICIAL NEURAL NETWORKS AND MACHINE LEARNING - ICANN 2019: DEEP LEARNING, PT II, 2019, 11728 : 153 - 164
  • [7] Blalock D, 2020, Arxiv, DOI arXiv:2003.03033
  • [8] Brandwood D. H., 1983, IEE Proceedings H (Microwaves, Optics and Antennas), V130, P11, DOI 10.1049/ip-h-1.1983.0004
  • [9] Cai CH, 2014, LECT NOTES ARTIF INT, V8862, P820, DOI 10.1007/978-3-319-13560-1_65
  • [10] Pixel-Wise PolSAR Image Classification via a Novel Complex-Valued Deep Fully Convolutional Network
    Cao, Yice
    Wu, Yan
    Zhang, Peng
    Liang, Wenkai
    Li, Ming
    [J]. REMOTE SENSING, 2019, 11 (22)