STOCHASTIC INFORMATION GRADIENT ALGORITHM WITH GENERALIZED GAUSSIAN DISTRIBUTION MODEL
|
|
- Jonathan Tucker
- 5 years ago
- Views:
Transcription
1 Journal of Circuits, Systems, and Computers Vol. 2, No. (22) 256 (6 pages) #.c World Scientific Publishing Company DOI:.42/S STOCHASTIC INFORMATION GRADIENT ALGORITHM WITH GENERALIZED GAUSSIAN DISTRIBUTION MODEL BADONG CHEN,, JOSE C. PRINCIPE,, JINCHUN HU, and YU ZHU, ** Department of Electrical and Computer Engineering, University of Florida, Gainesville, FL 326, USA Department of Precision Instruments and Mechanology, Tsinghua University, Beijing 84, P. R. China Received 26 October 29 Accepted 3 August 2 Published 29 February 22 This paper presents a parameterized version of the stochastic information gradient (SIG) algorithm, in which the error distribution is modeled by generalized Gaussian density (GGD), with location, shape, and dispersion parameters. Compared with the kernel-based SIG (SIG- Kernel) algorithm, the GGD-based SIG (SIG-GGD) algorithm does not involve kernel width selection. If the error is zero-mean, the SIG-GGD algorithm will become the least mean p-power (LMP) algorithm with adaptive order and variable step-size. Due to its well matched density estimation and automatic switching capability, the proposed algorithm is favorably in line with existing algorithms. Keywords: Minimum error entropy criterion (MEE); stochastic information gradient (SIG) algorithm; generalized Gaussian density (GGD); least mean p-power (LMP).. Introduction Stochastic gradient (SG) algorithm under error criteria has been extensively studied and widely used in the last decades.,2 The least mean square (LMS) algorithm, which is based on the minimum mean-square error (MSE) criterion, is perhaps the most popular algorithm because of low computation requirements and optimality for Gaussian cases. However, in the presence of non-gaussian *This paper was recommended by Regional Editor Majid Ahmadi. 256-
2 B. Chen et al. noises, the LMS algorithm will result in significant performance degradation. To improve the learning performance (degree of stability, convergence speed, misadjustment error, etc.) for non-gaussian situations, many stochastic gradient algorithms under non-mean-square error (Non-MSE) criteria are studied. 3 5 Among these algorithms, the least mean p-power (LMP) algorithm 5,6 deserves special attention due to its ease of mathematical tractability and satisfactory performance. In scenarios of adaptive system training, the LMP algorithm with order p is given by Wðk þ ðjeðkþjp Þ ¼ WðkÞ pjeðkþj eðkþ ; ðþ where WðkÞ ¼½w ðkþ; w 2 ðkþ;...; w M ðkþš T is the weight vector of the adaptive model at k iteration, > is the step-size, eðkþ denotes the prediction error, the eðkþ depends on the topology of the adaptive system under consideration. Well-known special cases of the LMP algorithm include the least absolute difference (LAD) algorithm (p ¼ ), 7 LMS algorithm (p ¼ 2), and the least mean fourth (LMF) algorithm (p ¼ 4). 3 More recently, researchers propose the minimum error entropy (MEE) criterion as an information theoretic alternative to traditional error criteria in supervised adaptive system training. 8,23 As shown in Fig., under MEE criterion, the weight vector of the adaptive model hðx; WÞ is adjusted online so that the error s entropy HðeðkÞÞ is minimized. The error s entropy measures the average uncertainty contained in the error signal. Its minimization forces the error samples to concentrate. Numerical examples suggested that the MEE criterion could be able to achieve a better error distribution, especially for higher-order statistics in the adaptive system training. 2 Under MEE criterion, the optimal weight vector W is x(k) f (x) z(k) + + Σ n(k) d(k) + Σ e(k) error entropy h(x,w) y(k) Fig.. Scheme of adaptive system training under MEE criterion, where dðkþ is the noisy observations of the system output, and is considered as the desired response of the adaptive model when driven by input signal xðkþ and perturbed by noise samples nðkþ
3 Stochastic Information Gradient Algorithm with Generalized Gaussian Distribution Model given by W ¼ arg min HðeÞ W2R M Z ¼ arg min W2R M R p e ðeþ log p e ðeþde ¼ arg min E e ð log p e ðeþþ ; ð2þ W2R M where p e ðþ denotes the PDF of the error, and E e ðþ denotes expectation operator. The optimal weight vector can be searched by stochastic information gradient (SIG) algorithm. 9 Wðk þ Þ log p : ð3þ The key problem of SIG algorithm is how to estimate the PDF of error eðkþ. The nonparametric kernel approach has been widely used due to its smoothing characteristics. 3 By kernel density estimation (KDE), the estimated error distribution is given by ^p e ðeðkþþ ¼ L X k i¼k Lþ K h ðeðkþ eðiþþ ; ð4þ where K h ð:þ represents the kernel function, h is the kernel width, and L denotes the sliding window length. By kernel approach, one is often confronted with the problem of how to choose a suitable value of the kernel width. An inappropriate choice of width will significantly deteriorate the behavior of the algorithm. 8, Though the effects of the kernel width on the shape of the performance surface and the eigenvalues of the Hessian at and around the optimal solution have been carefully investigated, the choice of the kernel width is still a difficult task. In Ref. 24, an adaptive kernel width is proposed, however, the computation of the kernel width is based on a gradient search algorithm which is even as complex as the adaptive algorithm itself. Hence, certain parameterized density estimation, which does not involve the choice of kernel width, might be more practical. Further, if some prior knowledge about the structure of the density is available, the parameterized density estimation may achieve a better accuracy than nonparametric ones. These facts motivate us to develop a parameterized SIG algorithm, in which the error s distribution is modeled by a certain parametric family of probability density functions. In this paper, the generalized Gaussian density (GGD) 4 is used to match the error s distribution. The reasons for the choice of GGD densities are at least twofold: () The GGD densities belong to a family of maximum entropy densities, which have simple yet flexible functional forms and could approximate a large number of different signals, comprising a wide range of statistical distributions 5 8 ; (2) The 256-3
4 B. Chen et al. GGD densities have only three parameters, i.e., location, shape and dispersion, which can be easily estimated from the sample data. 7,8 The paper is organized as follows: In Sec. 2, the generalized Gaussian distribution and its parametric estimation are introduced. In Sec. 3, the GGD-based SIG algorithm is derived, and the connections with LMP algorithm are discussed. In Sec. 4, a series of Monte Carlo simulation experiments are performed to demonstrate the performance of the new algorithm. Finally, concluding remarks are presented in Sec Generalized Gaussian PDF and Its Parametric Estimation The generalized Gaussian density (GGD) with location and standard deviation > is defined as jx j p X ðxþ ¼ 2 exp ; ð5þ in which sffiffiffiffiffiffiffiffiffiffi ¼ ð Þ ð 3 Þ ; ð6þ where ðþ denotes the Gamma function 9 given by ðzþ ¼ Z x z e x dx ; z > : ð7þ In Eq. (5), describes the exponential rate of decay, and hence the shape of PDF, while models the dispersion of the distribution. Usually, is called the shape parameter and is referred to as the dispersion or scale parameter. The GGD densities include Gaussian ( ¼ 2) and Laplacian ( ¼ ) distributions as special cases. Figure 2 shows the GGD distributions for several shape parameters with the same variance ( ¼ ). From Fig. 2, it is evident that smaller values of the shape parameter correspond to heavier tails and therefore to sharper distributions. In the limiting cases, as!, the GGD becomes close to the uniform distribution, whereas for! þ, it approaches an impulse function (-distribution). The GGD densities are the maximum entropy (Maxent) distributions, which satisfy the following constrained optimization (see Appendix A): 8 Z þ >< max pðxþ log pðxþdx >: s:t: p Z þ jx j pðxþdx ¼ m ¼ ; where m denotes the -order absolute central moment. ð8þ 256-4
5 Stochastic Information Gradient Algorithm with Generalized Gaussian Distribution Model α=3 α= α=2 α= Fig. 2. Generalized Gaussian distribution for different shape parameters with unit variance. According to Jaynes s maximum entropy principle (MEP), 2 the Maxent distribution is \uniquely determined as the one which is maximally noncommittal with regard to missing information, and that it agrees with what is known, but expresses maximum uncertainty with respect to all other matters". Therefore, the GGD distribution would be the most unbiased distribution that agrees with a certain absolute central moment constraint. Until now, a number of methods have been proposed for the parametric estimation of a GGD distribution. The state-of-the-art can be found in Refs. 4 and 8. Here we only introduce the moment-matching estimators (MMEs). A complete statistical description for the GGD-modeled distributions, by simply matching the underlying moments of the data set with those of the assumed distribution, is first obtained by Mallat. 2 For a GGD distribution, the r-order absolute central moment is calculated as Z þ m r ¼ E½jX j r Š¼ jx j r p X ðxþdx ¼ r ðr þ Þ : ð9þ p ffiffiffiffiffiffi The ratio of the mean absolute deviation (m ) to standard deviation ( )is MðÞ ¼ m 2 pffiffiffiffiffiffi ¼ q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ; ðþ m 2 3 m
6 B. Chen et al. which is a steadily increasing function of shape parameter. So an estimate for can be expressed as ^ ¼ M ^m pffiffiffiffiffiffi ; ðþ ^m 2 where M ðþ denotes the inverse function of MðÞ, ^m and ^m 2 represent the sample mean absolute deviation and the sample standard deviation, respectively. Given independent, identically distributed (i.i.d.) random samples fx ; x 2 ;...; x N g,we have 8 ^ ¼ X N x N i ; >< ^m ¼ N >: ^m 2 ¼ N i¼ X N i¼ X N i¼ jx i ^j ; ðx i ^Þ 2 : The shape parameter can also be expressed in terms of the kurtosis of the GGD distribution. 22 A similar ratio, which is equal to the reciprocal of the square root of kurtosis of the GGD distribution is given by KðÞ ¼ m 2 3 pffiffiffiffiffiffi ¼ q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : ð3þ m 4 Therefore, another estimate of could be ^ ¼ K ^m pffiffiffiffiffiffi 2 ; ð4þ ^m 4 where K ðþ denotes the inverse function of KðÞ, ^m 4 is the sample fourth-order central moment, that is, ^m 4 ¼ N X N i¼ 5 ðx i ^Þ 4 : To estimate the shape parameter, we have to solve the inverse functions M ðþ or K ðþ, whose curves are depicted in Fig. 3. As both the inverse functions have no explicit forms, some interpolation method should be used in practical applications. p ffiffiffiffiffiffi Plugging the estimate of into (6) and substituting for ^m 2, we may get the estimate of the dispersion parameter: s ^ ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ^m 2 ð ^ Þ ð 3^ Þ ð2þ ð5þ : ð6þ 256-6
7 Stochastic Information Gradient Algorithm with Generalized Gaussian Distribution Model 9 8 y=m - (x) y=k - (x) 7 6 y x Fig. 3. Curves of the inverse function M ðþ and K ðþ. 3. GGD-Based Stochastic Information Gradient Algorithm Using GGD density to approximate the error s distribution, we have! ^ ^p e ðeðkþþ ¼ k 2 ^ exp jeðkþ ^ ^k! kj ; ð7þ k ^ ^ k k where ^ k, ^ k,and^ k denote the estimated location, shape, and dispersion parameters at k iteration. By moment-matching estimators, say the kurtosis-based approach, we have 8 ^ k ¼ X k eðiþ ; L i¼k Lþ X k >< ^ k ¼ K B L i¼k Lþ ðeðiþ ^ kþ 2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi X A ; k i¼k Lþ ðeðiþ ^ kþ 4 ð8þ L vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi vffiffiffiffiffiffiffiffiffiffiffiffiffiffi u X ^ k ^ k ¼ t ðeðiþ ^ L k Þ 2 u k t ; >: i¼k Lþ 3^ k 256-7
8 B. Chen et al. where L denotes the sliding window length. Thus, the GGD-based SIG algorithm can be derived as 8! Wðk þ log ^ k exp jeðkþ ^ ^k! 9 < = kj : 2^ k ^ k ; ^ k ^k jeðkþ ^ k j ^ k ¼ WðkÞ ^ k ^ ^ jeðkþ ^ k k j ^ k signðeðkþ ^ eðkþ : k ð9þ In this paper, to make a distinction, we denote \SIG-Kernel" and \SIG-GGD" the SIG algorithms based on kernel approach and GGD densities, respectively. Compared with the SIG-Kernel algorithm, the SIG-GGD just needs to estimate the three parameters of GGD density, without resorting to the choice of kernel width and the calculation of kernel function. Comparing () and (9), we find that when ^ k, the SIG-GGD algorithm can be regarded as the LMP algorithm with adaptive order ^a k and variable step-size =^ ^ k k. In fact, under certain conditions, the SIG-GGD algorithm will converge to a specific LMP algorithm. Consider the finite impulse response (FIR) system identification case, in which the plant and the adaptive model are both FIR filters with the same order. In this case, the error signal eðkþ is given by eðkþ ¼W T XðkÞþnðkÞ W T ðkþxðkþ ¼ V T ðkþxðkþþnðkþ ; ð2þ where W ¼½w ; w 2 ;...; w M ŠT is the weight vector of unknown system, nðkþ is the interference noise, XðkÞ ¼½xðkÞ; xðk Þ;...; xðk M þ ÞŠ T is the input vector, and V ðkþ ¼W WðkÞ is the weight error vector. Then, near the convergence, we have WðkÞ W, and hence eðkþ ¼ðW WðkÞÞ T XðkÞþnðkÞ nðkþ : If fnðkþg is zero-mean white noise, we have ^ k and 8 X k ^ k K B L ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi i¼k Lþ n2ðiþ q X A ; k >< L i¼k Lþ n4 ðiþ vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi vffiffiffiffiffiffiffiffiffiffiffiffiffiffi u X ^ k ^ k t n 2 u k ðiþ t : >: L i¼k Lþ 3^ k ð2þ ð22þ 256-8
9 Stochastic Information Gradient Algorithm with Generalized Gaussian Distribution Model Assuming L as large enough, by the limit theorems of probability, 8 X k n 2 ðiþ!e½n 2 ðkþš ; >< L >: L i¼k Lþ X k i¼k Lþ n 4 ðiþ!e½n 4 ðkþš : Thus, the estimated parameters ^ k and ^ k shall approach certain constants, i.e., the SIG-GGD algorithm shall converge to a certain LMP algorithm with fixed order and step size. For example, if nðkþ is zero-mean Gaussian distributed (with shape parameter ¼ 2), we have ^ k 2 near the convergence, and consequently, the SIG- GGD algorithm converges to the LMS algorithm. Similarly, for Laplace noises, the SIG-GGD algorithm shall converge to the LAD algorithm. In Ref. 4, it has been shown that under slow adaptation, the LMS and LAD algorithms are, respectively, the optimum algorithms for the Gaussian and Laplace interfering noises. We may therefore conclude that the SIG-GGD algorithm has the ability to adjust its parameter pair ( ^ k, ^ k ) so as to switch to a certain optimum algorithm. This automatic switching property will be verified in the following section through a set of Monte- Carlo simulations. 4. Simulation Results To demonstrate the performance of the SIG-GGD algorithm, we perform a series of Monte-Carlo simulations for the FIR channel identification, in comparison with the SIG-Kernel, LAD, LMS, and LMF algorithms. Let the weight vector of unknown channel be W ¼½:; :3; :5; :3; :Š T, and the input signal be white Gaussian noise with unit power. For the SIG-Kernel algorithm, the Gaussian kernel is used, and the kernel width is chosen according to the well-known Silverman s rule of thumb. 3 For the SIG-GGD algorithm, the kurtosis-based moment-match estimator is used (see Ref. 8). To solve the inverse function K ðþ, we adopt the basic linear interpolation method, with step.5. In addition, to avoid \very large gradient", we set the upper bound of ^ k equal to 4, that is, if ^ k > 4, we set ^ k ¼ 4. In the experiments below, we focus on four special distributions of the disturbance noise, whose density functions are depicted in Fig. 4. Figures 5 8 illustrate the convergence curves of the mean-square deviation (MSD) E½V T ðkþv ðkþš averaged over independent Monte-Carlo runs for each disturbance noise. Figure 9 shows the averaged parameter k for Laplace, Gaussian, and Uniform noises. To compare the statistical results, we also summarize in Table the sample mean and standard deviation of the weight w 3 (w 3 ¼ :5). From the figures and table, we have the following observations: () The performances of the LAD, LMS, and LMF algorithms depend crucially on the distribution of the disturbance noise. The three algorithms may achieve the ð23þ 256-9
10 B. Chen et al..6.5 p(x) Gaussian Laplace Uniform MixNorm x Fig. 4. Four probability density functions of the disturbance noise. -5 SIG-Kernel SIG-GGD LAD LMS LMF LMF - MSD (db) -5 LMS SIG-Kernel SIG-GGD -2 LAD iterations Fig. 5. Convergence curves of mean square deviation (Laplace noise case). 256-
11 Stochastic Information Gradient Algorithm with Generalized Gaussian Distribution Model -5 SIG-Kernel SIG-GGD LAD LMS LMF - MSD (db) -5 SIG-Kernel LMF LAD LMS SIG-GGD iterations Fig. 6. Convergence curves of mean square deviation (Gaussian noise case). -5 SIG-Kernel SIG-GGD LAD LMS LMF - MSD (db) -5 LAD LMS SIG-Kernel LMF SIG-GGD iterations Fig. 7. Convergence curves of mean square deviation (Uniform noise case). 256-
12 B. Chen et al. -5 SIG-Kernel SIG-GGD LAD LMS LMF - MSD (db) -5 LAD LMS SIG-GGD LMF -2 SIG-Kernel iterations Fig. 8. Convergence curves of mean square deviation (MixNorm noise case). Table. Mean deviation results of w3 over Monte-Carlo simulations. SIG-Kernel SIG-GGD LAD LMS LMF Laplace Gaussian Uniform MixNorm smallest misadjustment for a certain noise distribution (e.g., the LMS algorithm for Gaussian noise); however, for other noise distributions, their performance will deteriorate dramatically. (2) The SIG-GGD and SIG-Kernel algorithms are both robust to the noise distribution. In the case of symmetric and unimodal noise (e.g., Laplace, Gaussian, Uniform), the SIG-GGD algorithm may achieve a smaller misadjustment than the SIG-Kernel algorithm. Though in the case of nonsymmetric and nonunimodal noise (e.g., MixNorm), the performance of SIG-GGD algorithm may be not as good as the SIG-Kernel algorithm, it is still better than most of the LMP algorithms. (3) Near the convergence, the parameter k converges approximately to, 2, and 4 (note that ^ k is restricted to ^ k 4 artificially) when disturbed by Laplace, 256-2
13 Stochastic Information Gradient Algorithm with Generalized Gaussian Distribution Model 2 α k iterations (a) Laplace Noise 2.5 α k iterations (b) Gaussian Noise 4 α k iterations (c) Uniform Noise Fig. 9. Parameter k averaged across independent Monte Carlo runs. Gaussian, and Uniform noise, which implies the SIG-GGD algorithm will switch approximately to the LAD, LMS, and LMF algorithm for, the Laplace, Gaussian, and Uniform noise, respectively. This agrees with the previous discussion in Sec Conclusions We have presented in this paper a new stochastic information gradient (SIG) algorithm, in which the error s distribution is modeled by the generalized Gaussian density (GGD). Different from the kernel-based SIG algorithm, the GGD-based algorithm does not resort to the choice of kernel width, and just relies on estimating the location, shape, and dispersion parameters during the adaptation. The relationships between the proposed algorithm and the least mean p-power (LMP) 256-3
14 B. Chen et al. algorithm are investigated. This new algorithm has the ability to adjust the parameters so as to switch approximately to a certain optimal LMP algorithm. Simulation results suggest that, for the case in which the noise has a symmetric and unimodal distribution, the SIG-GGD algorithm may outperform the SIG-Kernel algorithm. In the case of nonsymmetric and nonunimodal disturbance, the performance of the SIG-GGD algorithm may be not as good as the SIG-Kernel algorithm, however, it still outperforms most of the LMP algorithms. Acknowledgments This work was supported by the National Natural Science Foundation of China (No ), and was partially supported by NSF Grant ECCS 85644, NSF IIS 96497, and ONR N Appendix A To solve the optimization of (8), we create an unconstrained expression for the entropy using Lagrange multipliers: Z þ Z þ L ¼ pðxþ log pðxþdx ð Þ pðxþdx Z þ jx j pðxþdx : ða:þ Using calculus of variations, we maximize L with respect to pðxþ, f p log p ð Þp jx j pg¼ ) log p ð Þ jx j ¼ ) p ¼ expf jx j g ; where and are determined by 8Z þ >< expf jx j gdx ¼ ; >: Z þ jx j expf jx j gdx ¼ : Hence, it is easy to derive and as follows: ¼ logfð2 ð=þþ=g ; ¼ = : ða:2þ ða:3þ ða:4þ Combining (A.2) and (A.4) yields the GGD distribution. This completes the proof
15 Stochastic Information Gradient Algorithm with Generalized Gaussian Distribution Model References. S. Haykin, Adaptive Filtering Theory, 3rd edn. (Prentice Hall, NY, 996). 2. B. Widrow and S. D. Stearns, Adaptive Signal Processing (Prentice Hall, Englewood Cliffs, NJ, 985). 3. E. Walach and B. Widrow, The least mean fourth (LMF) adaptive algorithm and its family, IEEE Trans. Inform. Theory 3 (984) S. C. Douglas and H. Y. Meng, Stochastic gradient adaptation under general error criteria, IEEE Trans. Signal Process. 42 (994) S. C. Pei and C. C. Tseng, Least mean p-power error criterion for adaptive FIR filter, IEEE J. Select. Areas Commun. 2 (994) Y. Xiao, Y. Tadokoro and K. Shida, Adaptive algorithm based on least mean p-power error criterion for Fourier analysis in additive noise, IEEE Trans. Signal Process. 47 (999) V. J. Mathews and S. H. Cho, Improved convergence analysis of stochastic gradient adaptive filter using the sign algorithm, IEEE, Trans. Acoust. Speech, Signal Process. 35 (987) D. Erdogmus and J. C. Principe, Generalized information potential criterion for adaptive system training, IEEE Trans. Neural Ntw 3 (22) D. Erdogmus, E. H. Kenneth and J. C. Principe, Online entropy manipulation: Stochastic information gradient, IEEE Signal Process. Lett. (23) D. Erdogmus and J. C. Principe, Convergence properties and data efficiency of the minimum error entropy criterion in Adaline training, IEEE Trans. Signal Process. 5 (23) T. M. Cover and J. A. Thomas, Element of Information Theory (Wiley & Son, Inc., Chichester, 99). 2. D. Erdogmus and J. C. Principe, Comparison of entropy and mean square error criteria in adaptive system training using higher order statistics, Second Int. Workshop on Independent Component Analysis and Blind Signal Separation (2), pp B. W. Silverman, Density Estimation for Statistic and Data Analysis (Chapman & Hall, NY, 986). 4. M. K. Varanasi and B. Aazhang, Parametric generalized Gaussian density estimation, J. Acoust. Soc. Amer. 86 (989) R. L. Joshi and T. R. Fisher, Comparison of generalized Gaussian and Laplacian modeling in DCT image coding, IEEE Signal Process. Lett. 2 (995) F. Wang, H. Li and R. Li, Unified parametric and nonparametric ICA algorithms for hybrid source signals and stability analysis, Int. J. Innovative Comput. Inform. Control 4 (28) K. Sharifi and A. Leon-Garcia, Estimation of shape parameter for generalized Gaussian distributions in subband decompositions of video, IEEE Trans. Circuits Syst. Video Technol. 5 (995) K. Kokkinakis and A. K. Nandi, Exponent parameter estimation for generalized Gaussian probability density functions with application to speech modeling, Signal Process. 85 (25) M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions (Dover, New York, 97). 2. E. T. Jaynes, Information theory and statistical mechanics, Phys. Rev. 6 (957) S. Mallat, A theory for multiresolution signal decomposition: The wavelet representation, IEEE Trans. Pattern Anal. Machine Intell. (989)
16 B. Chen et al. 22. S. Choi, A. Cichocki and S. Amari, Flexible independent component analysis, J. VLSI Signal Process. 26 (2) B. Chen, J. Hu, L. Pu and Z. Sun, Stochastic gradient algorithm under (h;)-entropy criterion, Circuits Syst. Signal Process. 26 (27) A. Singh and J. C. Principe, Information theoretic learning with adaptive kernels, Signal Process. 9 (2)
STOCHASTIC INFORMATION GRADIENT ALGORITHM BASED ON MAXIMUM ENTROPY DENSITY ESTIMATION. Badong Chen, Yu Zhu, Jinchun Hu and Ming Zhang
ICIC Express Letters ICIC International c 2009 ISSN 1881-803X Volume 3, Number 3, September 2009 pp. 1 6 STOCHASTIC INFORMATION GRADIENT ALGORITHM BASED ON MAXIMUM ENTROPY DENSITY ESTIMATION Badong Chen,
More informationComputing Maximum Entropy Densities: A Hybrid Approach
Computing Maximum Entropy Densities: A Hybrid Approach Badong Chen Department of Precision Instruments and Mechanology Tsinghua University Beijing, 84, P. R. China Jinchun Hu Department of Precision Instruments
More informationRecursive Least Squares for an Entropy Regularized MSE Cost Function
Recursive Least Squares for an Entropy Regularized MSE Cost Function Deniz Erdogmus, Yadunandana N. Rao, Jose C. Principe Oscar Fontenla-Romero, Amparo Alonso-Betanzos Electrical Eng. Dept., University
More informationOn the steady-state mean squared error of the fixed-point LMS algorithm
Signal Processing 87 (2007) 3226 3233 Fast communication On the steady-state mean squared error of the fixed-point LMS algorithm Mohamed Ghanassi, Benoıˆt Champagne, Peter Kabal Department of Electrical
More informationSubmitted to Electronics Letters. Indexing terms: Signal Processing, Adaptive Filters. The Combined LMS/F Algorithm Shao-Jen Lim and John G. Harris Co
Submitted to Electronics Letters. Indexing terms: Signal Processing, Adaptive Filters. The Combined LMS/F Algorithm Shao-Jen Lim and John G. Harris Computational Neuro-Engineering Laboratory University
More informationStochastic error whitening algorithm for linear filter estimation with noisy data
Neural Networks 16 (2003) 873 880 2003 Special issue Stochastic error whitening algorithm for linear filter estimation with noisy data Yadunandana N. Rao*, Deniz Erdogmus, Geetha Y. Rao, Jose C. Principe
More informationRecurrent neural networks with trainable amplitude of activation functions
Neural Networks 16 (2003) 1095 1100 www.elsevier.com/locate/neunet Neural Networks letter Recurrent neural networks with trainable amplitude of activation functions Su Lee Goh*, Danilo P. Mandic Imperial
More informationError Entropy Criterion in Echo State Network Training
Error Entropy Criterion in Echo State Network Training Levy Boccato 1, Daniel G. Silva 1, Denis Fantinato 1, Kenji Nose Filho 1, Rafael Ferrari 1, Romis Attux 1, Aline Neves 2, Jugurta Montalvão 3 and
More informationRecursive Generalized Eigendecomposition for Independent Component Analysis
Recursive Generalized Eigendecomposition for Independent Component Analysis Umut Ozertem 1, Deniz Erdogmus 1,, ian Lan 1 CSEE Department, OGI, Oregon Health & Science University, Portland, OR, USA. {ozertemu,deniz}@csee.ogi.edu
More informationAn Error-Entropy Minimization Algorithm for Supervised Training of Nonlinear Adaptive Systems
1780 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 50, NO. 7, JULY 2002 An Error-Entropy Minimization Algorithm for Supervised Training of Nonlinear Adaptive Systems Deniz Erdogmus, Member, IEEE, and Jose
More informationMANY real-word applications require complex nonlinear
Proceedings of International Joint Conference on Neural Networks, San Jose, California, USA, July 31 August 5, 2011 Kernel Adaptive Filtering with Maximum Correntropy Criterion Songlin Zhao, Badong Chen,
More informationIndependent Component Analysis. Contents
Contents Preface xvii 1 Introduction 1 1.1 Linear representation of multivariate data 1 1.1.1 The general statistical setting 1 1.1.2 Dimension reduction methods 2 1.1.3 Independence as a guiding principle
More informationADAPTIVE FILTER THEORY
ADAPTIVE FILTER THEORY Fourth Edition Simon Haykin Communications Research Laboratory McMaster University Hamilton, Ontario, Canada Front ice Hall PRENTICE HALL Upper Saddle River, New Jersey 07458 Preface
More informationA METHOD OF ADAPTATION BETWEEN STEEPEST- DESCENT AND NEWTON S ALGORITHM FOR MULTI- CHANNEL ACTIVE CONTROL OF TONAL NOISE AND VIBRATION
A METHOD OF ADAPTATION BETWEEN STEEPEST- DESCENT AND NEWTON S ALGORITHM FOR MULTI- CHANNEL ACTIVE CONTROL OF TONAL NOISE AND VIBRATION Jordan Cheer and Stephen Daley Institute of Sound and Vibration Research,
More informationOn Information Maximization and Blind Signal Deconvolution
On Information Maximization and Blind Signal Deconvolution A Röbel Technical University of Berlin, Institute of Communication Sciences email: roebel@kgwtu-berlinde Abstract: In the following paper we investigate
More informationSTEADY-STATE MEAN SQUARE PERFORMANCE OF A SPARSIFIED KERNEL LEAST MEAN SQUARE ALGORITHM.
STEADY-STATE MEAN SQUARE PERFORMANCE OF A SPARSIFIED KERNEL LEAST MEAN SQUARE ALGORITHM Badong Chen 1, Zhengda Qin 1, Lei Sun 2 1 Institute of Artificial Intelligence and Robotics, Xi an Jiaotong University,
More information2262 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 47, NO. 8, AUGUST A General Class of Nonlinear Normalized Adaptive Filtering Algorithms
2262 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 47, NO. 8, AUGUST 1999 A General Class of Nonlinear Normalized Adaptive Filtering Algorithms Sudhakar Kalluri, Member, IEEE, and Gonzalo R. Arce, Senior
More informationPerformance Comparison of Two Implementations of the Leaky. LMS Adaptive Filter. Scott C. Douglas. University of Utah. Salt Lake City, Utah 84112
Performance Comparison of Two Implementations of the Leaky LMS Adaptive Filter Scott C. Douglas Department of Electrical Engineering University of Utah Salt Lake City, Utah 8411 Abstract{ The leaky LMS
More informationAn Improved Blind Spectrum Sensing Algorithm Based on QR Decomposition and SVM
An Improved Blind Spectrum Sensing Algorithm Based on QR Decomposition and SVM Yaqin Chen 1,(&), Xiaojun Jing 1,, Wenting Liu 1,, and Jia Li 3 1 School of Information and Communication Engineering, Beijing
More informationIS NEGATIVE STEP SIZE LMS ALGORITHM STABLE OPERATION POSSIBLE?
IS NEGATIVE STEP SIZE LMS ALGORITHM STABLE OPERATION POSSIBLE? Dariusz Bismor Institute of Automatic Control, Silesian University of Technology, ul. Akademicka 16, 44-100 Gliwice, Poland, e-mail: Dariusz.Bismor@polsl.pl
More informationConvergence Evaluation of a Random Step-Size NLMS Adaptive Algorithm in System Identification and Channel Equalization
Convergence Evaluation of a Random Step-Size NLMS Adaptive Algorithm in System Identification and Channel Equalization 1 Shihab Jimaa Khalifa University of Science, Technology and Research (KUSTAR) Faculty
More informationGeneralized Information Potential Criterion for Adaptive System Training
IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 13, NO. 5, SEPTEMBER 2002 1035 Generalized Information Potential Criterion for Adaptive System Training Deniz Erdogmus, Student Member, IEEE, and Jose C. Principe,
More informationStatistics: Learning models from data
DS-GA 1002 Lecture notes 5 October 19, 2015 Statistics: Learning models from data Learning models from data that are assumed to be generated probabilistically from a certain unknown distribution is a crucial
More informationIdentification of nonlinear discrete-time systems using raised-cosine radial basis function networks
International Journal of Systems Science volume 35, number 4, April 4, pages Identification of nonlinear discrete-time systems using raised-cosine radial basis function networks A. F. AL-AJLOUNI{, R.J.SCHILLING{*
More informationA Flexible ICA-Based Method for AEC Without Requiring Double-Talk Detection
APSIPA ASC 2011 Xi an A Flexible ICA-Based Method for AEC Without Requiring Double-Talk Detection Marko Kanadi, Muhammad Tahir Akhtar, Wataru Mitsuhashi Department of Information and Communication Engineering,
More informationMITIGATING UNCORRELATED PERIODIC DISTURBANCE IN NARROWBAND ACTIVE NOISE CONTROL SYSTEMS
17th European Signal Processing Conference (EUSIPCO 29) Glasgow, Scotland, August 24-28, 29 MITIGATING UNCORRELATED PERIODIC DISTURBANCE IN NARROWBAND ACTIVE NOISE CONTROL SYSTEMS Muhammad Tahir AKHTAR
More informationEstimation of the Optimum Rotational Parameter for the Fractional Fourier Transform Using Domain Decomposition
Estimation of the Optimum Rotational Parameter for the Fractional Fourier Transform Using Domain Decomposition Seema Sud 1 1 The Aerospace Corporation, 4851 Stonecroft Blvd. Chantilly, VA 20151 Abstract
More informationNOISE ROBUST RELATIVE TRANSFER FUNCTION ESTIMATION. M. Schwab, P. Noll, and T. Sikora. Technical University Berlin, Germany Communication System Group
NOISE ROBUST RELATIVE TRANSFER FUNCTION ESTIMATION M. Schwab, P. Noll, and T. Sikora Technical University Berlin, Germany Communication System Group Einsteinufer 17, 1557 Berlin (Germany) {schwab noll
More informationAlpha-Stable Distributions in Signal Processing of Audio Signals
Alpha-Stable Distributions in Signal Processing of Audio Signals Preben Kidmose, Department of Mathematical Modelling, Section for Digital Signal Processing, Technical University of Denmark, Building 3,
More informationAdaptive Filter Theory
0 Adaptive Filter heory Sung Ho Cho Hanyang University Seoul, Korea (Office) +8--0-0390 (Mobile) +8-10-541-5178 dragon@hanyang.ac.kr able of Contents 1 Wiener Filters Gradient Search by Steepest Descent
More informationPattern Recognition and Machine Learning. Bishop Chapter 2: Probability Distributions
Pattern Recognition and Machine Learning Chapter 2: Probability Distributions Cécile Amblard Alex Kläser Jakob Verbeek October 11, 27 Probability Distributions: General Density Estimation: given a finite
More information1 Introduction Independent component analysis (ICA) [10] is a statistical technique whose main applications are blind source separation, blind deconvo
The Fixed-Point Algorithm and Maximum Likelihood Estimation for Independent Component Analysis Aapo Hyvarinen Helsinki University of Technology Laboratory of Computer and Information Science P.O.Box 5400,
More informationNonparametric Bayesian Methods (Gaussian Processes)
[70240413 Statistical Machine Learning, Spring, 2015] Nonparametric Bayesian Methods (Gaussian Processes) Jun Zhu dcszj@mail.tsinghua.edu.cn http://bigml.cs.tsinghua.edu.cn/~jun State Key Lab of Intelligent
More informationMMSE DECODING FOR ANALOG JOINT SOURCE CHANNEL CODING USING MONTE CARLO IMPORTANCE SAMPLING
MMSE DECODING FOR ANALOG JOINT SOURCE CHANNEL CODING USING MONTE CARLO IMPORTANCE SAMPLING Yichuan Hu (), Javier Garcia-Frias () () Dept. of Elec. and Comp. Engineering University of Delaware Newark, DE
More informationA Canonical Genetic Algorithm for Blind Inversion of Linear Channels
A Canonical Genetic Algorithm for Blind Inversion of Linear Channels Fernando Rojas, Jordi Solé-Casals, Enric Monte-Moreno 3, Carlos G. Puntonet and Alberto Prieto Computer Architecture and Technology
More informationNew design of estimators using covariance information with uncertain observations in linear discrete-time systems
Applied Mathematics and Computation 135 (2003) 429 441 www.elsevier.com/locate/amc New design of estimators using covariance information with uncertain observations in linear discrete-time systems Seiichi
More informationDesign of Image Adaptive Wavelets for Denoising Applications
Design of Image Adaptive Wavelets for Denoising Applications Sanjeev Pragada and Jayanthi Sivaswamy Center for Visual Information Technology International Institute of Information Technology - Hyderabad,
More informationA Minimum Error Entropy criterion with Self Adjusting. Step-size (MEE-SAS)
Signal Processing - EURASIP (SUBMIED) A Minimum Error Entropy criterion with Self Adjusting Step-size (MEE-SAS) Seungju Han *, Sudhir Rao *, Deniz Erdogmus, Kyu-Hwa Jeong *, Jose Principe * Corresponding
More informationBlind Deconvolution by Modified Bussgang Algorithm
Reprinted from ISCAS'99, International Symposium on Circuits and Systems, Orlando, Florida, May 3-June 2, 1999 Blind Deconvolution by Modified Bussgang Algorithm Simone Fiori, Aurelio Uncini, Francesco
More informationinear Adaptive Inverse Control
Proceedings of the 36th Conference on Decision & Control San Diego, California USA December 1997 inear Adaptive nverse Control WM15 1:50 Bernard Widrow and Gregory L. Plett Department of Electrical Engineering,
More informationBlind Deconvolution via Maximum Kurtosis Adaptive Filtering
Blind Deconvolution via Maximum Kurtosis Adaptive Filtering Deborah Pereg Doron Benzvi The Jerusalem College of Engineering Jerusalem, Israel doronb@jce.ac.il, deborahpe@post.jce.ac.il ABSTRACT In this
More informationA low intricacy variable step-size partial update adaptive algorithm for Acoustic Echo Cancellation USNRao
ISSN: 77-3754 International Journal of Engineering and Innovative echnology (IJEI Volume 1, Issue, February 1 A low intricacy variable step-size partial update adaptive algorithm for Acoustic Echo Cancellation
More informationEM-algorithm for Training of State-space Models with Application to Time Series Prediction
EM-algorithm for Training of State-space Models with Application to Time Series Prediction Elia Liitiäinen, Nima Reyhani and Amaury Lendasse Helsinki University of Technology - Neural Networks Research
More informationON SOME EXTENSIONS OF THE NATURAL GRADIENT ALGORITHM. Brain Science Institute, RIKEN, Wako-shi, Saitama , Japan
ON SOME EXTENSIONS OF THE NATURAL GRADIENT ALGORITHM Pando Georgiev a, Andrzej Cichocki b and Shun-ichi Amari c Brain Science Institute, RIKEN, Wako-shi, Saitama 351-01, Japan a On leave from the Sofia
More informationRiccati difference equations to non linear extended Kalman filter constraints
International Journal of Scientific & Engineering Research Volume 3, Issue 12, December-2012 1 Riccati difference equations to non linear extended Kalman filter constraints Abstract Elizabeth.S 1 & Jothilakshmi.R
More informationCondensed Table of Contents for Introduction to Stochastic Search and Optimization: Estimation, Simulation, and Control by J. C.
Condensed Table of Contents for Introduction to Stochastic Search and Optimization: Estimation, Simulation, and Control by J. C. Spall John Wiley and Sons, Inc., 2003 Preface... xiii 1. Stochastic Search
More informationEfficient Use Of Sparse Adaptive Filters
Efficient Use Of Sparse Adaptive Filters Andy W.H. Khong and Patrick A. Naylor Department of Electrical and Electronic Engineering, Imperial College ondon Email: {andy.khong, p.naylor}@imperial.ac.uk Abstract
More informationEFFICIENT SHAPE OPTIMIZATION USING POLYNOMIAL CHAOS EXPANSION AND LOCAL SENSITIVITIES
9 th ASCE Specialty Conference on Probabilistic Mechanics and Structural Reliability EFFICIENT SHAPE OPTIMIZATION USING POLYNOMIAL CHAOS EXPANSION AND LOCAL SENSITIVITIES Nam H. Kim and Haoyu Wang University
More informationRobust extraction of specific signals with temporal structure
Robust extraction of specific signals with temporal structure Zhi-Lin Zhang, Zhang Yi Computational Intelligence Laboratory, School of Computer Science and Engineering, University of Electronic Science
More informationsquares based sparse system identification for the error in variables
Lim and Pang SpringerPlus 2016)5:1460 DOI 10.1186/s40064-016-3120-6 RESEARCH Open Access l 1 regularized recursive total least squares based sparse system identification for the error in variables Jun
More informationPILCO: A Model-Based and Data-Efficient Approach to Policy Search
PILCO: A Model-Based and Data-Efficient Approach to Policy Search (M.P. Deisenroth and C.E. Rasmussen) CSC2541 November 4, 2016 PILCO Graphical Model PILCO Probabilistic Inference for Learning COntrol
More informationLecture : Probabilistic Machine Learning
Lecture : Probabilistic Machine Learning Riashat Islam Reasoning and Learning Lab McGill University September 11, 2018 ML : Many Methods with Many Links Modelling Views of Machine Learning Machine Learning
More informationADAPTIVE FILTER THEORY
ADAPTIVE FILTER THEORY Fifth Edition Simon Haykin Communications Research Laboratory McMaster University Hamilton, Ontario, Canada International Edition contributions by Telagarapu Prabhakar Department
More informationLow-Complexity Image Denoising via Analytical Form of Generalized Gaussian Random Vectors in AWGN
Low-Complexity Image Denoising via Analytical Form of Generalized Gaussian Random Vectors in AWGN PICHID KITTISUWAN Rajamangala University of Technology (Ratanakosin), Department of Telecommunication Engineering,
More informationStochastic Optimization with Inequality Constraints Using Simultaneous Perturbations and Penalty Functions
International Journal of Control Vol. 00, No. 00, January 2007, 1 10 Stochastic Optimization with Inequality Constraints Using Simultaneous Perturbations and Penalty Functions I-JENG WANG and JAMES C.
More informationNonparametric Density Estimation. October 1, 2018
Nonparametric Density Estimation October 1, 2018 Introduction If we can t fit a distribution to our data, then we use nonparametric density estimation. Start with a histogram. But there are problems with
More informationDensity Estimation. Seungjin Choi
Density Estimation Seungjin Choi Department of Computer Science and Engineering Pohang University of Science and Technology 77 Cheongam-ro, Nam-gu, Pohang 37673, Korea seungjin@postech.ac.kr http://mlg.postech.ac.kr/
More informationSparse Least Mean Square Algorithm for Estimation of Truncated Volterra Kernels
Sparse Least Mean Square Algorithm for Estimation of Truncated Volterra Kernels Bijit Kumar Das 1, Mrityunjoy Chakraborty 2 Department of Electronics and Electrical Communication Engineering Indian Institute
More informationADAPTIVE signal processing algorithms (ASPA s) are
IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 46, NO. 12, DECEMBER 1998 3315 Locally Optimum Adaptive Signal Processing Algorithms George V. Moustakides Abstract We propose a new analytic method for comparing
More informationLTI Systems, Additive Noise, and Order Estimation
LTI Systems, Additive oise, and Order Estimation Soosan Beheshti, Munther A. Dahleh Laboratory for Information and Decision Systems Department of Electrical Engineering and Computer Science Massachusetts
More informationAn Adaptive Sensor Array Using an Affine Combination of Two Filters
An Adaptive Sensor Array Using an Affine Combination of Two Filters Tõnu Trump Tallinn University of Technology Department of Radio and Telecommunication Engineering Ehitajate tee 5, 19086 Tallinn Estonia
More informationA NOVEL OPTIMAL PROBABILITY DENSITY FUNCTION TRACKING FILTER DESIGN 1
A NOVEL OPTIMAL PROBABILITY DENSITY FUNCTION TRACKING FILTER DESIGN 1 Jinglin Zhou Hong Wang, Donghua Zhou Department of Automation, Tsinghua University, Beijing 100084, P. R. China Control Systems Centre,
More informationOutline Lecture 2 2(32)
Outline Lecture (3), Lecture Linear Regression and Classification it is our firm belief that an understanding of linear models is essential for understanding nonlinear ones Thomas Schön Division of Automatic
More informationHST.582J / 6.555J / J Biomedical Signal and Image Processing Spring 2007
MIT OpenCourseWare http://ocw.mit.edu HST.582J / 6.555J / 16.456J Biomedical Signal and Image Processing Spring 2007 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
More informationTime-domain noise reduction based on an orthogonal decomposition for desired signal extraction
Time-domain noise reduction based on an orthogonal decomposition for desired signal extraction Jacob Benesty INRS-EMT, University of Quebec, 800 de la Gauchetiere Ouest, Suite 6900, Montreal, Quebec H5A
More informationIndependent Component Analysis and Unsupervised Learning
Independent Component Analysis and Unsupervised Learning Jen-Tzung Chien National Cheng Kung University TABLE OF CONTENTS 1. Independent Component Analysis 2. Case Study I: Speech Recognition Independent
More informationBlind Source Separation Using Artificial immune system
American Journal of Engineering Research (AJER) e-issn : 2320-0847 p-issn : 2320-0936 Volume-03, Issue-02, pp-240-247 www.ajer.org Research Paper Open Access Blind Source Separation Using Artificial immune
More informationMixture Models and EM
Mixture Models and EM Goal: Introduction to probabilistic mixture models and the expectationmaximization (EM) algorithm. Motivation: simultaneous fitting of multiple model instances unsupervised clustering
More informationOn the Behavior of Information Theoretic Criteria for Model Order Selection
IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 49, NO. 8, AUGUST 2001 1689 On the Behavior of Information Theoretic Criteria for Model Order Selection Athanasios P. Liavas, Member, IEEE, and Phillip A. Regalia,
More informationExpectation propagation for signal detection in flat-fading channels
Expectation propagation for signal detection in flat-fading channels Yuan Qi MIT Media Lab Cambridge, MA, 02139 USA yuanqi@media.mit.edu Thomas Minka CMU Statistics Department Pittsburgh, PA 15213 USA
More informationELEG 833. Nonlinear Signal Processing
Nonlinear Signal Processing ELEG 833 Gonzalo R. Arce Department of Electrical and Computer Engineering University of Delaware arce@ee.udel.edu February 15, 2005 1 INTRODUCTION 1 Introduction Signal processing
More informationTHE problem of phase noise and its influence on oscillators
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II: EXPRESS BRIEFS, VOL. 54, NO. 5, MAY 2007 435 Phase Diffusion Coefficient for Oscillators Perturbed by Colored Noise Fergal O Doherty and James P. Gleeson Abstract
More informationOn the Use of A Priori Knowledge in Adaptive Inverse Control
54 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS PART I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL 47, NO 1, JANUARY 2000 On the Use of A Priori Knowledge in Adaptive Inverse Control August Kaelin, Member,
More informationThe Laplacian PDF Distance: A Cost Function for Clustering in a Kernel Feature Space
The Laplacian PDF Distance: A Cost Function for Clustering in a Kernel Feature Space Robert Jenssen, Deniz Erdogmus 2, Jose Principe 2, Torbjørn Eltoft Department of Physics, University of Tromsø, Norway
More informationAlgorithm for Multiple Model Adaptive Control Based on Input-Output Plant Model
BULGARIAN ACADEMY OF SCIENCES CYBERNEICS AND INFORMAION ECHNOLOGIES Volume No Sofia Algorithm for Multiple Model Adaptive Control Based on Input-Output Plant Model sonyo Slavov Department of Automatics
More informationRECONSTRUCTED QUANTIZED COEFFICIENTS MODELED WITH GENERALIZED GAUSSIAN DISTRIBUTION WITH EXPONENT 1/3
Image Processing & Communications, vol. 2, no. 4, pp.5-2 DOI:.55/ipc-26-9 5 RECONSTRUCTED QUANTIZED COEFFICIENTS MODELED WITH GENERALIZED GAUSSIAN DISTRIBUTION WITH EXPONENT /3 ROBERT KRUPIŃSKI West-Pomeranian
More information798 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 44, NO. 10, OCTOBER 1997
798 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL 44, NO 10, OCTOBER 1997 Stochastic Analysis of the Modulator Differential Pulse Code Modulator Rajesh Sharma,
More informationOutput regulation of general discrete-time linear systems with saturation nonlinearities
INTERNATIONAL JOURNAL OF ROBUST AND NONLINEAR CONTROL Int. J. Robust Nonlinear Control 2002; 12:1129 1143 (DOI: 10.1002/rnc.647) Output regulation of general discrete-time linear systems with saturation
More informationTitle without the persistently exciting c. works must be obtained from the IEE
Title Exact convergence analysis of adapt without the persistently exciting c Author(s) Sakai, H; Yang, JM; Oka, T Citation IEEE TRANSACTIONS ON SIGNAL 55(5): 2077-2083 PROCESS Issue Date 2007-05 URL http://hdl.handle.net/2433/50544
More informationSpatially Smoothed Kernel Density Estimation via Generalized Empirical Likelihood
Spatially Smoothed Kernel Density Estimation via Generalized Empirical Likelihood Kuangyu Wen & Ximing Wu Texas A&M University Info-Metrics Institute Conference: Recent Innovations in Info-Metrics October
More informationComparison of DDE and ETDGE for. Time-Varying Delay Estimation. H. C. So. Department of Electronic Engineering, City University of Hong Kong
Comparison of DDE and ETDGE for Time-Varying Delay Estimation H. C. So Department of Electronic Engineering, City University of Hong Kong Tat Chee Avenue, Kowloon, Hong Kong Email : hcso@ee.cityu.edu.hk
More informationV. Adaptive filtering Widrow-Hopf Learning Rule LMS and Adaline
V. Adaptive filtering Widrow-Hopf Learning Rule LMS and Adaline Goals Introduce Wiener-Hopf (WH) equations Introduce application of the steepest descent method to the WH problem Approximation to the Least
More informationSystem Identification and Adaptive Filtering in the Short-Time Fourier Transform Domain
System Identification and Adaptive Filtering in the Short-Time Fourier Transform Domain Electrical Engineering Department Technion - Israel Institute of Technology Supervised by: Prof. Israel Cohen Outline
More informationANALYSIS OF STEADY-STATE EXCESS MEAN-SQUARE-ERROR OF THE LEAST MEAN KURTOSIS ADAPTIVE ALGORITHM
ANALYSIS OF SEADY-SAE EXCESS MEAN-SQUARE-ERROR OF HE LEAS MEAN KUROSIS ADAPIVE ALGORIHM Junibakti Sanubari Department of Electronics Engineering, Satya Wacana University Jalan Diponegoro 5 60, Salatiga,
More informationSTA 4273H: Statistical Machine Learning
STA 4273H: Statistical Machine Learning Russ Salakhutdinov Department of Statistics! rsalakhu@utstat.toronto.edu! http://www.utstat.utoronto.ca/~rsalakhu/ Sidney Smith Hall, Room 6002 Lecture 7 Approximate
More informationPattern Recognition and Machine Learning
Christopher M. Bishop Pattern Recognition and Machine Learning ÖSpri inger Contents Preface Mathematical notation Contents vii xi xiii 1 Introduction 1 1.1 Example: Polynomial Curve Fitting 4 1.2 Probability
More informationCOMPLEX WAVELET TRANSFORM IN SIGNAL AND IMAGE ANALYSIS
COMPLEX WAVELET TRANSFORM IN SIGNAL AND IMAGE ANALYSIS MUSOKO VICTOR, PROCHÁZKA ALEŠ Institute of Chemical Technology, Department of Computing and Control Engineering Technická 905, 66 8 Prague 6, Cech
More informationAsymptotic Analysis of the Generalized Coherence Estimate
IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 49, NO. 1, JANUARY 2001 45 Asymptotic Analysis of the Generalized Coherence Estimate Axel Clausen, Member, IEEE, and Douglas Cochran, Senior Member, IEEE Abstract
More informationAdaptive Filtering for Flow-Cytometric Particles
126 Part. Part. Syst. Charact. 17 (2000) 126±133 Adaptive Filtering for Flow-Cytometric Particles alek Adjouadi*, Carlos Reyes**, John Riley***, Patricio Vidal* (Received 15 June 2000; resubmitted 26 July
More informationA ROBUST BEAMFORMER BASED ON WEIGHTED SPARSE CONSTRAINT
Progress In Electromagnetics Research Letters, Vol. 16, 53 60, 2010 A ROBUST BEAMFORMER BASED ON WEIGHTED SPARSE CONSTRAINT Y. P. Liu and Q. Wan School of Electronic Engineering University of Electronic
More informationLinear Least-Squares Based Methods for Neural Networks Learning
Linear Least-Squares Based Methods for Neural Networks Learning Oscar Fontenla-Romero 1, Deniz Erdogmus 2, JC Principe 2, Amparo Alonso-Betanzos 1, and Enrique Castillo 3 1 Laboratory for Research and
More informationSimultaneous Diagonalization in the Frequency Domain (SDIF) for Source Separation
Simultaneous Diagonalization in the Frequency Domain (SDIF) for Source Separation Hsiao-Chun Wu and Jose C. Principe Computational Neuro-Engineering Laboratory Department of Electrical and Computer Engineering
More informationNo. 6 Determining the input dimension of a To model a nonlinear time series with the widely used feed-forward neural network means to fit the a
Vol 12 No 6, June 2003 cfl 2003 Chin. Phys. Soc. 1009-1963/2003/12(06)/0594-05 Chinese Physics and IOP Publishing Ltd Determining the input dimension of a neural network for nonlinear time series prediction
More informationIndependent Component Analysis and Unsupervised Learning. Jen-Tzung Chien
Independent Component Analysis and Unsupervised Learning Jen-Tzung Chien TABLE OF CONTENTS 1. Independent Component Analysis 2. Case Study I: Speech Recognition Independent voices Nonparametric likelihood
More informationReading Group on Deep Learning Session 1
Reading Group on Deep Learning Session 1 Stephane Lathuiliere & Pablo Mesejo 2 June 2016 1/31 Contents Introduction to Artificial Neural Networks to understand, and to be able to efficiently use, the popular
More informationEVALUATING SYMMETRIC INFORMATION GAP BETWEEN DYNAMICAL SYSTEMS USING PARTICLE FILTER
EVALUATING SYMMETRIC INFORMATION GAP BETWEEN DYNAMICAL SYSTEMS USING PARTICLE FILTER Zhen Zhen 1, Jun Young Lee 2, and Abdus Saboor 3 1 Mingde College, Guizhou University, China zhenz2000@21cn.com 2 Department
More informationNew Recursive-Least-Squares Algorithms for Nonlinear Active Control of Sound and Vibration Using Neural Networks
IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 12, NO. 1, JANUARY 2001 135 New Recursive-Least-Squares Algorithms for Nonlinear Active Control of Sound and Vibration Using Neural Networks Martin Bouchard,
More informationPCA & ICA. CE-717: Machine Learning Sharif University of Technology Spring Soleymani
PCA & ICA CE-717: Machine Learning Sharif University of Technology Spring 2015 Soleymani Dimensionality Reduction: Feature Selection vs. Feature Extraction Feature selection Select a subset of a given
More informationParametric Techniques
Parametric Techniques Jason J. Corso SUNY at Buffalo J. Corso (SUNY at Buffalo) Parametric Techniques 1 / 39 Introduction When covering Bayesian Decision Theory, we assumed the full probabilistic structure
More informationA NEW INFORMATION THEORETIC APPROACH TO ORDER ESTIMATION PROBLEM. Massachusetts Institute of Technology, Cambridge, MA 02139, U.S.A.
A EW IFORMATIO THEORETIC APPROACH TO ORDER ESTIMATIO PROBLEM Soosan Beheshti Munther A. Dahleh Massachusetts Institute of Technology, Cambridge, MA 0239, U.S.A. Abstract: We introduce a new method of model
More information