Robust techniques for independent component analysis (ICA) with noisy data

Size: px
Start display at page:

Download "Robust techniques for independent component analysis (ICA) with noisy data"

Transcription

1 Neurocomputing 22 (1998) Robust techniques for independent component analysis (ICA) with noisy data A. Cichocki*, S.C. Douglas, S. Amari Brain Science Institute Riken, Brain Information Processing Group, 2-1 Hirosawa, Wako-shi, Saitama , Japan Department of Electrical Engineering, University of Utah, Salt Lake City, UT 84112, USA Accepted 3 July 1998 Abstract In this contribution, we propose approaches to independent component analysis (ICA) when the measured signals are contaminated by additive noise. We extend existing adaptive algorithms with equivariant properties in order to considerably reduce the bias in the demixing matrix caused by measurement noise. Moreover, we describe a novel recurrent dynamic neural network for simultaneous estimation of the unknown mixing matrix, blind source separation, and reduction of noise in the extracted output signals. We discuss the optimal choice of nonlinear activation functions for various noise distributions assuming a generalized Gaussian-distributed noise model. Computer simulations of a selected approach are provided that confirm its usefulness and excellent performance Elsevier Science B.V. All rights reserved. Keywords: Independent component analysis (ICA); Bias removal; Noise cancellation; Natural gradient; Blind source separation; Maximum likelihood 1. Introduction Recently, a number of efficient adaptive, on-line learning algorithms have been developed for ICA [1 19,21 31]. Although the underlying principles and approaches are different, many of the techniques have very similar forms. Most of these algorithms assume that any measurement noise within the mixed signals can be neglected. * Corresponding author. cia@brain.riken.go.jp. A. Cichocki is on leave from the Warsaw University of Technology, Poland /98/$ see front matter 1998 Elsevier Science B.V. All rights reserved. PII S (98)

2 114 A. Cichocki et al./neurocomputing 22 (1998) However, in real-world applications most measured signals are contaminated by additive noise. Thus, the problem of efficiently reducing the influence of noise on the performance of algorithms for ICA arises, and in particular, methods are desired to reduce noise in the stochastically independent extracted components. This paper addresses this difficult and challenging problem [6,7,21,27]. There are at least three definitions of ICA. In this paper, we employ the two distinct definitions given below: Definition 1. The ICA of a noisy random vector x(t)"[x (t)2x (t)] is obtained by finding an nm, (with m5n), full rank, linear transformation matrix W such that the output signal vector y(t)"[y (t)2y (t)], defined as y(t)"wx(t), (1) contains source components that are as independent as possible, as measured by an information-theoretic cost function such as minimum Kullback Leibler divergence. Definition 2. For a random noisy vector x(t) defined as x(t)"hs(t)#ν(t), (2) where H is an (mn) mixing matrix (m5n), s(t)"[s (t)2s (t)] is a source vector of stochastically independent signals, and ν(t) is a vector of uncorrelated noises, ICA is obtained by estimating both the mixing matrix H and the additive noise ν(t). As the estimation of a separating or de-mixing matrix W and/or a mixing matrix H in the presence of noise is rather difficult, the majority of past research efforts were devoted to the noiseless case where ν(t)"0. The objective of this paper is to develop novel approaches and learning algorithms that are more robust with respect to noise than existing techniques or that can reduce the noise in the estimated output vector y(t). In this paper, we assume that the source signals and additive noise components are mutually stochastically independent. 2. Bias removal techniques for pre-whitening and ICA algorithms 2.1. Bias removal for pre-whitening algorithms Consider the standard decorrelation or pre-whitening algorithm for x(t) given by [15,22] W(t#1)"W(t)#η(t)[I!y(t)y(t)]W(t) (3) or its averaged version given by W(t)"η(t)[I!Ey(t)y(t)]W(t), (4) where y(t)"w(t)x(t) and E ) denotes statistical expectation. When x(t) is noisy such that x(t)"xl (t)#ν(t) and xl (t) and yl (t)"w(t)xl (t) are the noiseless estimates of the input

3 A. Cichocki et al./neurocomputing 22 (1998) and output vectors, respectively, it is easy to show that the additive noise ν(t) within x(t) introduces a bias in the estimated decorrelation matrix W. The covariance matrix of the output can be evaluated as R "Ey(t)y(t)"WR L L W#WR W, (5) where R L L "ExL (t)xl (t) and R "Eν(t)ν(t). Assuming that the covariance matrix of the noise is known (e.g. R "σi) or can be estimated, a proposed modified algorithm employing bias removal is given by W(t)"η(t)[I!Ey(t)y(t)#W(t)R W(t)]W(t). (6) The stochastic gradient version of this algorithm for R "σi is W(t)"η(t)[I!y(t)y(t)#σW(t)W(t)]W(t). (7) 2.2. Bias removal for ICA adaptive algorithms A similar technique as described above can be applied to remove the coefficient bias for a class of natural gradient algorithms for ICA [17]. In what follows, we outline these algorithms and the proposed modifications. To formulate the ICA problem, one must define an appropriate loss or cost function that depends on the parameters of the specified neural network ICA model. Minimization of such a loss function should cause the outputs of the model to satisfy the desired statistical conditions of stochastic independence and/or temporal and spatial mutual decorrelation [2,4,5,19,26]. Minimum entropy, ICA, and maximum likelihood lead to similar expected loss functions that measure the mutual stochastic independence of the system s output signals. A unifying loss or risk function is given by the Kullback Leibler divergence [2] E(y,W)" p (y)log p (y) dy, (8) p (y) where p (y) is the joint probability distribution of the output signal vector and p (y)"p (y ) is the model distribution assuming that y contains independent components. When used in a stochastic-gradient-type algorithm, this loss function can be expressed in a simple form as (y,w)"! log det(ww)! log p (y ), (9) where p (y ) is the assumed form of the probability density function (p.d.f.) of the ith output signal, det(ww) is the determinant of symmetric positive-definite matrix WW and ( ) ) is the transpose operator. The natural gradient search method [2] has emerged as a particularly useful technique for solving iterative optimization problems. Taking into account that the gradient of the loss function can be expressed as (W) "!(WW)W#f( y)x, (10) W

4 116 A. Cichocki et al./neurocomputing 22 (1998) the natural gradient learning rule is [2] W(t)"W(t#1)!W(t)"!η (W) W WW "η(t)[i!f(y(t))y(t)]w(t), (11) where f(y)"[f )2f )] with f )"! log(p )) y "! pr ) p ). (12) Typical p.d.f. and corresponding optimal nonlinear activation functions are shown in Table 1. Alternatively, we can use the following pre-conditioning filtered gradient rule [3,8,13] W(t)"!η(t)W(t) (W(t)) W W(t)"η(t)[I!y(t)u(y(t))]W(t), (13) Table 1 Typical probability density functions p(y) and corresponding activation functions f (y)"!d log p(y)/dy Name Density function p(y) Activation function f (y) Gaussian Laplace Cauchy Hyperbolic cosine Unimodal Triangular Generalized Gaussian Robust generalized Gaussian 1 2πσ exp(!y/2σ) 1 2σ exp(!y/σ) 1 πσ(1#(y/σ)) 1 π cosh(γy) exp(!2γy) (1#exp(!2γy)) 1 a (1!y/a) y(a r 2aΓ(1/r) exp(!(y/a)) r51 r 2aΓ(1/r) exp(!ρ(y)/a) r51, ρ(y) robust function y sign(y) 2y σ#y tanh(γy) tanh(γy) sign(y) a(1!y/a) y sign(y) ρ(y)ρ y

5 A. Cichocki et al./neurocomputing 22 (1998) where u(y)"[g (y )2g (y )] with nonlinearities g (y ) are now inverse (dual) to function f (y )"!p (y )/p (y ) (e.g. instead of f (y )"y we use cubic function g (y )"y or instead f (y )"tanh(y ) we can use inverse function g )"artanh )" log 1#y 1!y. The performances of these learning algorithms strongly depend on the shapes of the activation functions f ) and g ). Moreover, the optimal selection of these functions depends on the p.d.f. s of the source signals. It has been indicated via analysis and simulation that for the specific choices of nonlinearities f )"α y #tanh(β y ), the learning rule in Eq. (11) is able to successfully separate the source signals if all the source p.d.f. s are heavy-tailed, not unlike super-gaussian signals, whereas the learning rule in Eq. (13) can separate source signals with light-tailed p.d.f. s similar to sub-gaussian signals. Alternatively, when f )"α y #y, the algorithm in Eq. (11) can separate sub-gaussian signals, whereas the algorithm in Eq. (13) can separate super-gaussian signals. However, if the measured signals x (k) contain mixtures of both sub-gaussian and super-gaussian sources, then these algorithms may fail to separate these signals reliably, in which case other approaches have been suggested [16,19]. Note that the above two learning rules can be combined to form a more general and flexible universal learning rule [11 13] W(t)"η(t)[I!f[y(t)]u[y(t)]]W(t), (14) where f[y(t)] and u[y(t)] are suitably designed nonlinear functions, e.g. f (y )" tanh(β y ) sign(y )y g (y )" sign )y tanh(β y ) for κ )'δ, otherwise, for κ )'!δ, otherwise, (15) (16) where r 52, κ )"Ey /Ey!3 is the normalized value of kurtosis and δ50 is a small threshold. The value of the kurtosis can be evaluated on-line using Ey (k#1)"(1!η)ey(k)#ηy (k), (q"2,4). (17) The above learning algorithm (14) (16) monitors and estimates the statistics of each output signal and depending on sign or value of its normalized kurtosis (which is the measure of distance from the Gaussianity) automatically selects (or switches) suitable nonlinear activation functions, such that successful (stable) separation of all non- Gaussian source signals is possible. In this approach activation functions are adaptive time-varying nonlinearities. It should be noted that nonlinearities of the form f )"tanh(β y ) or g )"tanh(β y ) provide a degree of robustness to outliers that is not shared by

6 118 A. Cichocki et al./neurocomputing 22 (1998) nonlinearities of the form f (y )"sign(y )y (for r 53) or f (y )"αy # sign(κ(y ))tanh(βy ). For these choices, the parameters β 52 can be either fixed in value or adapted during the learning process as [9,24,30] log(p (y )) β (k)"!η "η β β η log(cosh(β (k)y (k)))/β (k)!y (k) β (k)!0.729/(β (k)#1.397). (18) The learning algorithms (11) (14) have been shown to possess excellent performance when separating noiseless signal mixtures; however, its performance deteriorates with noisy measurements due to undesirable coefficient biases and the existence of noise in the separated signals. In order to estimate the coefficient biases, we determine Taylor series expansions of the nonlinearities f (y ) and g (y ) about the estimated noiseless values yl. The generalized covariance matrix R can be approximately evaluated as [17] R "Ef[y(t)]u[y(t)]"Ef[yL (t)]u[yl (t)]#k WR Wk, (19) where k and k are diagonal matrices with entries k "Edf (t))/dy and k "Edg (t))/dy, respectively. Thus, a modified adaptive learning algorithm with reduced coefficient bias has the form W(t)"η(t)[I!f[y(t)]u[y(t)]#k W(t)R W(t)k ]W(t) "η(t)[i!f[y(t)]u[y(t)]#c W(t)R W(t)]W(t), (20) where C"[c ] is an nn scaling matrix with entries c "k k and means Hadamard product. In the special case when all of the source distributions are identical, f )"f ) i, g )"g ) i, and R "σ I, the bias correction term simplifies to B"σ k k WW. It is interesting to note that we can almost always select nonlinearities such that the global scaling coefficient c"k k can be close to zero for a wide class of signals. For example, when f )"y sign ) and g )"tanh(βy ) are chosen, or when f )"tanh(βy ) and g )"y sign ) are chosen, the scaling coefficient is equal to c"k k "rβey (t)[1!etanh(βy (t))] for r51, which is smaller over the range y 41 than would be the case if g )"y were chosen. Moreover, we can optimally design the parameters r and β so that within a specified range of y the absolute value of the scaling coefficient c"k k is minimal. Another possible solution to mitigate coefficient bias is to employ nonlinearities of the form fi )"f )!α y and g )"y with α 50. The motivation behind the use of linear terms!α y is to reduce the values of the scaling coefficients as c "k!α as well as to reduce the influence of large outliers. Alternatively, we can use the generalized Fahlman functions given by tanh(β y )!α y for either f )org ), where appropriate [18,19]. One disadvantage of these proposed techniques for bias removal is that a few equivariant properties for the resulting algorithm are lost when a bias compensating

7 term is added, and thus the algorithm may perform poorly or even fail to separate sources if the mixing matrix is very ill-conditioned. For this reason, it is necessary to design nonlinearities which correspond as closely as possible to those produced from the true p.d.f. s of the source signals while also maximally reducing the coefficient bias caused by noise Computer simulation experiments A. Cichocki et al./neurocomputing 22 (1998) We now illustrate the behavior of the bias removal algorithm in Eq. (20) via simulation. More illustrative examples are provided in [17]. In this example, a 33 mixing matrix given by H" (21) is employed. Three independent random sources one uniform-[!1,1]-distributed and two binary-$1-distributed are generated, and Eq. (2) is used to create x(t), where each ν(t) is a jointly Gaussian random vector with covariance R "σi with σ"0.01. The condition number of HEs(t)s(t)H is Here, f (y)"y and g (y)"y for all 14i43 and η(t)" Twenty trials were run, in which W(0) was a different random orthogonal matrix such that W(0)W(0)"0.25I, and ensemble averages were taken in each case. Fig. 1 shows the evolution of the performance factor ζ(t) defined as ζ(t)" 1 m b(t) max b (t)!1, l Ol for ioj (22) for each algorithm, where m"3 and b (t) is the (i, j)th element of the combined system matrix W(t)H. The value of ζ(t) measures the average source signal crosstalk in the output signals y (t) if no noise were present. As can be seen, the original algorithm yields a biased estimate of W(t), whereas the bias removal algorithm achieves a crosstalk level that is about 7 db lower. Also, shown for comparison is the original algorithm with no measurement noise, showing that the new algorithm s performance approaches this idealized case for small learning rates. 3. Recurrent neural network approach for noise cancellation 3.1. Basic concept and algorithm derivation Assume that we have successfully estimated an unbiased estimate of the separating matrix W via one of the previously described approaches. Then, we can estimate a mixing matrix HK "W"HPD, where W is the pseudo-inverse of W, P is any nn permutation matrix, and D is an nn non-singular diagonal scaling matrix. We now

8 120 A. Cichocki et al./neurocomputing 22 (1998) Fig. 1. Ensemble-averaged value of the performance factors for uncorrelated measurement noise in the first example: dotted line original algorithm (14) with noise, dashed line bias removal algorithm (20) with noise, continous line original algorithm (14) without noise. propose approaches for cancelling the effects of noise in the estimated source signals. In order to develop a viable neural network approach for noise cancellation, we define the error vector e(t)"x(t)!hk yl (t), (23) where e(t)"[e (t)2e (t)] and yl (t) an estimate of the source s(t). To compute yl (t), consider discussing the minimum entropy (ME) cost function E(e(t))"! Elog[p (e (t))], (24) where p (e ) is the true p.d.f. of the additive noise ν (t). It should be noted that we have assumed that the noise sources are i.i.d.; thus, stochastic gradient descent of the ME function yields stochastic independence of the error components as well as the minimization of their magnitude in an optimal way. The resulting system of differential equations is dyl (t) "μ(t)hk Ψ[e(t)], (25) dt

9 A. Cichocki et al./neurocomputing 22 (1998) where Ψ[e(t)]"[Ψ [e (t)]2ψ [e (t)]] with nonlinearities Ψ (e )"! log p (e ). (26) e A block diagram illustrating the implementation of the above algorithm is shown in Fig. 2, where Learning Algorithm denotes an appropriate bias removal learning rule (20). In the proposed algorithm, the optimal choices of nonlinearities Ψ (e ) depend on the noise distributions. Assume that all of the noise signals have generalized Gaussian distributions of the form [20] p (e )" r 2σ Γ(1/r ) exp!1 r e, (27) σ where r '0 is a variable parameter, Γ(r)"u exp(!u)du is the gamma function and σ"ee is a generalized measure of the noise variance known as the dispersion. Note that a unity value of r yields a Laplacian distribution, a value of r "2 yields the standard Gaussian distribution, and r PR yields a uniform distribution. In general, we can select any value of r 51, in which case the locally optimal nonlinear activation functions are of the form Ψ (e )"! log(p (e )) "e sign(e ), r 51. (28) e For very impulsive (spiky) sources with a high value of kurtosis, the optimal parameter r typically takes a value between zero and one. In such cases, we can use the modified activation functions Ψ (e )"e /[e #ε], where ε is a small positive constant, to avoid the singularity of the function at e "0. Moreover, when we do not Fig. 2. Neural network architecture for estimating the separating matrix and efficient noise reduction.

10 122 A. Cichocki et al./neurocomputing 22 (1998) have exact a priori knowledge about the noise distributions, we can adapt the value of r (t) for each error signal e (t) according to its estimated distance from Gaussianity. A simple gradient-based rule for adjusting each parameter r (t) is log(p (e )) r (t)"!η "η (29) r r 0.1r (t)#e (t)(1!log(e (t))) η. (30) r(t) Similar methods can be applied for other parameterized noise distributions. For example, when p (e ) is a generalized Cauchy distribution, then Ψ (e )"[(vr #1)/ (va(r )#e )]e sgn(e ). Similarly, for the generalized Rayleigh distribution, one obtains Ψ (e )"e e for complex-valued signals and coefficients. It should be noted that the continuous time algorithm in Eq. (25) can be easily converted to a discrete time algorithm as yl (t#1)"yl (t)#η(t)hk (t)ψ[e(t)]. (31) The proposed system in Fig. 2 can be considered as a form of nonlinear postprocessing that effectively reduces the additive noise component in the estimated source signals. In the next subsection, we propose a more efficient architecture that simultaneously estimates the mixing matrix H while reducing the amount of noise in the separated sources Simultaneous estimation of a mixing matrix and noise reduction Consider a recurrent neural network for noisy ICA with the same number of inputs and outputs (m"n) described by y(t)"x(t)!wk (t)y(t). (32) For m"n this model is equivalent to the previously described model since y(t)"[i#wk (t)]x(t)"w(t)x(t) with W(t)"[I#WK (t)]. It is easy to derive an equivariant learning algorithm for this algorithm, given by [13]. dwk (t) "!μ (t)[i#wk (t)][i!f[y(t)]u[y(t)]]. (33) dt Since the estimating mixing matrix HK can be expressed as HK "W"WK #I, (34) we replace the output vector y(t) by an improved estimate yl (t) to derive a novel learning algorithm as (see Fig. 3) dhk (t) dt "!μ (t)hk (t)[i!f[yl (t)]u[yl (t)]] (35)

11 A. Cichocki et al./neurocomputing 22 (1998) Fig. 3. Neural network architecture for simultaneous noise reduction and mixing matrix estimation for discrete-time t"k (k"0, 1, 2, 2 ). and dyl (t) "μ(t)hk Ψ[e(t)] (36) dt or in discrete time, and where HK (t)"hk (t#1)!hk (t)"η (t)hk (t)[i!f[yl (t)]u[yl (t)]] (37) yl (t#1)"yl (t)#η(t)hk (t)ψ[e(t)], (38) e(t)"x(t)!hk (t)sl (t) and x(t)"xl (t)#ν(t). A functional block diagram illustrating this algorithm s implementation is shown in Fig Pre-whitening and principal component analysis (PCA) As an enhancement to the above approaches, we could perform as preprocessing either pre-whitening or principal component analysis of the measured sensor signals either to reduce the effects of data conditioning or to reduce the effects of noise when m5n. This preprocessing step is represented in Fig. 3 by the nm matrix Q. Pre-whitening for noisy data can be performed using the learning algorithm (7) Q(t)"η(t)[I!xJ (t)xj (t)#σ Q(t)Q(t)]Q(t), (39)

12 124 A. Cichocki et al./neurocomputing 22 (1998) where xj (t)"qx(t)"q(hs(t)#ν(t)). Alternatively, for a nonsingular covariance matrix R "Ex(t)x(t) with m"n, we can use the standard numerical algorithm Q"ΛV"( )xx, (40) where ) denotes time averaging, Λ"diagλ 2λ is a diagonal matrix containing the n largest eigenvalues of R, and V"[ 2 ] is an orthogonal matrix of the corresponding eigenvectors of R. As an alternative to computing the PCA whitening matrix using a numerical algorithm, we could also use the fast adaptive RLS algorithm given by [10] (see Fig. 4) xn (t)" (t)x (t), (41) (t)"λη(t!1)#xn η (t#1)" (t)# xn (t) (t), (42) η(t) (k)!xn (t) (t)], (43) x (t)"x (t)!xn (t) *, (44) x (t)"x(t). (45) After applying the above PCA learning procedure, the output signals xn (t) are uncorrelated with variance λ "ExN. To normalize them to unit variance, we use the procedure xj (t)"λ xn (t)"λ x (t). (46) * It is interesting to note that after pre-whitening the global mixing matrix A"QH is an orthogonal matrix, since R "I for normalized sources and R J J "ExJ xj "AA#σI. Fig. 4. Functional block diagram illustrating implementation of the fast adaptive PCA learning algorithm.

13 4. Computer simulation experiments A. Cichocki et al./neurocomputing 22 (1998) Due to space we will present only two illustrative examples indicating the performance of the techniques. The three sub-gaussian source signals shown in Fig. 5 have been mixed using the mixing matrix whose rows are h "[0.8! ], h "[ ], and h "[! ]. Uncorrelated Gaussian noise signals Fig. 5. Exemplary on-line simulation results of the neural network in Fig. 3 for Gaussian noise. The first three signals are the original sources, the next three signals are the measured signals, and the last three signals are the estimated source signals using the learning rule in Eqs. (37) and (38). The horizontal axis represents time in seconds.

14 126 A. Cichocki et al./neurocomputing 22 (1998) with variance 1.6 was added to each of the elements of x(t). The neural network model depicted in Fig. 3 with associated learning rules in Eqs. (37) and (38) and nonlinearities f )"y, g )"tanh(10y ) and Ψ(e )"e was used to separate these signals, where HK (0)"I. Shown in Fig. 5 are the resulting separated signals, in which the source signals are accurately estimated. The resulting three rows of the combined system matrix HK H after 400 ms (with sampling period ) are [0.0034! ], [! !!0.0142] and Fig. 6. Exemplary on-line simulation results of the neural network in Fig. 3 for impulsive noise. The first three signals are the mixed sensors signals contaminated by impulsive (Laplacian) noise, the next three signals are the separated signals, using the learning rule (14) and the last three signals are the estimated source signals using the learning rule in Eqs. (37) and (38).

15 A. Cichocki et al./neurocomputing 22 (1998) [!0.2975!0.0061!0.0683], respectively, indicating that separation has been achieved. Note that standard algorithms that assume noiseless measurements fail to separate such noisy signals. In the second illustrative example the sensor signals were contaminated by additive impulsive (spiky) noise as is shown in Fig. 6. The same learning rule has been employed but with nonlinear functions Ψ(e )"tanh(10e ). The neural network of Fig. 3 was able to considerably reduce the influence of the noise in separating signals. 5. Conclusions In this paper, robust methods for performing independent component analysis in the presence of measurement noise are described. These methods simultaneously perform unbiased estimation of the separating matrix and noise reduction on the extracted sources. In addition, gradient-based rules for adjusting the shape parameters of the nonlinearities within the algorithms are given. Simulations indicate that the algorithms perform robust estimation of the independent components when noises are present. References [1] S. Amari, A. Cichocki, H.H. Yang, Recurrent neural networks for blind separation of sources, Proc. Int. Symp. on Nonlinear Theory and its Applications, NOLTA-95, Las Vegas, NV, 1995, pp [2] S. Amari, A. Cichocki, H.H. Yang, A new learning algorithm for blind signal separation, in: D.S. Touretzky, M.C. Mozer, M.E. Hasselmo (Eds.), Advances in Neural Information Processing Systems, MIT Press, Cambridge, MA, 1996, pp [3] J.J. Atick, A.N. Redlich, Convergent algorithm for sensory receptive field development, Neural Comput. 5 (1993) [4] A.J. Bell, T.J. Sejnowski, An information-maximization approach to blind separation and blind deconvolution, Neural Comput. 7 (1995) [5] J.F. Cardoso, B. Laheld, Equivariant adaptive source separation, IEEE Trans. Signal Process. 44 (1996) [6] A. Cichocki, S.C. Douglas, S. Amari, P. Mierzejewski, Independent component analysis for noisy data, in: C. Fyfe (Ed.), Proc. Int. Workshop on Independence and artificial Neural Networks, Tenerife, Spain, 9 10 February 1998, pp [7] A. Cichocki, W. Kasprzak, S. Amari, Adaptive approach to blind source separation with cancellation of additive and convolutional noise, ICSP 96, 3rd Int. Conf. on Signal Processing, Proc. IEEE Press/PHEI Beijing, vol. I, September 1996, pp [8] A. Cichocki, I. Sabala, S. Choi, B. Orsier, R. Szupiluk, Self-adaptive independent component analysis for sub-gaussian and super-gaussian mixtures with an unknown number of sources, Int. Symp. on Nonlinear Theory and Applications NOLTA 97, Honolulu, USA, 1997, pp [9] A. Cichocki, I. Sabala, S. Amari, Intelligent neural networks for blind signal separation with unknown number of sources, Proc. Int. Conf. Engineering of Intelligent Systems, Tenerife, Spain, February 1988, pp [10] A. Cichocki, R. Unbehauen, Robust estimation of principal components in real time, Electron. Lett. 29 (1993) [11] A. Cichocki, R. Unbehauen, E. Rummert, Robust learning algorithm for blind separation of signals, Electron. Lett. 30 (1994)

16 128 A. Cichocki et al./neurocomputing 22 (1998) [12] A. Cichocki, R. Unbehauen, L. Moszczyński, E. Rummert, A new on-line adaptive learning algorithm for blind separation of source signals, Proc. Int. Symp. Artificial Neural Networks, ISANN-94, Tainan, Taiwan, 1994, pp [13] A. Cichocki, R. Unbehauen, Robust neural networks with on-line learning for blind identification and blind separation of sources, IEEE Trans. Circuits Systems I 43 (1996) [14] P. Comon, Independent component analysis: a new concept?, Signal Process. 36 (1994) [15] S.C. Douglas, A. Cichocki, Neural networks for blind decorrelation of signals, IEEE Trans. Signal Process. 45 (1997) [16] S.C. Douglas, A. Cichocki, S. Amari, Multichannel blind separation and deconvolution of sources with arbitrary distributions, Proc. IEEE Workshop on Neural Networks for Signal Processing, Amelia Island, FL, 1987, pp [17] S.C. Douglas, A. Cichocki, S. Amari, Bias removal for blind source separation with noisy measurements, Electron. Lett. 34 (14) (1998) [18] M. Girolami, C. Fyfe, Stochastic ICA contrast maximization using Oja s nonlinear PCA algorithm, Int. J. Neural Systems (1997), in press. [19] M. Girolami, C. Fyfe, Extraction of independent signals sources using a deflationary exploratory projection pursuit network with lateral inhibition, IEE Proc. Vision, Image Signal Process. (1997), in press. [20] W.C. Gray, Variable norm deconvolution, Ph.D. Dissertation, Stanford Univ., Stanford, CA, [21] A. Hyvarinen, Independent component analysis in the presence of noise: a maximum likelihood approach, in: C. Fyfe (Ed.), Proc. Int. Workshop on Independence and Artificial Neural Networks, Tenerife, Spain, 9 10 February 1998, pp [22] J. Karhunen, Neural approaches to independent component analysis and source separation, Proc. European Symp. on Artificial Neural Networks, ESANN-96, Bruges, Belgium, 1996, pp [23] J. Karhunen, A. Cichocki, W. Kasprzak, P. Pajunen, On neural blind separation with noise suppression and redundancy reduction, Int. J. Neural Systems 8 (2) (1997) [24] D.J.C. MacKay, Maximum likelihood and covariant algorithms for independent component analysis, Internal Report, Cavendish Laboratory, Cambridge Univ., [25] Z. Maluche, O. Macchi, Adaptive separation of unknown number of sources, Proc. IEEE Workshop on Higher Order Statistics, Los Alamitos, CA, 1997, pp [26] K. Matsuoka, M. Ohya, M. Kawamoto, A neural net for blind separation of non-stationary signals, Neural Networks 8 (1995) [27] E. Moulines, J.F. Cardoso, E. Gassiat, Maximum likelihood for blind separation and deconvolution of noisy signals using mixture models, Proc. Int. Conf. Acoust. Speech, Signal Processing, ICASSP-97, Munich, Germany, 1997, pp [28] E. Oja, J. Karhunen, Signal separation by nonlinear Hebbian learning, in: M. Palaniswami et al. (Eds.), Computational Intelligence A Dynamic System Perspective, IEEE Press, New York, 1995, pp [29] E. Oja, The nonlinear PCA learning rule in independent component analysis, Neurocomputing 17 (1997) [30] L. Xu, C.-C. Cheung, J. Ruan, S. Amari, Nonlinearity and separation capability: further justification for the ICA algorithm with a learned mixture of parametric densities, Proc. European Symp. on Artificial Neural Networks, ESANN 97, Bruges, Belgium, 1997, pp [31] L. Xu, C.C. Chung, H.H. Yang, S. Amari, Independent component analysis by the informationtheoretic approach with mixture of parametric densities, Proc. IEEE Int. Conf. on Nural Networks, vol. III, Houston, TX, USA, 9 12 June, 1997, pp

17 A. Cichocki et al./neurocomputing 22 (1998) Andrzej Cichocki received the M.Sc. (with honors), Ph.D., and Habilitate Doctorate (Dr.Sc.) degrees, all in electrical engineering, from Warsaw University of Technology (Poland) in 1972, 1975, and 1982, respectively. Since 1972, he has been with the Institute of Theory of Electrical Engineering and Electrical Measurements at the Warsaw University of Technology, where he became a Professor in He is the co-author of two books: MOS Switched-Capacitor and Continuous-Time Integrated Circuits and Systems (Springer-Verlag, 1989) and Neural Networks for Optimization and Signal Processing (Teubner Wiley, 1993/94) and more than 150 research papers. He spent at University Erlangen-Nuernberg (Germany) a few years as Alexander Humboldt Research Fellow and Guest Professor, at Lehrstuhl fuer Allgemeine und Theortische Elektrotechnik directed by Professor Rolf Unbehauen. In he worked as a team leader of the laboratory for Artificial Brain Systems, at Frontier Research Program, Riken, Japan. He is currently working as a head of the laboratory for Open Information Systems, at Brain Science Institute, Riken in the Brain-Style Information Processing Group directed by Professor Shun-ichi Amari. He is currently an Associate Editor for IEEE ¹ransactions on Neural Networks. His current research interests include adaptive semi-blind signal processing, especially intelligent processing of biomedical signals, dynamic independent component analysis, neural networks and nonlinear dynamic systems theory. Scott C. Douglas received the B.S. (with distinction), M.S., and Ph.D. degrees in Electrical Engineering from Stanford University, Stanford, CA, in 1988, 1989, and 1992, respectively. From April to September 1992, he was with the Acoustics and Radar Technology Laboratory at SRI International, Menlo Park, CA. Since September 1992, he has been an assistant professor in the Department of Electrical Engineering at the University of Utah, Salt Lake City, UT. His research activities include adaptive filtering, active noise control, blind equalization and source separation, neural networks, and VLSI implementations of digital signal processing systems. Dr. Douglas received the Hughes Masters Fellowship Award in 1988 and the NSF Graduate Fellowship Award in He was a recipient of the NSF CAREER Award in He has published more than 60 articles in journals and conference proceedings. He served as a section editor for ¹he Digital Signal Processing Handbook (CRC/IEEE Press, 1998) and is currently an Associate Editor for IEEE Signal Processing etters. He has served on the organizing committees of several international conferences and workshops and co-founded the Signal Processing/Communications Chapter of the Utah Section of the IEEE. He is a frequent consultant to industry in the areas of signal processing and adaptive filtering and is a member of Phi Beta Kappa and Tau Beta Pi. Shun-ichi Amari graduated from the University of Tokyo in 1958 majoring in mathematical engineering and received the Dr.Eng. degree from the University of Tokyo in He has worked at Kyushu University and then at University of Tokyo at which he is now a Professor Emeritus. He is currently the Group Director of the Brain-Style Information Processing Group, Brain Science Institute, Riken (The Institute of Physical and Chemical Research), Japan. He has been engaged in research in wide areas of mathematical engineering or applied mathematics, such as topological network theory, differential geometry of continuum mechanics, pattern recognition and information sciences. In particular, he has devoted himself to mathematical foundations of neural network theory, including statistical neurodynamical, dynamical theory of neural fields, associative memory, self-organization, and general learning theory. Another main subject of his research is information geometry proposed by himself, which develops and applies modern differential geometry to statistical inference, information theory, and modern control theory, proposing a new powerful method of information sciences and probability theory. Dr. Amari is the Past President of the International Neural Networks Society. He was the Conference Chair of first International Joint Conference on Neural Networks, and he gave the Mahalanobis lecture on Differential Geometry of Statistical Inference at the 47th session of the International Statistical Inference at the 47th session of the International Statistical Institute. He is an IEEE Fellow, and the recipient of the 1995 Japan Academy Award, 1992 IEEE Neural Networks Pioneer Award, 1997 IEEE Piore Award.

ON SOME EXTENSIONS OF THE NATURAL GRADIENT ALGORITHM. Brain Science Institute, RIKEN, Wako-shi, Saitama , Japan

ON 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 information

Natural Gradient Learning for Over- and Under-Complete Bases in ICA

Natural Gradient Learning for Over- and Under-Complete Bases in ICA NOTE Communicated by Jean-François Cardoso Natural Gradient Learning for Over- and Under-Complete Bases in ICA Shun-ichi Amari RIKEN Brain Science Institute, Wako-shi, Hirosawa, Saitama 351-01, Japan Independent

More information

1 Introduction Independent component analysis (ICA) [10] is a statistical technique whose main applications are blind source separation, blind deconvo

1 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 information

where A 2 IR m n is the mixing matrix, s(t) is the n-dimensional source vector (n» m), and v(t) is additive white noise that is statistically independ

where A 2 IR m n is the mixing matrix, s(t) is the n-dimensional source vector (n» m), and v(t) is additive white noise that is statistically independ BLIND SEPARATION OF NONSTATIONARY AND TEMPORALLY CORRELATED SOURCES FROM NOISY MIXTURES Seungjin CHOI x and Andrzej CICHOCKI y x Department of Electrical Engineering Chungbuk National University, KOREA

More information

To appear in Proceedings of the ICA'99, Aussois, France, A 2 R mn is an unknown mixture matrix of full rank, v(t) is the vector of noises. The

To appear in Proceedings of the ICA'99, Aussois, France, A 2 R mn is an unknown mixture matrix of full rank, v(t) is the vector of noises. The To appear in Proceedings of the ICA'99, Aussois, France, 1999 1 NATURAL GRADIENT APPROACH TO BLIND SEPARATION OF OVER- AND UNDER-COMPLETE MIXTURES L.-Q. Zhang, S. Amari and A. Cichocki Brain-style Information

More information

ICA [6] ICA) [7, 8] ICA ICA ICA [9, 10] J-F. Cardoso. [13] Matlab ICA. Comon[3], Amari & Cardoso[4] ICA ICA

ICA [6] ICA) [7, 8] ICA ICA ICA [9, 10] J-F. Cardoso. [13] Matlab ICA. Comon[3], Amari & Cardoso[4] ICA ICA 16 1 (Independent Component Analysis: ICA) 198 9 ICA ICA ICA 1 ICA 198 Jutten Herault Comon[3], Amari & Cardoso[4] ICA Comon (PCA) projection persuit projection persuit ICA ICA ICA 1 [1] [] ICA ICA EEG

More information

MULTICHANNEL BLIND SEPARATION AND. Scott C. Douglas 1, Andrzej Cichocki 2, and Shun-ichi Amari 2

MULTICHANNEL BLIND SEPARATION AND. Scott C. Douglas 1, Andrzej Cichocki 2, and Shun-ichi Amari 2 MULTICHANNEL BLIND SEPARATION AND DECONVOLUTION OF SOURCES WITH ARBITRARY DISTRIBUTIONS Scott C. Douglas 1, Andrzej Cichoci, and Shun-ichi Amari 1 Department of Electrical Engineering, University of Utah

More information

Fundamentals of Principal Component Analysis (PCA), Independent Component Analysis (ICA), and Independent Vector Analysis (IVA)

Fundamentals of Principal Component Analysis (PCA), Independent Component Analysis (ICA), and Independent Vector Analysis (IVA) Fundamentals of Principal Component Analysis (PCA),, and Independent Vector Analysis (IVA) Dr Mohsen Naqvi Lecturer in Signal and Information Processing, School of Electrical and Electronic Engineering,

More information

Nonholonomic Orthogonal Learning Algorithms for Blind Source Separation

Nonholonomic Orthogonal Learning Algorithms for Blind Source Separation LETTER Communicated by Klaus Obermayer Nonholonomic Orthogonal Learning Algorithms for Blind Source Separation Shun-ichi Amari Tian-Ping Chen Andrzej Cichocki RIKEN Brain Science Institute, Brain-Style

More information

Independent Component Analysis and Its Applications. By Qing Xue, 10/15/2004

Independent Component Analysis and Its Applications. By Qing Xue, 10/15/2004 Independent Component Analysis and Its Applications By Qing Xue, 10/15/2004 Outline Motivation of ICA Applications of ICA Principles of ICA estimation Algorithms for ICA Extensions of basic ICA framework

More information

Neural networks for blind separation with unknown number of sources

Neural networks for blind separation with unknown number of sources Neurocomputing 24 (1999) 55 93 Neural networks for blind separation with unknown number of sources Andrzej Cichocki*, Juha Karhunen, Wlodzimierz Kasprzak, Ricardo Vigário Laboratory for Open Information

More information

Independent Component Analysis. Contents

Independent 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 information

An Improved Cumulant Based Method for Independent Component Analysis

An Improved Cumulant Based Method for Independent Component Analysis An Improved Cumulant Based Method for Independent Component Analysis Tobias Blaschke and Laurenz Wiskott Institute for Theoretical Biology Humboldt University Berlin Invalidenstraße 43 D - 0 5 Berlin Germany

More information

Gatsby Theoretical Neuroscience Lectures: Non-Gaussian statistics and natural images Parts I-II

Gatsby Theoretical Neuroscience Lectures: Non-Gaussian statistics and natural images Parts I-II Gatsby Theoretical Neuroscience Lectures: Non-Gaussian statistics and natural images Parts I-II Gatsby Unit University College London 27 Feb 2017 Outline Part I: Theory of ICA Definition and difference

More information

Massoud BABAIE-ZADEH. Blind Source Separation (BSS) and Independent Componen Analysis (ICA) p.1/39

Massoud BABAIE-ZADEH. Blind Source Separation (BSS) and Independent Componen Analysis (ICA) p.1/39 Blind Source Separation (BSS) and Independent Componen Analysis (ICA) Massoud BABAIE-ZADEH Blind Source Separation (BSS) and Independent Componen Analysis (ICA) p.1/39 Outline Part I Part II Introduction

More information

1 Introduction Blind source separation (BSS) is a fundamental problem which is encountered in a variety of signal processing problems where multiple s

1 Introduction Blind source separation (BSS) is a fundamental problem which is encountered in a variety of signal processing problems where multiple s Blind Separation of Nonstationary Sources in Noisy Mixtures Seungjin CHOI x1 and Andrzej CICHOCKI y x Department of Electrical Engineering Chungbuk National University 48 Kaeshin-dong, Cheongju Chungbuk

More information

A METHOD OF ICA IN TIME-FREQUENCY DOMAIN

A METHOD OF ICA IN TIME-FREQUENCY DOMAIN A METHOD OF ICA IN TIME-FREQUENCY DOMAIN Shiro Ikeda PRESTO, JST Hirosawa 2-, Wako, 35-98, Japan Shiro.Ikeda@brain.riken.go.jp Noboru Murata RIKEN BSI Hirosawa 2-, Wako, 35-98, Japan Noboru.Murata@brain.riken.go.jp

More information

PROPERTIES OF THE EMPIRICAL CHARACTERISTIC FUNCTION AND ITS APPLICATION TO TESTING FOR INDEPENDENCE. Noboru Murata

PROPERTIES OF THE EMPIRICAL CHARACTERISTIC FUNCTION AND ITS APPLICATION TO TESTING FOR INDEPENDENCE. Noboru Murata ' / PROPERTIES OF THE EMPIRICAL CHARACTERISTIC FUNCTION AND ITS APPLICATION TO TESTING FOR INDEPENDENCE Noboru Murata Waseda University Department of Electrical Electronics and Computer Engineering 3--

More information

One-unit Learning Rules for Independent Component Analysis

One-unit Learning Rules for Independent Component Analysis One-unit Learning Rules for Independent Component Analysis Aapo Hyvarinen and Erkki Oja Helsinki University of Technology Laboratory of Computer and Information Science Rakentajanaukio 2 C, FIN-02150 Espoo,

More information

TWO METHODS FOR ESTIMATING OVERCOMPLETE INDEPENDENT COMPONENT BASES. Mika Inki and Aapo Hyvärinen

TWO METHODS FOR ESTIMATING OVERCOMPLETE INDEPENDENT COMPONENT BASES. Mika Inki and Aapo Hyvärinen TWO METHODS FOR ESTIMATING OVERCOMPLETE INDEPENDENT COMPONENT BASES Mika Inki and Aapo Hyvärinen Neural Networks Research Centre Helsinki University of Technology P.O. Box 54, FIN-215 HUT, Finland ABSTRACT

More information

SPARSE REPRESENTATION AND BLIND DECONVOLUTION OF DYNAMICAL SYSTEMS. Liqing Zhang and Andrzej Cichocki

SPARSE REPRESENTATION AND BLIND DECONVOLUTION OF DYNAMICAL SYSTEMS. Liqing Zhang and Andrzej Cichocki SPARSE REPRESENTATON AND BLND DECONVOLUTON OF DYNAMCAL SYSTEMS Liqing Zhang and Andrzej Cichocki Lab for Advanced Brain Signal Processing RKEN Brain Science nstitute Wako shi, Saitama, 351-198, apan zha,cia

More information

CIFAR Lectures: Non-Gaussian statistics and natural images

CIFAR Lectures: Non-Gaussian statistics and natural images CIFAR Lectures: Non-Gaussian statistics and natural images Dept of Computer Science University of Helsinki, Finland Outline Part I: Theory of ICA Definition and difference to PCA Importance of non-gaussianity

More information

A Constrained EM Algorithm for Independent Component Analysis

A Constrained EM Algorithm for Independent Component Analysis LETTER Communicated by Hagai Attias A Constrained EM Algorithm for Independent Component Analysis Max Welling Markus Weber California Institute of Technology, Pasadena, CA 91125, U.S.A. We introduce a

More information

Recursive Generalized Eigendecomposition for Independent Component Analysis

Recursive 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 information

Analytical solution of the blind source separation problem using derivatives

Analytical solution of the blind source separation problem using derivatives Analytical solution of the blind source separation problem using derivatives Sebastien Lagrange 1,2, Luc Jaulin 2, Vincent Vigneron 1, and Christian Jutten 1 1 Laboratoire Images et Signaux, Institut National

More information

Estimation of linear non-gaussian acyclic models for latent factors

Estimation of linear non-gaussian acyclic models for latent factors Estimation of linear non-gaussian acyclic models for latent factors Shohei Shimizu a Patrik O. Hoyer b Aapo Hyvärinen b,c a The Institute of Scientific and Industrial Research, Osaka University Mihogaoka

More information

Advanced Introduction to Machine Learning CMU-10715

Advanced Introduction to Machine Learning CMU-10715 Advanced Introduction to Machine Learning CMU-10715 Independent Component Analysis Barnabás Póczos Independent Component Analysis 2 Independent Component Analysis Model original signals Observations (Mixtures)

More information

MULTICHANNEL SIGNAL PROCESSING USING SPATIAL RANK COVARIANCE MATRICES

MULTICHANNEL SIGNAL PROCESSING USING SPATIAL RANK COVARIANCE MATRICES MULTICHANNEL SIGNAL PROCESSING USING SPATIAL RANK COVARIANCE MATRICES S. Visuri 1 H. Oja V. Koivunen 1 1 Signal Processing Lab. Dept. of Statistics Tampere Univ. of Technology University of Jyväskylä P.O.

More information

Constrained Projection Approximation Algorithms for Principal Component Analysis

Constrained Projection Approximation Algorithms for Principal Component Analysis Constrained Projection Approximation Algorithms for Principal Component Analysis Seungjin Choi, Jong-Hoon Ahn, Andrzej Cichocki Department of Computer Science, Pohang University of Science and Technology,

More information

Non-Euclidean Independent Component Analysis and Oja's Learning

Non-Euclidean Independent Component Analysis and Oja's Learning Non-Euclidean Independent Component Analysis and Oja's Learning M. Lange 1, M. Biehl 2, and T. Villmann 1 1- University of Appl. Sciences Mittweida - Dept. of Mathematics Mittweida, Saxonia - Germany 2-

More information

Principal Component Analysis

Principal Component Analysis Principal Component Analysis Introduction Consider a zero mean random vector R n with autocorrelation matri R = E( T ). R has eigenvectors q(1),,q(n) and associated eigenvalues λ(1) λ(n). Let Q = [ q(1)

More information

Blind Extraction of Singularly Mixed Source Signals

Blind Extraction of Singularly Mixed Source Signals IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL 11, NO 6, NOVEMBER 2000 1413 Blind Extraction of Singularly Mixed Source Signals Yuanqing Li, Jun Wang, Senior Member, IEEE, and Jacek M Zurada, Fellow, IEEE Abstract

More information

Second-order statistics based blind source separation using a bank of subband filters

Second-order statistics based blind source separation using a bank of subband filters Digital Signal Processing 13 (2003) 252 274 www.elsevier.com/locate/dsp Second-order statistics based blind source separation using a bank of subband filters R.R. Gharieb a,b, and A. Cichocki a,c a Laboratory

More information

On Information Maximization and Blind Signal Deconvolution

On 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 information

x 1 (t) Spectrogram t s

x 1 (t) Spectrogram t s A METHOD OF ICA IN TIME-FREQUENCY DOMAIN Shiro Ikeda PRESTO, JST Hirosawa 2-, Wako, 35-98, Japan Shiro.Ikeda@brain.riken.go.jp Noboru Murata RIKEN BSI Hirosawa 2-, Wako, 35-98, Japan Noboru.Murata@brain.riken.go.jp

More information

Undercomplete Independent Component. Analysis for Signal Separation and. Dimension Reduction. Category: Algorithms and Architectures.

Undercomplete Independent Component. Analysis for Signal Separation and. Dimension Reduction. Category: Algorithms and Architectures. Undercomplete Independent Component Analysis for Signal Separation and Dimension Reduction John Porrill and James V Stone Psychology Department, Sheeld University, Sheeld, S10 2UR, England. Tel: 0114 222

More information

A Canonical Genetic Algorithm for Blind Inversion of Linear Channels

A 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 information

IEICE TRANS. FUNDAMENTALS, VOL.E83 A, NO.12 DECEMBER PAPER Natural Gradient Learning for Spatio-temporal Decorrelation: Recurrent Network Λ Seu

IEICE TRANS. FUNDAMENTALS, VOL.E83 A, NO.12 DECEMBER PAPER Natural Gradient Learning for Spatio-temporal Decorrelation: Recurrent Network Λ Seu PAPER Natural Gradient Learning for Spatio-temporal Decorrelation: Recurrent Network Λ Seungjin CHOI y, Nonmember, Shunichi AMARI yy, and Andrzej CICHOCKI yy, Members SUMMARY Spatio-temporal decorrelation

More information

ACENTRAL problem in neural-network research, as well

ACENTRAL problem in neural-network research, as well 626 IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 10, NO. 3, MAY 1999 Fast and Robust Fixed-Point Algorithms for Independent Component Analysis Aapo Hyvärinen Abstract Independent component analysis (ICA)

More information

Independent Component Analysis on the Basis of Helmholtz Machine

Independent Component Analysis on the Basis of Helmholtz Machine Independent Component Analysis on the Basis of Helmholtz Machine Masashi OHATA *1 ohatama@bmc.riken.go.jp Toshiharu MUKAI *1 tosh@bmc.riken.go.jp Kiyotoshi MATSUOKA *2 matsuoka@brain.kyutech.ac.jp *1 Biologically

More information

Independent Component Analysis of Incomplete Data

Independent Component Analysis of Incomplete Data Independent Component Analysis of Incomplete Data Max Welling Markus Weber California Institute of Technology 136-93 Pasadena, CA 91125 fwelling,rmwg@vision.caltech.edu Keywords: EM, Missing Data, ICA

More information

Simultaneous Diagonalization in the Frequency Domain (SDIF) for Source Separation

Simultaneous 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 information

ORIENTED PCA AND BLIND SIGNAL SEPARATION

ORIENTED PCA AND BLIND SIGNAL SEPARATION ORIENTED PCA AND BLIND SIGNAL SEPARATION K. I. Diamantaras Department of Informatics TEI of Thessaloniki Sindos 54101, Greece kdiamant@it.teithe.gr Th. Papadimitriou Department of Int. Economic Relat.

More information

Using Kernel PCA for Initialisation of Variational Bayesian Nonlinear Blind Source Separation Method

Using Kernel PCA for Initialisation of Variational Bayesian Nonlinear Blind Source Separation Method Using Kernel PCA for Initialisation of Variational Bayesian Nonlinear Blind Source Separation Method Antti Honkela 1, Stefan Harmeling 2, Leo Lundqvist 1, and Harri Valpola 1 1 Helsinki University of Technology,

More information

Independent Component Analysis and Blind Source Separation

Independent Component Analysis and Blind Source Separation Independent Component Analysis and Blind Source Separation Aapo Hyvärinen University of Helsinki and Helsinki Institute of Information Technology 1 Blind source separation Four source signals : 1.5 2 3

More information

BLIND SEPARATION OF INSTANTANEOUS MIXTURES OF NON STATIONARY SOURCES

BLIND SEPARATION OF INSTANTANEOUS MIXTURES OF NON STATIONARY SOURCES BLIND SEPARATION OF INSTANTANEOUS MIXTURES OF NON STATIONARY SOURCES Dinh-Tuan Pham Laboratoire de Modélisation et Calcul URA 397, CNRS/UJF/INPG BP 53X, 38041 Grenoble cédex, France Dinh-Tuan.Pham@imag.fr

More information

Independent Component Analysis. PhD Seminar Jörgen Ungh

Independent Component Analysis. PhD Seminar Jörgen Ungh Independent Component Analysis PhD Seminar Jörgen Ungh Agenda Background a motivater Independence ICA vs. PCA Gaussian data ICA theory Examples Background & motivation The cocktail party problem Bla bla

More information

Blind Machine Separation Te-Won Lee

Blind Machine Separation Te-Won Lee Blind Machine Separation Te-Won Lee University of California, San Diego Institute for Neural Computation Blind Machine Separation Problem we want to solve: Single microphone blind source separation & deconvolution

More information

ON-LINE BLIND SEPARATION OF NON-STATIONARY SIGNALS

ON-LINE BLIND SEPARATION OF NON-STATIONARY SIGNALS Yugoslav Journal of Operations Research 5 (25), Number, 79-95 ON-LINE BLIND SEPARATION OF NON-STATIONARY SIGNALS Slavica TODOROVIĆ-ZARKULA EI Professional Electronics, Niš, bssmtod@eunet.yu Branimir TODOROVIĆ,

More information

BLIND DECONVOLUTION ALGORITHMS FOR MIMO-FIR SYSTEMS DRIVEN BY FOURTH-ORDER COLORED SIGNALS

BLIND DECONVOLUTION ALGORITHMS FOR MIMO-FIR SYSTEMS DRIVEN BY FOURTH-ORDER COLORED SIGNALS BLIND DECONVOLUTION ALGORITHMS FOR MIMO-FIR SYSTEMS DRIVEN BY FOURTH-ORDER COLORED SIGNALS M. Kawamoto 1,2, Y. Inouye 1, A. Mansour 2, and R.-W. Liu 3 1. Department of Electronic and Control Systems Engineering,

More information

Independent Component Analysis (ICA) Bhaskar D Rao University of California, San Diego

Independent Component Analysis (ICA) Bhaskar D Rao University of California, San Diego Independent Component Analysis (ICA) Bhaskar D Rao University of California, San Diego Email: brao@ucsdedu References 1 Hyvarinen, A, Karhunen, J, & Oja, E (2004) Independent component analysis (Vol 46)

More information

Bayesian ensemble learning of generative models

Bayesian ensemble learning of generative models Chapter Bayesian ensemble learning of generative models Harri Valpola, Antti Honkela, Juha Karhunen, Tapani Raiko, Xavier Giannakopoulos, Alexander Ilin, Erkki Oja 65 66 Bayesian ensemble learning of generative

More information

1 Introduction Consider the following: given a cost function J (w) for the parameter vector w = [w1 w2 w n ] T, maximize J (w) (1) such that jjwjj = C

1 Introduction Consider the following: given a cost function J (w) for the parameter vector w = [w1 w2 w n ] T, maximize J (w) (1) such that jjwjj = C On Gradient Adaptation With Unit-Norm Constraints Scott C. Douglas 1, Shun-ichi Amari 2, and S.-Y. Kung 3 1 Department of Electrical Engineering, Southern Methodist University Dallas, Texas 75275 USA 2

More information

Robust extraction of specific signals with temporal structure

Robust 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 information

BLIND SEPARATION USING ABSOLUTE MOMENTS BASED ADAPTIVE ESTIMATING FUNCTION. Juha Karvanen and Visa Koivunen

BLIND SEPARATION USING ABSOLUTE MOMENTS BASED ADAPTIVE ESTIMATING FUNCTION. Juha Karvanen and Visa Koivunen BLIND SEPARATION USING ABSOLUTE MOMENTS BASED ADAPTIVE ESTIMATING UNCTION Juha Karvanen and Visa Koivunen Signal Processing Laboratory Helsinki University of Technology P.O. Box 3, IN-215 HUT, inland Tel.

More information

BLIND SEPARATION OF POSITIVE SOURCES USING NON-NEGATIVE PCA

BLIND SEPARATION OF POSITIVE SOURCES USING NON-NEGATIVE PCA BLIND SEPARATION OF POSITIVE SOURCES USING NON-NEGATIVE PCA Erkki Oja Neural Networks Research Centre Helsinki University of Technology P.O.Box 54, 215 HUT, Finland erkki.oja@hut.fi Mark Plumbley Department

More information

Independent Component Analysis

Independent Component Analysis Independent Component Analysis Seungjin Choi Department of Computer Science Pohang University of Science and Technology, Korea seungjin@postech.ac.kr March 4, 2009 1 / 78 Outline Theory and Preliminaries

More information

Blind Source Separation for Changing Source Number: A Neural Network Approach with a Variable Structure 1

Blind Source Separation for Changing Source Number: A Neural Network Approach with a Variable Structure 1 Blind Source Separation for Changing Source Number: A Neural Network Approach with a Variable Structure 1 Shun-Tian Lou, Member, IEEE and Xian-Da Zhang, Senior Member, IEEE Key Lab for Radar Signal Processing,

More information

Speed and Accuracy Enhancement of Linear ICA Techniques Using Rational Nonlinear Functions

Speed and Accuracy Enhancement of Linear ICA Techniques Using Rational Nonlinear Functions Speed and Accuracy Enhancement of Linear ICA Techniques Using Rational Nonlinear Functions Petr Tichavský 1, Zbyněk Koldovský 1,2, and Erkki Oja 3 1 Institute of Information Theory and Automation, Pod

More information

A Convex Cauchy-Schwarz Divergence Measure for Blind Source Separation

A Convex Cauchy-Schwarz Divergence Measure for Blind Source Separation INTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING Volume, 8 A Convex Cauchy-Schwarz Divergence Measure for Blind Source Separation Zaid Albataineh and Fathi M. Salem Abstract We propose

More information

Independent Component Analysis

Independent Component Analysis Independent Component Analysis James V. Stone November 4, 24 Sheffield University, Sheffield, UK Keywords: independent component analysis, independence, blind source separation, projection pursuit, complexity

More information

ADAPTIVE LATERAL INHIBITION FOR NON-NEGATIVE ICA. Mark Plumbley

ADAPTIVE LATERAL INHIBITION FOR NON-NEGATIVE ICA. Mark Plumbley Submitteed to the International Conference on Independent Component Analysis and Blind Signal Separation (ICA2) ADAPTIVE LATERAL INHIBITION FOR NON-NEGATIVE ICA Mark Plumbley Audio & Music Lab Department

More information

Unsupervised learning: beyond simple clustering and PCA

Unsupervised learning: beyond simple clustering and PCA Unsupervised learning: beyond simple clustering and PCA Liza Rebrova Self organizing maps (SOM) Goal: approximate data points in R p by a low-dimensional manifold Unlike PCA, the manifold does not have

More information

Blind Source Separation Using Artificial immune system

Blind 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 information

MINIMIZATION-PROJECTION (MP) APPROACH FOR BLIND SOURCE SEPARATION IN DIFFERENT MIXING MODELS

MINIMIZATION-PROJECTION (MP) APPROACH FOR BLIND SOURCE SEPARATION IN DIFFERENT MIXING MODELS MINIMIZATION-PROJECTION (MP) APPROACH FOR BLIND SOURCE SEPARATION IN DIFFERENT MIXING MODELS Massoud Babaie-Zadeh ;2, Christian Jutten, Kambiz Nayebi 2 Institut National Polytechnique de Grenoble (INPG),

More information

Comparative Analysis of ICA Based Features

Comparative Analysis of ICA Based Features International Journal of Emerging Engineering Research and Technology Volume 2, Issue 7, October 2014, PP 267-273 ISSN 2349-4395 (Print) & ISSN 2349-4409 (Online) Comparative Analysis of ICA Based Features

More information

Emergence of Phase- and Shift-Invariant Features by Decomposition of Natural Images into Independent Feature Subspaces

Emergence of Phase- and Shift-Invariant Features by Decomposition of Natural Images into Independent Feature Subspaces LETTER Communicated by Bartlett Mel Emergence of Phase- and Shift-Invariant Features by Decomposition of Natural Images into Independent Feature Subspaces Aapo Hyvärinen Patrik Hoyer Helsinki University

More information

Independent Component Analysis and Its Application on Accelerator Physics

Independent Component Analysis and Its Application on Accelerator Physics Independent Component Analysis and Its Application on Accelerator Physics Xiaoying Pang LA-UR-12-20069 ICA and PCA Similarities: Blind source separation method (BSS) no model Observed signals are linear

More information

Blind signal processing algorithms

Blind signal processing algorithms 12th Int. Workshop on Systems, Signals & Image Processing, 22-24 September 2005, Chalkida, Greece 105 Blind signal processing algorithms Athanasios Margaris and Efthimios Kotsialos Department of Applied

More information

CONTROL SYSTEMS ANALYSIS VIA BLIND SOURCE DECONVOLUTION. Kenji Sugimoto and Yoshito Kikkawa

CONTROL SYSTEMS ANALYSIS VIA BLIND SOURCE DECONVOLUTION. Kenji Sugimoto and Yoshito Kikkawa CONTROL SYSTEMS ANALYSIS VIA LIND SOURCE DECONVOLUTION Kenji Sugimoto and Yoshito Kikkawa Nara Institute of Science and Technology Graduate School of Information Science 896-5 Takayama-cho, Ikoma-city,

More information

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 54, NO. 2, FEBRUARY

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 54, NO. 2, FEBRUARY IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL 54, NO 2, FEBRUARY 2006 423 Underdetermined Blind Source Separation Based on Sparse Representation Yuanqing Li, Shun-Ichi Amari, Fellow, IEEE, Andrzej Cichocki,

More information

Higher Order Statistics

Higher Order Statistics Higher Order Statistics Matthias Hennig Neural Information Processing School of Informatics, University of Edinburgh February 12, 2018 1 0 Based on Mark van Rossum s and Chris Williams s old NIP slides

More information

Independent Component Analysis

Independent Component Analysis A Short Introduction to Independent Component Analysis with Some Recent Advances Aapo Hyvärinen Dept of Computer Science Dept of Mathematics and Statistics University of Helsinki Problem of blind source

More information

Comparative Performance Analysis of Three Algorithms for Principal Component Analysis

Comparative Performance Analysis of Three Algorithms for Principal Component Analysis 84 R. LANDQVIST, A. MOHAMMED, COMPARATIVE PERFORMANCE ANALYSIS OF THR ALGORITHMS Comparative Performance Analysis of Three Algorithms for Principal Component Analysis Ronnie LANDQVIST, Abbas MOHAMMED Dept.

More information

HST.582J / 6.555J / J Biomedical Signal and Image Processing Spring 2007

HST.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 information

ARTEFACT DETECTION IN ASTROPHYSICAL IMAGE DATA USING INDEPENDENT COMPONENT ANALYSIS. Maria Funaro, Erkki Oja, and Harri Valpola

ARTEFACT DETECTION IN ASTROPHYSICAL IMAGE DATA USING INDEPENDENT COMPONENT ANALYSIS. Maria Funaro, Erkki Oja, and Harri Valpola ARTEFACT DETECTION IN ASTROPHYSICAL IMAGE DATA USING INDEPENDENT COMPONENT ANALYSIS Maria Funaro, Erkki Oja, and Harri Valpola Neural Networks Research Centre, Helsinki University of Technology P.O.Box

More information

Independent component analysis: algorithms and applications

Independent component analysis: algorithms and applications PERGAMON Neural Networks 13 (2000) 411 430 Invited article Independent component analysis: algorithms and applications A. Hyvärinen, E. Oja* Neural Networks Research Centre, Helsinki University of Technology,

More information

Gaussian process for nonstationary time series prediction

Gaussian process for nonstationary time series prediction Computational Statistics & Data Analysis 47 (2004) 705 712 www.elsevier.com/locate/csda Gaussian process for nonstationary time series prediction Soane Brahim-Belhouari, Amine Bermak EEE Department, Hong

More information

Blind channel deconvolution of real world signals using source separation techniques

Blind channel deconvolution of real world signals using source separation techniques Blind channel deconvolution of real world signals using source separation techniques Jordi Solé-Casals 1, Enric Monte-Moreno 2 1 Signal Processing Group, University of Vic, Sagrada Família 7, 08500, Vic

More information

Blind separation of sources that have spatiotemporal variance dependencies

Blind separation of sources that have spatiotemporal variance dependencies Blind separation of sources that have spatiotemporal variance dependencies Aapo Hyvärinen a b Jarmo Hurri a a Neural Networks Research Centre, Helsinki University of Technology, Finland b Helsinki Institute

More information

Nonlinear Blind Source Separation Using a Radial Basis Function Network

Nonlinear Blind Source Separation Using a Radial Basis Function Network 124 IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 12, NO. 1, JANUARY 2001 Nonlinear Blind Source Separation Using a Radial Basis Function Network Ying Tan, Member, IEEE, Jun Wang, Senior Member, IEEE, and

More information

Neural Learning in Structured Parameter Spaces Natural Riemannian Gradient

Neural Learning in Structured Parameter Spaces Natural Riemannian Gradient Neural Learning in Structured Parameter Spaces Natural Riemannian Gradient Shun-ichi Amari RIKEN Frontier Research Program, RIKEN, Hirosawa 2-1, Wako-shi 351-01, Japan amari@zoo.riken.go.jp Abstract The

More information

Independent Component Analysis

Independent Component Analysis A Short Introduction to Independent Component Analysis Aapo Hyvärinen Helsinki Institute for Information Technology and Depts of Computer Science and Psychology University of Helsinki Problem of blind

More information

HST.582J/6.555J/16.456J

HST.582J/6.555J/16.456J Blind Source Separation: PCA & ICA HST.582J/6.555J/16.456J Gari D. Clifford gari [at] mit. edu http://www.mit.edu/~gari G. D. Clifford 2005-2009 What is BSS? Assume an observation (signal) is a linear

More information

Independent Component Analysis (ICA)

Independent Component Analysis (ICA) Independent Component Analysis (ICA) Université catholique de Louvain (Belgium) Machine Learning Group http://www.dice.ucl ucl.ac.be/.ac.be/mlg/ 1 Overview Uncorrelation vs Independence Blind source separation

More information

NONLINEAR INDEPENDENT FACTOR ANALYSIS BY HIERARCHICAL MODELS

NONLINEAR INDEPENDENT FACTOR ANALYSIS BY HIERARCHICAL MODELS NONLINEAR INDEPENDENT FACTOR ANALYSIS BY HIERARCHICAL MODELS Harri Valpola, Tomas Östman and Juha Karhunen Helsinki University of Technology, Neural Networks Research Centre P.O. Box 5400, FIN-02015 HUT,

More information

POLYNOMIAL SINGULAR VALUES FOR NUMBER OF WIDEBAND SOURCES ESTIMATION AND PRINCIPAL COMPONENT ANALYSIS

POLYNOMIAL SINGULAR VALUES FOR NUMBER OF WIDEBAND SOURCES ESTIMATION AND PRINCIPAL COMPONENT ANALYSIS POLYNOMIAL SINGULAR VALUES FOR NUMBER OF WIDEBAND SOURCES ESTIMATION AND PRINCIPAL COMPONENT ANALYSIS Russell H. Lambert RF and Advanced Mixed Signal Unit Broadcom Pasadena, CA USA russ@broadcom.com Marcel

More information

GENERALIZED DEFLATION ALGORITHMS FOR THE BLIND SOURCE-FACTOR SEPARATION OF MIMO-FIR CHANNELS. Mitsuru Kawamoto 1,2 and Yujiro Inouye 1

GENERALIZED DEFLATION ALGORITHMS FOR THE BLIND SOURCE-FACTOR SEPARATION OF MIMO-FIR CHANNELS. Mitsuru Kawamoto 1,2 and Yujiro Inouye 1 GENERALIZED DEFLATION ALGORITHMS FOR THE BLIND SOURCE-FACTOR SEPARATION OF MIMO-FIR CHANNELS Mitsuru Kawamoto,2 and Yuiro Inouye. Dept. of Electronic and Control Systems Engineering, Shimane University,

More information

Blind Source Separation with a Time-Varying Mixing Matrix

Blind Source Separation with a Time-Varying Mixing Matrix Blind Source Separation with a Time-Varying Mixing Matrix Marcus R DeYoung and Brian L Evans Department of Electrical and Computer Engineering The University of Texas at Austin 1 University Station, Austin,

More information

Independent Component Analysis and Unsupervised Learning

Independent 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 information

Introduction to Machine Learning

Introduction to Machine Learning 10-701 Introduction to Machine Learning PCA Slides based on 18-661 Fall 2018 PCA Raw data can be Complex, High-dimensional To understand a phenomenon we measure various related quantities If we knew what

More information

Independent Component Analysis (ICA)

Independent Component Analysis (ICA) Independent Component Analysis (ICA) 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

More information

Artificial Intelligence Module 2. Feature Selection. Andrea Torsello

Artificial Intelligence Module 2. Feature Selection. Andrea Torsello Artificial Intelligence Module 2 Feature Selection Andrea Torsello We have seen that high dimensional data is hard to classify (curse of dimensionality) Often however, the data does not fill all the space

More information

BLIND SEPARATION OF INSTANTANEOUS MIXTURES OF NON STATIONARY SOURCES

BLIND SEPARATION OF INSTANTANEOUS MIXTURES OF NON STATIONARY SOURCES BLIND SEPARATION OF INSTANTANEOUS MIXTURES OF NON STATIONARY SOURCES Dinh-Tuan Pham Laboratoire de Modélisation et Calcul URA 397, CNRS/UJF/INPG BP 53X, 38041 Grenoble cédex, France Dinh-Tuan.Pham@imag.fr

More information

Introduction PCA classic Generative models Beyond and summary. PCA, ICA and beyond

Introduction PCA classic Generative models Beyond and summary. PCA, ICA and beyond PCA, ICA and beyond Summer School on Manifold Learning in Image and Signal Analysis, August 17-21, 2009, Hven Technical University of Denmark (DTU) & University of Copenhagen (KU) August 18, 2009 Motivation

More information

A Novel Approach For Sensor Noise Elimination In Real Time Process

A Novel Approach For Sensor Noise Elimination In Real Time Process A Novel Approach For Sensor Noise Elimination In Real Time Process K.UMASURESH *, K.SURESH MANIC Abstract A review on the current literature on sensor noise elimination algorithms, reports the eistence

More information

Independent Component Analysis and Unsupervised Learning. Jen-Tzung Chien

Independent 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 information

Second Order Nonstationary Source Separation

Second Order Nonstationary Source Separation Journal of VLSI Signal Processing,?, 1 13 (2001) cfl 2001 Kluwer Academic Publishers, Boston. Manufactured in The Netherlands. Second Order Nonstationary Source Separation SEUNGJIN CHOI Department of Computer

More information

Equivalence Probability and Sparsity of Two Sparse Solutions in Sparse Representation

Equivalence Probability and Sparsity of Two Sparse Solutions in Sparse Representation IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 19, NO. 12, DECEMBER 2008 2009 Equivalence Probability and Sparsity of Two Sparse Solutions in Sparse Representation Yuanqing Li, Member, IEEE, Andrzej Cichocki,

More information