Showing papers on "Independent component analysis published in 2013"
Book•
16 Feb 2013TL;DR: This work presents a Unifying Information-Theoretic Framework for ICA, a novel and scalable framework for independent component analysis that combines supervised and unsupervised classification with ICA Mixture Models.
Abstract: . Preface. Acknowledgments. List of Figures. List of Tables. Abbreviations and Symbols. Introduction. Part I: Independent Component Analysis: Theory. 1. Basics. 2. Independent Component Analysis. 3. A Unifying Information-Theoretic Framework for ICA. 4. Blind Separation of Time-Delayed and Convolved Sources. 5. ICA Using Overcomplete Representations. 6. First Steps towards Nonlinear ICA. Part II: Independent Component Analysis: Applications. 7. Biomedical Applications of ICA. 8. ICA for Feature Extraction. 9. Unsupervised Classification with ICA Mixture Models. 10. Conclusions and Future Research. Bibliography. About the Author. Index.
772 citations
TL;DR: Five types of beat classes of arrhythmia as recommended by Association for Advancement of Medical Instrumentation (AAMI) were analyzed and dimensionality reduced features were fed to the Support Vector Machine, neural network and probabilistic neural network (PNN) classifiers for automated diagnosis.
586 citations
TL;DR: An improved method is suggested, which utilizes second order Independent Components Analysis (also known as time-structure based Independent components Analysis, or tICA), to construct the state-space of Markov State Models, and the resulting model is an improvement over previously built models using conventional distance metrics.
Abstract: Markov State Models (MSMs) provide an automated framework to investigate the dynamical properties of high-dimensional molecular simulations. These models can provide a human-comprehensible picture of the underlying process and have been successfully used to study protein folding, protein aggregation, protein ligand binding, and other biophysical systems. The MSM requires the construction of a discrete state-space such that two points are in the same state if they can interconvert rapidly. In the following, we suggest an improved method, which utilizes second order Independent Component Analysis (also known as time-structure based Independent Component Analysis, or tICA), to construct the state-space. We apply this method to simulations of NTL9 (provided by Lindorff-Larsen et al. Science2011, 334, 517–520) and show that the MSM is an improvement over previously built models using conventional distance metrics. Additionally, the resulting model provides insight into the role of non-native contacts by reveal...
540 citations
TL;DR: An overview of some recent developments in the theory of independent component analysis is provided, including analysis of causal relations, testing independent components, analysing multiple datasets (three-way data), modelling dependencies between the components and improved methods for estimating the basic model.
Abstract: Independent component analysis is a probabilistic method for learning a linear transform of a random vector. The goal is to find components that are maximally independent and non-Gaussian (non-normal). Its fundamental difference to classical multi-variate statistical methods is in the assumption of non-Gaussianity, which enables the identification of original, underlying components, in contrast to classical methods. The basic theory of independent component analysis was mainly developed in the 1990s and summarized, for example, in our monograph in 2001. Here, we provide an overview of some recent developments in the theory since the year 2000. The main topics are: analysis of causal relations, testing independent components, analysing multiple datasets (three-way data), modelling dependencies between the components and improved methods for estimating the basic model.
311 citations
TL;DR: In order to preserve independence of subject-specific independent components at the subject level and simultaneously establish correspondence of ICs across subjects, a new framework for obtaining subject specific ICs, which is coined group-information guided ICA (GIG-ICA), was proposed in this article.
276 citations
TL;DR: This paper describes a novel artifact removal technique known as ensemble empirical mode decomposition with canonical correlation analysis (EEMD-CCA) which is capable of operating on single-channel measurements and is shown to produce significantly improved results.
Abstract: Biosignal measurement and processing is increasingly being deployed in ambulatory situations particularly in connected health applications. Such an environment dramatically increases the likelihood of artifacts which can occlude features of interest and reduce the quality of information available in the signal. If multichannel recordings are available for a given signal source, then there are currently a considerable range of methods which can suppress or in some cases remove the distorting effect of such artifacts. There are, however, considerably fewer techniques available if only a single-channel measurement is available and yet single-channel measurements are important where minimal instrumentation complexity is required. This paper describes a novel artifact removal technique for use in such a context. The technique known as ensemble empirical mode decomposition with canonical correlation analysis (EEMD-CCA) is capable of operating on single-channel measurements. The EEMD technique is first used to decompose the single-channel signal into a multidimensional signal. The CCA technique is then employed to isolate the artifact components from the underlying signal using second-order statistics. The new technique is tested against the currently available wavelet denoising and EEMD-ICA techniques using both electroencephalography and functional near-infrared spectroscopy data and is shown to produce significantly improved results.
234 citations
TL;DR: In this paper, the Joint Approximate Diagonalization of Eigenmatrices (JADE) algorithm is used to compute ICA parameters, and three examples are used to illustrate its performance, and highlight the differences between ICA results and those of other methods.
Abstract: Independent Components Analysis (ICA) is a relatively recent method, with an increasing number of applications in chemometrics. Of the many algorithms available to compute ICA parameters, the Joint Approximate Diagonalization of Eigenmatrices (JADE) algorithm is presented here in detail. Three examples are used to illustrate its performance, and highlight the differences between ICA results and those of other methods, such as Principal Components Analysis. A comparison with Parallel Factor Analysis (PARAFAC) is also presented in the case of a three-way data set to show that ICA applied on an unfolded high-order array can give results comparable with those of PARAFAC. (c) 2013 Elsevier Ltd. All rights reserved.
176 citations
TL;DR: This study investigates the effect of model order on the distribution of functional properties across networks as a method for identifying the most informative decompositions of BrainMap-based ICA components and suggests dimensionality of 20 for low model order ICA to examine large-scale brain networks.
Abstract: Independent component analysis (ICA) has become a widely used method for extracting functional networks in the brain during rest and task. Historically, preferred ICA dimensionality has widely varied within the neuroimaging community, but typically varies between 20 and 100 components. This can be problematic when comparing results across multiple studies because of the impact ICA dimensionality has on the topology of its resultant components. Recent studies have demonstrated that ICA can be applied to peak activation coordinates archived in a large neuroimaging database (i.e., BrainMap Database) to yield whole-brain task-based co-activation networks. A strength of applying ICA to BrainMap data is that the vast amount of metadata in BrainMap can be used to quantitatively assess tasks and cognitive processes contributing to each component. In this study, we investigated the effect of model order on the distribution of functional properties across networks as a method for identifying the most informative decompositions of BrainMap-based ICA components. Our findings suggest dimensionality of 20 for low model order ICA to examine large-scale brain networks, and dimensionality of 70 to provide insight into how large-scale networks fractionate into sub-networks. We also provide a functional and organizational assessment of visual, motor, emotion, and interoceptive task co-activation networks as they fractionate from low to high model-orders.
168 citations
TL;DR: The results showed the applicability of the ICA-based method to noise-contamination reduction in brain mapping by identifying the original hemodynamic response in the presence of noises.
Abstract: Functional near-infrared spectroscopy (fNIRS) is used to detect concentration changes of oxy-hemoglobin and deoxy-hemoglobin in the human brain. The main difficulty entailed in the analysis of fNIRS signals is the fact that the hemodynamic response to a specific neuronal activation is contaminated by physiological and instrument noises, motion artifacts, and other interferences. This paper proposes independent component analysis (ICA) as a means of identifying the original hemodynamic response in the presence of noises. The original hemodynamic response was reconstructed using the primary independent component (IC) and other, less-weighting-coefficient ICs. In order to generate experimental brain stimuli, arithmetic tasks were administered to eight volunteer subjects. The t-value of the reconstructed hemodynamic response was improved by using the ICs found in the measured data. The best t-value out of 16 low-pass-filtered signals was 37, and that of the reconstructed one was 51. Also, the average t-value of the eight subjects’ reconstructed signals was 40, whereas that of all of their low-pass-filtered signals was only 20. Overall, the results showed the applicability of the ICA-based method to noise-contamination reduction in brain mapping.
148 citations
TL;DR: The approach is able to identify the cluster of generators and buses following a disturbance in the system and is robust in the presence of noise in measured signal, which is an important factor to be considered for assessing the effectiveness of any measurement-based technique.
Abstract: This paper presents a novel approach to the coherency identification technique in interconnected power system using independent component analysis (ICA). The ICA is applied to the generator speed and bus angle data to identify the coherent areas of the system. The results of the application of ICA using simulated data from 16-machine 68-bus system model and on data gathered through U.K. University-based Wide-Area Measurement System are presented. The approach is able to identify the cluster of generators and buses following a disturbance in the system. It is also demonstrated that the approach is robust in the presence of noise in measured signal, which is an important factor to be considered for assessing the effectiveness of any measurement-based technique.
118 citations
TL;DR: It is demonstrated that simple extensions of TRCA can provide most distinctive signals for two tasks and can integrate multiple modalities of information to remove task-unrelated artifacts.
TL;DR: In this paper, the modal identification problem is transformed into a time-frequency framework, and the sparse timefrequency representations of the monotone modal responses are proposed as the targeted independent sources hidden in those of the system responses which have been short-time Fourier-transformed (STFT).
Abstract: Output-only algorithms are needed for modal identification when only structural responses are available. The recent years have witnessed the fast development of blind source separation (BSS) as a promising signal processing technique, pursuing to recover the sources using only the measured mixtures. As the most popular tool solving the BSS problem, independent component analysis (ICA) is able to directly extract the time-domain modal responses, which are viewed as virtual sources, from the observed system responses; however, it has been shown that ICA loses accuracy in the presence of higher-level damping. In this study, the modal identification issue, which is incorporated into the BSS formulation, is transformed into a time-frequency framework. The sparse time-frequency representations of the monotone modal responses are proposed as the targeted independent sources hidden in those of the system responses which have been short-time Fourier-transformed (STFT); they can then be efficiently extracte...
Posted Content•
TL;DR: New and emerging models and approaches for tensor decompositions in applications to group and linked multiway BSS/ICA, feature extraction, classification and Multiway Parti al Least Squares (MPLS) are overviewed.
Abstract: Matrix factorizations and their extensions to tensor factorizations and decompositions have become prominent techniques for linear and multilinear blind source separation (BSS), especially multiway Independent Component Analysis (ICA), NonnegativeMatrix and Tensor Factorization (NMF/NTF), Smooth Component Analysis (SmoCA) and Sparse Component Analysis (SCA). Moreover, tensor decompositions have many other potential applications beyond multilinear BSS, especially feature extraction, classification, dimensionality reduction and multiway clustering. In this paper, we briefly overview new and emerging models and approaches for tensor decompositions in applications to group and linked multiway BSS/ICA, feature extraction, classification andMultiway Partial Least Squares (MPLS) regression problems. Keywords: Multilinear BSS, linked multiway BSS/ICA, tensor factorizations and decompositions, constrained Tucker and CP models, Penalized Tensor Decompositions (PTD), feature extraction, classification, multiway PLS and CCA.
21 Jul 2013
TL;DR: The approach is able to identify the cluster of generators and buses following a disturbance in the system and is robust in the presence of noise in measured signal, an important factor to be considered for assessing the effectiveness of any measurement-based technique.
Abstract: Summary form only given. This paper presents a novel approach to the coherency identification technique in interconnected power system using independent component analysis (ICA). The ICA is applied to the generator speed and bus angle data to identify the coherent areas of the system. The results of the application of ICA using simulated data from 16-machine 68-bus system model and on data gathered through U.K. University-based Wide-Area Measurement System are presented. The approach is able to identify the cluster of generators and buses following a disturbance in the system. It is also demonstrated that the approach is robust in the presence of noise in measured signal, which is an important factor to be considered for assessing the effectiveness of any measurement-based technique.
TL;DR: This work demonstrates the application of MPA to a multi-subject EEG study decomposed using independent component analysis (ICA), compares the results to k-means IC clustering in EEGLAB, and uses surrogate data to test MPA robustness.
TL;DR: This paper proposes a new method for NBI suppression in the data domain based on the independent component analysis (ICA), which copes well with the time-varying NBI with little signal loss.
Abstract: The narrow-band interference (NBI) is a common jamming signal against synthetic aperture radar (SAR), which can degrade the imaging quality severely. This paper proposes a new method for NBI suppression in the data domain based on the independent component analysis (ICA). In this method, echoes contaminated by the NBI are identified in the frequency domain. Next, time filtering and whitening are performed to the identified echoes. Then, the ICA is carried out to decompose the echoes into a series of basis signals, and the jamming components are selected by thresholding. Finally, the NBI is reconstructed and subtracted from the echoes, and the well-focused SAR imagery is obtained by conventional imaging methods. The proposed method copes well with the time-varying NBI with little signal loss. Results of simulated and measured data have proved the validity of the proposed method.
TL;DR: In this paper, the authors used independent component analysis (ICA) to extract the transmission spectrum of the hot-Jupiter HD189733b recorded by the Hubble/NICMOS instrument and showed that spectroscopic errors only about 10%-30% larger than parametric methods can be obtained for 11 spectral bins with bin sizes of ∼0.09 μm.
Abstract: Blind-source separation techniques are used to extract the transmission spectrum of the hot-Jupiter HD189733b recorded by the Hubble/NICMOS instrument. Such a “blind” analysis of the data is based on the concept of independent component analysis. The detrending of Hubble/NICMOS data using the sole assumption that nongaussian systematic noise is statistically independent from the desired light-curve signals is presented. By not assuming any prior or auxiliary information but the data themselves, it is shown that spectroscopic errors only about 10%–30% larger than parametric methods can be obtained for 11 spectral bins with bin sizes of ∼0.09 μm. This represents a reasonable trade-off between a higher degree of objectivity for the non-parametric methods and smaller standard errors for the parametric de-trending. Results are discussed in light of previous analyses published in the literature. The fact that three very different analysis techniques yield comparable spectra is a strong indication of the stability of these results.
TL;DR: A gene expression based approach for the prediction of Parkinson's disease (PD) using 'projection based learning for meta-cognitive radial basis function network (PBL-McRBFN)', inspired by human meta- cognitive learning principles.
Abstract: In this paper, we propose a gene expression based approach for the prediction of Parkinson's disease (PD) using 'projection based learning for meta-cognitive radial basis function network (PBL-McRBFN)'. McRBFN is inspired by human meta-cognitive learning principles. McRBFN has two components, a cognitive component and a meta-cognitive component. The cognitive component is a radial basis function network with evolving architecture. In the cognitive component, the PBL algorithm computes the optimal output weights with least computational effort. The meta-cognitive component controls the learning process in the cognitive component by choosing the best learning strategy for the current sample and adapts the learning strategies by implementing self-regulation. The interaction of cognitive component and meta-cognitive component address the what-to-learn, when-to-learn and how-to-learn of human learning principles efficiently. PBL-McRBFN classifier is used to predict PD using micro-array gene expression data obtained from ParkDB database. The performance of PBL-McRBFN classifier has been evaluated using Independent Component Analysis (ICA) reduced features sets from the complete genes and selected genes with two different significance levels. Further, the performance of PBL-McRBFN classifier is statistically compared with existing classifiers using one-way repeated ANOVA test. Further, it is also used in PD prediction using the standard vocal and gait PD data sets. In all these data sets, the performance of PBL-McRBFN is compared against existing results in the literature. Performance results clearly highlight the superior performance of our proposed approach.
TL;DR: This paper intends to improve the ICA statistical monitoring method by incorporating the ensemble learning approach and the Bayesian inference strategy, and a new performance-driven approach for IC number selection is proposed.
TL;DR: Independent component analysis (ICA), principal component analysis, auto- and cross-correlation are investigated and compared with respect to their effectiveness in extracting the relevant information from video recordings and it is found that ICA produces the most consistent results.
Abstract: Imaging photoplethysmography is an emerging technique for the extraction of biometric information from people using video recordings. The focus is on extracting the cardiac heart rate of the subject by analysing the luminance of the colour video signal and identifying periodic components. Advanced signal processing is needed to recover the information required. In this paper, independent component analysis (ICA), principal component analysis, auto- and cross-correlation are investigated and compared with respect to their effectiveness in extracting the relevant information from video recordings. Results obtained are compared with those recorded by a modern commercial finger pulse oximeter. It is found that ICA produces the most consistent results.
TL;DR: It is shown that these experiments fall short of proving claims that two independent component analysis algorithms, Infomax and FastICA, select for sparsity rather than independence, and that the ICA algorithms are indeed doing what they are designed to do: identify maximally independent sources.
Abstract: A recent paper by Daubechies et al. claims that two independent component analysis (ICA) algorithms, Infomax and FastICA, which are widely used for functional magnetic resonance imaging (fMRI) analysis, select for sparsity rather than independence. The argument was supported by a series of experiments on synthetic data. We show that these experiments fall short of proving this claim and that the ICA algorithms are indeed doing what they are designed to do: identify maximally independent sources.
Book•
07 Aug 2013TL;DR: The Independent Component Analysis (ICA) as mentioned in this paper is a recently developed technique for component extraction, which requires the statistical independence of the extracted components, a stronger constraint that uses higher-order statistics, instead of the classical decorrelation, a weaker constraint that using only secondorder statistics.
Abstract: The Independent Component Analysis is a recently developed technique for component extraction. This new method requires the statistical independence of the extracted components, a stronger constraint that uses higher-order statistics, instead of the classical decorrelation, a weaker constraint that uses only second-order statistics. This technique has been used recently for the analysis of geophysical time series with the goal of investigating the causes of variability in observed data (i.e. exploratory approach). We demonstrate with a data simulation experiment that, if initialized with a Principal Component Analysis, the Independent Component Analysis performs a rotation of the classical PCA (or EOF) solution. This rotation uses no localization criterion like other Rotation Techniques (RT), only the global generalization of decorrelation by statistical independence is used. This rotation of the PCA solution seems to be able to solve the tendency of PCA to mix several physical phenomena, even when the signal is just their linear sum.
TL;DR: In this article, it is shown that the ICA algorithm with diagonalizing the 4th order cumulant tensor, through the rotation of experimental orthogonal functions, will indeed perfectly separate an unknown mixture of trend and sinusoidal signals in the data, provided that the length of the data set is infinite.
Abstract: Independent Component Analysis (ICA) represents a higher-order statistical technique that is often used to separate mixtures of stochastic random signals into statistically independent sources. Its benefit is that it only relies on the information contained in the observations, i.e. no parametric a-priori models are prescribed to extract the source signals. The mathematical foundation of ICA, however, is rooted in the theory of random signals. This has led to questions whether the application of ICA to deterministic signals can be justified at all? In this context, the possibility of using ICA to separate deterministic signals such as complex sinusoidal cycles has been subjected to previous studies. In many geophysical and geodetic applications, however, understanding long-term trend in the presence of periodical components of an observed phenomenon is desirable. In this study, therefore, we extend the previous studies with mathematically proving that the ICA algorithm with diagonalizing the 4th order cumulant tensor, through the rotation of experimental orthogonal functions, will indeed perfectly separate an unknown mixture of trend and sinusoidal signals in the data, provided that the length of the data set is infinite. In other words, we justify the application of ICA to those deterministic signals that are most relevant in geodetic and geophysical applications.
TL;DR: This work shows that the use of variance information in source space projected Hilbert envelope time series yields important spatial information, and is of significant functional relevance, and shows that employing this information in functional connectivity analyses improves the spatial delineation of network nodes using both seed based and ICA approaches.
TL;DR: This paper utilizes a hierarchical method including a blind source separation (BSS) step, in particular independent component analysis, to eliminate the effect of the baseline wander in electrocardiogram signals.
Abstract: Noise can compromise the extraction of some fundamental and important features from biomedical signals and hence prohibit accurate analysis of these signals. Baseline wander in electrocardiogram (ECG) signals is one such example, which can be caused by factors such as respiration, variations in electrode impedance, and excessive body movements. Unless baseline wander is effectively removed, the accuracy of any feature extracted from the ECG, such as timing and duration of the ST-segment, is compromised. This paper approaches this filtering task from a novel standpoint by assuming that the ECG baseline wander comes from an independent and unknown source. The technique utilizes a hierarchical method including a blind source separation (BSS) step, in particular independent component analysis, to eliminate the effect of the baseline wander. We examine the specifics of the components causing the baseline wander and the factors that affect the separation process. Experimental results reveal the superiority of the proposed algorithm in removing the baseline wander.
TL;DR: Two main geometric objects related to the separator output samples, namely Principal Hyper-Ellipsoid and Bounding Hyper-Rectangle, are introduced as relevant optimization problems for Bounded Component Analysis algorithms.
Abstract: Bounded Component Analysis (BCA) is a recent approach which enables the separation of both dependent and independent signals from their mixtures. In this approach, under the practical source boundedness assumption, the widely used statistical independence assumption is replaced by a more generic domain separability assumption. This article introduces a geometric framework for the development of Bounded Component Analysis algorithms. Two main geometric objects related to the separator output samples, namely Principal Hyper-Ellipsoid and Bounding Hyper-Rectangle, are introduced. The maximization of the volume ratio of these objects, and its extensions, are introduced as relevant optimization problems for Bounded Component Analysis. The article also provides corresponding iterative algorithms for both real and complex sources. The numerical examples illustrate the potential advantage of the proposed BCA framework in terms of correlated source separation capability as well as performance improvement for short data records.
Book•
15 Jul 2013
TL;DR: The book presents research work on face recognition using edge information as features for face recognition with ICA algorithms and provides insights for advance research work in the area of image processing and biometrics.
Abstract: The book presents research work on face recognition using edge information as features for face recognition with ICA algorithms. The independent components are extracted from edge information. These independent components are used with classifiers to match the facial images for recognition purpose. In their study, authors have explored Canny and LOG edge detectors as standard edge detection methods. Oriented Laplacian of Gaussian (OLOG) method is explored to extract the edge information with different orientations of Laplacian pyramid. Multiscale wavelet model for edge detection is also proposed to extract edge information. The book provides insights for advance research work in the area of image processing and biometrics.
TL;DR: This paper presents a novel single-mixture blind source separation method based on edge effect elimination of EEMD, principal component analysis (PCA) and independent components analysis (ICA), which outperforms the two latter algorithms with higher correlation coefficient and lower relative root mean square error (RRMSE).
Abstract: Blind source separation (BSS) of single-channel mixed recording is a challenging task that has applications in the fields of speech, audio and bio-signal processing. Ensemble empirical mode decomposition (EEMD)-based methods are commonly used for blind separation of single input multiple outputs. However, all of these EEMD-based methods appear in the edge effect problem when cubic spline interpolation is used to fit the upper and lower envelopes of the given signals. It is therefore imperative to have good methods to explore a more suitable design choice, which can avoid the problems mentioned above as much as possible. In this paper we present a novel single-mixture blind source separation method based on edge effect elimination of EEMD, principal component analysis (PCA) and independent component analysis (ICA). EEMD represents any time-domain signal as the sum of a finite set of oscillatory components called intrinsic mode functions (IMFs). In extreme point symmetry extension (EPSE), optimum values of endpoints are obtained by minimizing the deviation evaluation function of signal and signal envelope. Edge effect is turned away from signal by abandoning both ends’ extension parts of IMFs. PCA is applied to reduce dimensions of IMFs. ICA finds the independent components by maximizing the statistical independence of the dimensionality reduction of IMFs. The separated performance of edge EPSE-EEMD-PCA-ICA algorithm is compared with EEMD-ICA and EEMD-PCA-ICA algorithms through simulations, and experimental results show that the former algorithm outperforms the two latter algorithms with higher correlation coefficient and lower relative root mean square error (RRMSE).
01 Nov 2013
TL;DR: This paper will try to summarize those approaches taken previously to solve blind source separation and an experiment of source separation which will mix using Independent Component Analysis and then de-mix those source signals using ICA as the basic/prime approach.
Abstract: Blind Source Separation (BSS) refers to a problem where both the sources and the mixing methodology are unknown, only mixture signals are available for further separation process. In several situations it is desirable to recover all individual sources from the mixed signal, or at least to segregate a particular source. In laboratory condition, most of the algorithms works very fine where input signals, no. of source present in the mixture, mixing methodology etc are well known to the separation process. But in real-life scenario the problem is much more complicated and it begins with the input signal, a mixture where most of the parameters are unknown. This paper will try to summarize those approaches taken previously to solve this problem and an experiment of source separation which will mix using Independent Component Analysis (ICA) and then de-mix those source signals using ICA as the basic/prime approach.
TL;DR: This paper presents an online learning algorithm for the Gaussian process (GP) and establishes a separation procedure in the presence of nonstationary and temporally correlated mixing coefficients and source signals and captures the evolved statistics from sequential signals according to online Bayesian learning.
Abstract: Independent component analysis (ICA) is a popular approach for blind source separation where the mixing process is assumed to be unchanged with a fixed set of stationary source signals. However, the mixing system and source signals are nonstationary in real-world applications, e.g., the source signals may abruptly appear or disappear, the sources may be replaced by new ones or even moving by time. This paper presents an online learning algorithm for the Gaussian process (GP) and establishes a separation procedure in the presence of nonstationary and temporally correlated mixing coefficients and source signals. In this procedure, we capture the evolved statistics from sequential signals according to online Bayesian learning. The activity of nonstationary sources is reflected by an automatic relevance determination, which is incrementally estimated at each frame and continuously propagated to the next frame. We employ the GP to characterize the temporal structures of time-varying mixing coefficients and source signals. A variational Bayesian inference is developed to approximate the true posterior for estimating the nonstationary ICA parameters and for characterizing the activity of latent sources. The differences between this ICA method and the sequential Monte Carlo ICA are illustrated. In the experiments, the proposed algorithm outperforms the other ICA methods for the separation of audio signals in the presence of different nonstationary scenarios.