基于累積量的兩層前饋神經(jīng)網(wǎng)絡(luò)盲辨識(shí)

戴憲華

基于累積量的兩層前饋神經(jīng)網(wǎng)絡(luò)盲辨識(shí)

戴憲華

計(jì)量
- 文章訪問(wèn)數(shù): 2220
- HTML全文瀏覽量: 125
- PDF下載量: 465
- 被引次數(shù): 0
出版歷程
- 收稿日期: 1999-09-13
- 修回日期: 2000-07-07
- 刊出日期: 2002-01-19

Cumulant-bansed blind identification of two-layer feedforward neural networks

Dai Xianhua

摘要

摘要: 由于非線性系統(tǒng)輸出是其參數(shù)的非線性函數(shù),直接利用高階累積量辨識(shí)兩層前饋神經(jīng)網(wǎng)絡(luò)(FNN)通常是十分困難的。為解決這一問(wèn)題,該文提出兩種基于四階累積量的FNN辨識(shí)方法。第一種方法,FNN的隱元在其輸入空間利用多個(gè)線性系統(tǒng)近似,進(jìn)而FNN利用一統(tǒng)計(jì)模型混合專(zhuān)家(ME)網(wǎng)絡(luò)重新描述?；贛E模型,FNN參數(shù)可利用統(tǒng)計(jì)期望值最大化(EM)算法獲得估計(jì)。第二種方法,為簡(jiǎn)化FNN的ME模型,引入隱含觀測(cè)量?；陔[含觀測(cè)量估計(jì),FNN被分解為多個(gè)單隱元的訓(xùn)練問(wèn)題,進(jìn)而整體FNN可利用一兩階層ME描述。基于單隱元的參數(shù)估計(jì),FNN可利用一具有更快收斂速度的簡(jiǎn)化算法獲得估計(jì)。
- ME模型; EM算法; 基于累積量的系統(tǒng)辨識(shí)
Abstract: It is always difficult to train a two-layer Feedforward Neural Networks (FNN) based on the cumulant match criterion because cumulants are nonlinear implicit functions of the FNN parameters. In this work, two new cumulant-based training methods for two-layer FNN are developed. In the first method, the hidden units of two-layer FNN are approximated with multiple linear systems, and further total FNN is modeled with a mixture of experts (ME) architecture. With the ME model, FNN parameters are estimated with expectation-maximization (EM) algorithm. The second method, for simplifying the two-layer FNN statistical model, proposes a simplified two-level hierarchical ME to remodel the FNN, in which hidden variables are introduced such that training total FNN is changed to training a set of single neurons. Based on training single neuron, total FNN is estimated in a simplified version with a faster convergence speed.

HTML全文

參考文獻(xiàn)(1)

J.M. Mendel, Tutorial on high-order statistics (spectra) in signal processing and system theory,theoretic results and some applications, Proc.[J]. IEEE.1991,79:278-[2]Hames A. Cadzow, Blind deconvolution via cumulant extrema, IEEE Signal Processing Magazine,1996, 13(3), 24-42.[3]D. Hatzinakos, C. L. Nikias, Blind equalization using a tricespectrum-based algorithm, IEEE Trans. on Comm., 1991, COM-39(5), 669-681.[4]B. Port, B. Ftiedlander, Blind equalization of digital communication channels using higher-order moments, IEEE Trans. on ASSP, 1991, ASSP-39(2), 522-526.[5]J.K. Tugnait, Identification of linear stochastic systems via second and fourth-order cumulants matching, IEEE Trans. on Information Theory, 1987, 33(5), 393-407.[6]Jitendra K. Tugnait, Blind equalization and channel estimation with partial response input signals, IEEE Trans. on Comm., 1997, COM-45(9), 1025-1442.[7]Vojin, Zivojnvic, Minimum fisher information of moment-constrained distributions with application to robust blind identification.[J]. Signal Processing.1998,65:297-[8]Michael I. Jordan, Lei Xu, Convergence results for the EM approach to mixtures of experts architectures, Neural Networks, 1995, 8(9), 1409-1431.[9]S.I. Amari, Information geometry of the EM and EM algorithms for neural network, Neural Networks, 1995, 8(9), 1379-1408.[10]Sheng Ma, James Farmer, An efficient EM-based training algorithm for feedforward neural networks, Neural Networks, 1997, 10(2), 243-256.[11]M.I. Jordan, Hierarchical mixtures of experts and EM algorithm, Neural Computation, 1994,6(2), 181-241.[12]R.A. Jacobs, Adaptive mixtures of local experts, Neural Computation, 1991, 3(1), 79-87.[13]B. Widrow, 30 years of adaptive neual networks, perceptron, madaline, and backpropagation,Proc. IEEE, 1990, 78(9), 1415-1442.

相關(guān)文章

施引文獻(xiàn)

資源附件(0)

訪問(wèn)統(tǒng)計(jì)