Search

Linear Discriminant Analysis (LDA) is a very powerful method in pattern recognition. But it is difficult to realize online processing for data stream. In this paper, a new adaptive LDA method is proposed. We decompose the online LDA problem into two adaptive PCA problems and develop a fixed point adaptive PCA to implement adaptive LDA. Online updating of in-class scatter matrix Sw(t) and covariance matrix Cx(t) are derived in this paper. Simulation results show that the proposed method has no learning rate, fast convergence and less time-consuming. [ABSTRACT FROM AUTHOR] more...

Locally linear embedding (LLE) is one of the methods intended for dimensionality reduction, which relates to the number K of nearest-neighbors points to be initially chosen. So, in this paper, we want that the parameter K has little influence on the dimension reduction, that is to say, the parameter K can be widely chosen while not influence the effect of dimension reduction. Therefore, we propose a method of improved LLE, which uses new distance computing for weight of K nearest-neighbors points in LLE. Thus, even when the number K is little, the improved LLE can get good results of dimension reduction, while the traditional LLE needs a larger number of K to get the same results. When the number K of the nearest neighbors gets larger, test in this paper has proved that the improved LLE can still get correct results. [ABSTRACT FROM AUTHOR] more...

For a well-trained radial basis function neural network, this paper proposes a novel input stochastic sensitivity definition and gives its computational formula assuming the inputs are modelled by normal distribution random variables. Based on this formula, one can calculate the magnitude of sensitivity for each input (i.e. feature), which indicates the degree of importance of input to the output of neural network. When there are redundant inputs in the training set, one always wants to remove those redundant features to avoid a large network. This paper shows that removing redundant features or selecting significant features can be completed by choosing features with sensitivity over a predefined threshold. Numerical experiment shows that the new approach to feature selection performs well. [ABSTRACT FROM AUTHOR] more...

Blind identification of mixing matrix approach and the corresponding algorithm are proposed in this paper. Usually, many conventional Blind Source Separation (BSS) methods separate the source signals by estimating separated matrix. Different from this way, we present a new BSS approach in this paper, which achieves BSS by directly identifying the mixing matrix, especially for underdetermined case. Some experiments are conducted to check the validity of the theory and availability of the algorithm in this paper. [ABSTRACT FROM AUTHOR] more...

Comparative study of extreme learning machine (ELM) and support vector machine (SVM) is investigated in this paper. A cross validation method for determining the appropriate number of neurons in the hidden layer is also proposed in this paper. ELM proposed by Huang, et al [3] is a novel machine-learning algorithm for single hidden-layer feedforward neural network (SLFN), which randomly chooses the input weights and hidden-layer bias, and analytically determines the output weights optimally instead of tuning them. This algorithm tends to produce good generalization ability and obtain least experience risk simultaneously with solid foundations. Benchmark tests of a real Tennessee Eastman Process (TEP) are carried out to validate its superiority. Compared with SVM, this proposed algorithm is much faster and has better generalization performance than SVM in the case studied in this paper. [ABSTRACT FROM AUTHOR] more...

This paper presents an SVM classification algorithm with predesigned error correction ability by incorporating the error control coding schemes used in digital communications into the classification algorithm. The algorithm is applied to face recognition problems in the paper. Simulation experiments are conducted for different SVM-based classification algorithms using both PCA and Fisherface features as input vectors respectively to represent the images with dimensional reduction, and performance analysis is made among different approaches. Experiment results show that the error correction SVM classifier of the paper outperforms other commonly used SVM-based classifiers both in recognition rate and error tolerance. [ABSTRACT FROM AUTHOR] more...

In this paper, a clustering algorithm based on Gradient Based Fuzzy C-Means with a Mercer Kernel, called GBFCM (MK), is proposed. The kernel method adopted in this paper implicitly performs nonlinear mapping of the input data into a high-dimensional feature space. The proposed GBFCM(MK) algorithm is capable of dealing with nonlinear separation boundaries among clusters. Experiments on a synthetic data set and several real MPEG data sets show that the proposed algorithm gives better classification accuracies than both the conventional k-means algorithm and the GBFCM. [ABSTRACT FROM AUTHOR] more...

In view of the character of saturation linearity of output functions of neurons of the cellular neural networks, the method decomposing the state space to sub-regions is adopted to study almost sure exponential stability on delayed cellular neural networks which are in the noised environment. When perturbed terms in the model of the neural network satisfy Lipschitz condition, some algebraic criteria are obtained. The results obtained in this paper show that if an equilibrium of the neural network is the interior point of a sub-region, and an appropriate matrix related to this equilibrium has some stable degree to stabilize the perturbation, then the equilibrium of the delayed cellular neural network can still remain the property of exponential stability. All results in the paper is only to compute eigenvalues of matrices. [ABSTRACT FROM AUTHOR] more...

In this paper, the global exponential stability of a class of recurrent cellular neural networks with reaction-diffusion and variable time delays was studied. When neural networks contain unbounded activation functions, it may happen that equilibrium point does not exist at all. In this paper, without assuming the boundedness, monotonicity and differentiability of the active functions, the algebraic criteria ensuring existence, uniqueness and global exponential stability of the equilibrium point of neural networks are obtained. [ABSTRACT FROM AUTHOR] more...

In this paper, we discuss delayed Cohen-Grossberg neural networks with time-varying and distributed delays and investigate their global asymptotical stability of the equilibrium point. The model proposed in this paper is universal. A set of sufficient conditions ensuring global convergence and globally exponential convergence for the Cohen-Grossberg neural networks with time-varying and distributed delays are given. Most of the existing models and global stability results for Cohen-Grossberg neural networks, Hopfield neural networks and cellular neural networks can be obtained from the theorems given in this paper. [ABSTRACT FROM AUTHOR] more...

In this paper, the global uniform asymptotical stability is studied for delayed neutral-type neural networks by constructing appropriate Lyapunov functional and using the linear matrix inequality (LMI) approach. The main condition given in this paper is dependent on the size of the measure of the space, which is usually less conservative than space-independent ones. Finally, a numerical example is provided to demonstrate the effectiveness and applicability of the proposed criteria. [ABSTRACT FROM AUTHOR] more...

In this paper we studied a new dynamic evolution model on phase encoding in population of neuronal oscillators under condition of different phase, and investigated neural information processing in cerebral cortex and dynamic evolution under action of different stimulation signal. It is obtained that evolution of the averaging number density along with time in space of three dimensions is described in different cluster of neuronal oscillators firing action potential at different phase space by means of method of numerical analysis. The results of numerical analysis show that the dynamic model proposed in this paper can be used to describe mechanism of neurodynamics on attention and memory. [ABSTRACT FROM AUTHOR] more...

In this paper we propose a new principal component extracting algorithm based on the PAST. A prewhitening procedure is introduced, which makes it numerically robust. The estimation capability of the proposed algorithm is demonstrated by computer simulations of DOA (Degree of Arrival) estimation. The estimation results of the proposed PAST outperform those of the ordinary PAST. [ABSTRACT FROM AUTHOR] more...

Dimensionality reduction including both feature selection and feature extraction techniques are useful for improving the performance of neural networks. In this paper, particle swarm optimization (PSO) algorithm was proposed for simultaneous feature extraction and feature selection. First PSO was used to simultaneous feature extraction and selection in conjunction with knearest- neighbor (k-NN) for individual fitness evaluation. With the derived feature set, PSO was then used to evolve RBF networks dynamically. Experimental results on four datasets show that RBF networks evolved with the derived feature set by our proposed algorithm have more simple architecture and stronger generalization ability with the similar classification performance when compared with the networks evolved with the full feature set. [ABSTRACT FROM AUTHOR] more...

In this paper, a novel supervised information feature compression algorithm based on divergence is set up. Firstly, according to the information theory, the concept and its properties of the divergence, i.e. average separability information (ASI) is studied, and a concept of symmetry average separability information (SASI) is proposed, and proved that the SASI here is a kind of distance measure, i.e. the SASI satisfies three requests of distance axiomatization, which can be used to measure the difference degree of a two-class problem. Secondly, based on the SASI, a compression theorem is given, and can be used to design information feature compression algorithm. Based on these discussions, we design a novel supervised information feature compression algorithm based on the SASI. At last, the experimental results demonstrate that the algorithm here is valid and reliable. [ABSTRACT FROM AUTHOR] more...

This paper presents a new target recognition scheme via the neural network based on Hidden Markov Model (HMM), which processes the multiaspect features. The target features are extracted by the adaptive gaussian representation (AGR) from the view of physics. Discrimination results are presented for ISAR radar return signal. [ABSTRACT FROM AUTHOR] more...

Non-negative matrix factorization (NMF) is a recently proposed parts-based representation method, and because of its non-negativity constraints, it is mostly used to learn parts of faces and semantic features of text. In this paper, non-negative matrix factorization is first applied to extract features of underground nuclear explosion signals and natural earthquake signals, then a novel kernel-based non-negative matrix factorization (KNMF) method is proposed and also applied to extract features. To compare practical classification ability of these features extracted by NMF and KNMF, linear support vector machine (LSVM) is applied to distinguish nuclear explosions from natural earthquakes. Theoretical analysis and practical experimental results indicate that kernel-based non-negative matrix factorization is more appropriate for the feature extraction of underground nuclear explosions and natural earthquakes. [ABSTRACT FROM AUTHOR] more...

Many feature extraction algorithms have been proposed for time series classification. However, most of the proposed algorithms in time series data mining community belong to the unsupervised approach, without considering the class separability capability of features that is important for classification. In this paper we propose a supervised feature extraction approach by selecting discriminating wavelet coefficients to improve the time series classification accuracy. After wavelet transformation, few wavelet coefficients with higher class separability capability are selected as features. We apply three feature evaluation criteria, i.e., Fisher's discriminant ratio, divergence, and Bhattacharyya distance. Experiments performed on several benchmark time series datasets demonstrate the effectiveness of the proposed approach. [ABSTRACT FROM AUTHOR] more...

Most existing feature extraction algorithms aim at best preserving information in the original data or at improving the separability of data, but fail to consider the possibility of further reducing the number of used features. In this paper, we propose a parsimonious feature extraction algorithm. Its motivation is using as few features as possible to achieve the same or even better classification performance. It searches for the optimal features using a genetic algorithm and evaluates the features referring to Support Vector Machines. We tested the proposed algorithm by face recognition on the Yale and FERET databases. The experimental results proved its effectiveness and demonstrated that parsimoniousness should be a significant factor in developing efficient feature extraction algorithms. [ABSTRACT FROM AUTHOR] more...

Text classification is a problem of assigning a document into one or more predefined classes. One of the most interesting issues in text categorization is feature selection. This paper proposes a novel approach in feature selection based on support vector machine(SVM) and latent semantic indexing(LSI), which can identify LSI-subspace that is suited for classification. Experimental results show that the proposed method can achieve higher classification accuracies and is of less training and prediction time. [ABSTRACT FROM AUTHOR] more...

As a feature selection method, support vector machines-recursive feature elimination (SVM-RFE) can remove irrelevance features but don't take redundant features into consideration. In this paper, it is shown why this method can't remove redundant features and an improved technique is presented. Correlation coefficient is introduced to measure the redundancy in the selected subset with SVM-RFE. The features which have a great correlation coefficient with some important feature are removed. Experimental results show that there actually are several strongly redundant features in the selected subsets by SVM-RFE. The coefficients are high to 0.99. The proposed method can not only reduce the number of features, but also keep the classification accuracy. [ABSTRACT FROM AUTHOR] more...

This paper proposes a new method for fractal intrinsic dimension(ID) estimation. Local fractal features are extracted and combined to obtain the estimation. Compared with the contemporary methods for fractal ID estimation, the proposed method requires lower computation, can reach accurate results and can be flexibly extended. Both the theoretical analysis and experimental results show its validity. [ABSTRACT FROM AUTHOR] more...

The neural networks are successfully applied to many applications in different domains. However, due to the results made by the neural networks are difficult to explain the decision process of neural networks is supposed as a black box. The explanation of reasoning is important to some applications such like credit approval application and medical diagnosing software. Therefore, the rule extraction algorithm is becoming more and more important in explaining the extracted rules from the neural networks. In this paper, a decompositional algorithm is analyzed and designed to extract rules from neural networks. The algorithm is simple but efficient; can reduce the extracted rules but improve the efficiency of the algorithm at the same time. Moreover, the algorithm is compared to the other two algorithms, M-of-N and Garcez, by solving the MONK's problem. [ABSTRACT FROM AUTHOR] more...

Based on Jordan Curve Theorem, a universal classification method called HyperSurface Classifier (HSC) has recently been proposed. Experimental results show that in three-dimensional space, this method works fairly well in both accuracy and efficiency even for large size data up to 107. However, what we really need is an algorithm that can deal with data not only of massive size but also of high dimensionality. In this paper, an approach based on the idea of classifiers ensemble by dimension dividing without dimension reduction for high dimensional data is proposed. The most important difference between HSC ensemble and the traditional ensemble is that the sub-datasets are obtained by dividing the features rather than by dividing the sample set. Experimental results show that this method has a preferable performance on high dimensional datasets. [ABSTRACT FROM AUTHOR] more...

The method of discounting coefficient is an efficient way to solve the problem of evidence conflicts. In this paper a new method to calculate the discounting coefficient of evidence based on evidence clustering by the way of fuzzy ART neural network is proposed. The discounted evidence is taken into account in belief function combination. A numerical example is shown to illustrate the use of the proposed method to handle conflicting evidence. [ABSTRACT FROM AUTHOR] more...

A new off-line learning method of single-hidden layer feed-forward neural networks (SLFN) called Extreme Learning Machine (ELM) was introduced by Huang et al. [1, 2, 3, 4] . ELM is not the same as traditional BP methods as it can achieve good generalization performance at extremely fast learning speed. In ELM, the hidden neuron parameters (the input weights and hidden biases or the RBF centers and impact factors) were pre-assigned randomly so there may be a set of non-optimized parameters that avoid ELM achieving global minimum in some applications. Adopting the ideas in [5] that a single layer feed-forward neural network can be trained using a hybrid approach which takes advantages of both ELM and the evolutionary algorithm, this paper introduces a new kind of evolutionary algorithm called particle swarm optimization (PSO) which can train the network more suitable for some prediction problems using the ideas of ELM. [ABSTRACT FROM AUTHOR] more...

In this paper, a new competitive learning algorithm called Dynamic Competitive Learning (DCL) is presented. DCL is a supervised learning method that dynamically generates output neurons and initializes automatically the weight vectors from training patterns. It introduces a new parameter called LOG (Limit of Grade) to decide whether an output neuron is created or not. If the class of at least one among LOG number of nearest output neurons is the same as the class of the present training pattern, then DCL adjusts the weight vector associated with the output neuron to learn the pattern. If the classes of all the nearest output neurons are different from the class of the training pattern, a new output neuron is created and the given training pattern is used to initialize the weight vector of the created neuron. The proposed method is significantly different from the previous competitive learning algorithms in the point that the selected neuron for learning is not limited only to the winner and the output neurons are dynamically generated during the learning process. In addition, the proposed algorithm has a small number of parameters, which are easy to be determined and applied to real-world problems. Experimental results demon- strate the superiority of DCL in comparison to the conventional competitive learning methods. [ABSTRACT FROM AUTHOR] more...

Training strategy for negative sample collection and robust learning algorithm for large-scale samples set are critical issues for visual information retrieval problem. In this paper, an improved one class support vector classifier (SVC) and its boosting chain learning algorithm is proposed. Different from the one class SVC, this algorithm considers negative samples information, and integrates the bootstrap training and boosting algorithm into its learning procedure. The performances of the SVC can be successively boosted by repeat important sampling large negative set. Compared with traditional methods, it has the merits of higher detection rate and lower false positive rate, and is suitable for object detection and information retrieval. Experimental results show that the proposed boosting SVM chain learning method is efficient and effective. [ABSTRACT FROM AUTHOR] more...

Support Vector Data Description (SVDD) concerns the characterization of a data set. A good description covers all target data but includes no superfluous space. The boundary of a data set can be used to detect outliers. SVDD is affected by noises during being trained. In this paper, Grid-based Fuzzy Support Vector Data Description (G-FSVDD) is presented to deal with the problem. G-FSVDD reduces the effects of noises by a new fuzzy membership model, which is based on grids. Each grid is a hypercube in data set. After obtaining enough grids, Apriori algorithm is used to find grids with high density. In G-FSVDD, different training data make different contributions to the domain description according to their density. The advantage of G-FSVDD is shown in the experiment. [ABSTRACT FROM AUTHOR] more...

Keyphrase extraction is a task with many applications in information retrieval, text mining, and natural language processing. In this paper, a keyphrase extraction approach based on neural network ensemble is proposed. To determine whether a phrase is a keyphrase, the following features of a phrase in a given document are adopted: its term frequency, whether to appear in the title, abstract or headings (subheadings), and its frequency appearing in the paragraphs of the given document. The approach is evaluated by the standard information retrieval metrics of precision and recall. Experiment results show that the ensemble learning can significantly increase the precision and recall. [ABSTRACT FROM AUTHOR] more...

Text features are usually expressed as a huge dimensional vector in text mining, LSA can reduce dimensionality of text features effectively, and emerges the semantic relations between texts and terms. This paper presents a Conscientious Rival Penalized Competitive Learning (CRPCL) text clustering algorithm, which uses LSA to reduce the dimensionality and improves RPCL to set a conscientious threshold to restrict a winner that won too many times and to make every neural unit win the competition at near ideal probability. The experiments demonstrate good performance of this method. [ABSTRACT FROM AUTHOR] more...

In this paper hierarchical clustering and self-organizing maps (SOM) clustering are compared by using molecular data of large size sets. The hierarchical clustering can represent a multi-level hierarchy which show the tree relation of cluster distance. SOM can adapt the winner node and its neighborhood nodes, it can learn topology and represent roughly equal distributive regions of the input space, and similar inputs are mapped to neighboring neurons. By calculating distances between neighboring units and Davies-Boulding clustering index, the cluster boundaries of SOM are decided by the best Davies-Boulding clustering index. The experimental results show the effectiveness of clustering for molecular data, between-cluster distance of low energy samples from transition networks is far bigger than that of "local sampling" samples, the former has a better cluster result, "local sampling" data nevertheless exhibit some clusters. [ABSTRACT FROM AUTHOR] more...

In this paper competitive learning cluster are used for molecular data of large size sets. The competitive learning network can cluster the input data, it only adapts to the node of winner, the winning node is more likely to win the competition again when a similar input is presented, thus similar inputs are clustered into same a class and dissimilar inputs are clustered into different classes. The experimental results show that the competitive learning network has a good clustering reproducible, indicates the effectiveness of clusters for molecular data, the conscience learning algorithm can effectively cancel the dead nodes when the output nodes increasing, the kinds of network indicates the effectiveness of clusters for molecular data of large size sets. [ABSTRACT FROM AUTHOR] more...

In this paper, we focus on developing a new clustering method, robust kernel-based deterministic annealing (RKDA) algorithm, for data clustering in mercer kernel-induced feature space. A nonlinear version of the standard deterministic annealing (DA) algorithm is first constructed by means of a Gaussian kernel, which can reveal the structure in the data that may go unnoticed if DA is performed in the original input space. After that, a robust pruning method, the maximization of the mutual information against the constrained input data points, is performed to phase out noise and outliers. The good aspects of the proposed method for data clustering are supported by experimental results. [ABSTRACT FROM AUTHOR] more...

A novel self Adaptive Support Vector Clustering algorithm (ASVC) is proposed in this paper to cluster dataset with diverse dispersions. And a Kernel function is defined to measure affinity between multi-relational data. Task of clustering multi-relational data is addressed by integrating the designed Kernel into ASVC. Experimental results indicate that the designed Kernel can capture structured features well and ASVC is of fine performance. [ABSTRACT FROM AUTHOR] more...

How to determine the number of clusters and the dimensions of local principal subspaces is an important and challenging problem in various applications. Based on a probabilistic model of local PCA, this problem can be solved by one of existing statistical model selection criteria in a two-phase procedure. However, such a two-phase procedure is too time-consuming especially when there is no prior knowledge. The BYY harmony learning has provided a promising mechanism to make automatic model selection in parallel with parameter learning. This paper investigates the BYY harmony learning with automatic model selection on a mixture PCA model in comparison with three typical model selection criteria: AIC, CAIC, and MDL. This comparative study is made by experiments for different model selection tasks on simulated data sets under different conditions. Experiments have shown that automatic model selection by the BYY harmony learning are not only as good as or even better than conventional methods in terms of performances, but also considerably supervisory in terms of much less computational cost. [ABSTRACT FROM AUTHOR] more...

In this paper, we integrate orthogonal frequency-division multiplexing (OFDM) technique with vertical Bell Labs layered space-time (V-BLAST) architecture as a promising solution for enhancing the data rates of wireless communication systems, and propose a new blind deconvolution method. A two-stage algorithm is developed to estimate the channel parameters. At first stage, we propose an algorithm based on the second order statistics to decorrelate the sensor signals. After decorrelation, we apply instantaneous demixing algorithm to separate the signals at the second stage. Simulation results demonstrate the validity and the performance of the proposed algorithms. [ABSTRACT FROM AUTHOR] more...

In our previous work [11], we introduced a filter decomposition method for blind deconvolution in non-minimum phase system. To simplify the deconvolution procedure, we further study the demixing filter and modify the cascade structure of demixing filter. In this paper, we introduce a novel two-stage algorithm for blind deconvolution. In first stage, we present a permutable cascade structure which constructed by a causal filter and an anti-causal scalar filter. Then, we develop SOS-based algorithm for causal filter and derive a natural gradient algorithm for anti-causal scalar filter. At second stage, we apply an instantaneous ICA algorithm to eliminate the residual instantaneous mixtures. Computer simulations show the validity and effectiveness of this approach. [ABSTRACT FROM AUTHOR] more...

In this paper, a convolutive blind source separation (BSS) algorithm based on a double-iteration method is proposed to process the convolutive mixed non-white broadband signals. By sliding Fourier transform (SFT), the convolutive mixture problem is changed into instantaneous case in time-frequent domain, which can be solved by applying an instantaneous separation method for every frequent bin. A novel cost function for each frequent bin based on joint diagonalization of a set of correlation matrices with multiple time-lags is constructed. Through combination of the proposed double-iteration method with a restriction on the length of inverse filter in time domain, the inverse of transfer channel or separation matrix, which has consistent permutations for all frequencies, can be estimated. Then it is easy to calculate the recovered source signals. The results of simulations also illustrate the algorithm has not only fast convergence performance, but also higher recovered accuracy and output SER (Signal to Error of reconstruction Ratio). [ABSTRACT FROM AUTHOR] more...

Passive acoustic source location and tracking can be implemented by microphone array, which makes it have a broad application on target detection and fault diagnosis etc. Five-element planar cross array is an effective structure for acoustic source location, but the bad location results often obtained by conventional relative acoustic source location methods in complicated environments. This paper proposed a novel passive acoustic source location method based on blind source separation and thoroughly analyzed this new location algorithm. Experiments show that a more stable and accurate location results may be obtained by this new location method. The implementation of this location method is relatively simple and effective, which can satisfy the demand of engineering application. [ABSTRACT FROM AUTHOR] more...

Underdetermined blind delayed source problem is studied in this paper. Firstly, on the basis of the searching-and-averaging-based method in frequency domain, the algorithm was extended to blind delay source model. Secondly, a new cost function for estimating the delay of observed signal was present; the delay was inferred in the single-signal intervals. Finally, the delayed sound experiments demonstrate its performance. [ABSTRACT FROM AUTHOR] more...

In this paper, a new method for identifying the independent components of an alpha-stable random vector for under-determined mixtures is proposed. The method is based on an estimate of the discrete Borel measure for the characteristic function of an alpha-stable random vector. Simulations demonstrate that the proposed method can identify independent components and the basis vectors of mixing matrix in the so-called under-determined case of more sources than mixtures. [ABSTRACT FROM AUTHOR] more...

This paper proposes a novel algorithm based on minimizing mutual information for a special case of nonlinear blind source separation: post-nonlinear blind source separation. A network composed of a set of radial basis function (RBF) networks, a set of multilayer perceptron and a linear network is used as a demixing system to separate sources in post-nonlinear mixtures. The experimental results show that our proposed method is effective, and they also show that the local character of the RBF network's units allows a significant speedup in the training of the system. [ABSTRACT FROM AUTHOR] more...

Independent component analysis (ICA) is a widely applicable and effective approach in blind source separation (BSS) for basic ICA model, but with limitations that sources should be statistically independent, while more common situation is BSS for non-negative linear (NNL) model where observations are linear combinations of non-negative sources with non-negative coefficients and sources may be statistically dependent. By recognizing the fact that BSS for basic ICA model corresponds to matrix factorization problem, in this paper, a novel idea of BSS for NNL model is proposed that the BSS for NNL corresponds to a non-negative matrix factorization problem and the non-negative matrix factorization (NMF) technique is utilized. For better expression of the patterns of the sources, the NMF is further extended to pattern expression NMF (PE-NMF) and its algorithm is presented. Finally, the experimental results are presented which show the effectiveness and efficiency of the PE-NMF to BSS for a variety of applications which follow NNL model. [ABSTRACT FROM AUTHOR] more...

In this paper, a novel blind source separation (BSS) algorithm based on generalized variance is proposed according to the property of multivariable statistical analysis. This separation contrast function of this algorithm is based on second order moments. It can complete the blind separation of supergaussian and subgaussian signals at the same time without adjusting the learning function The restriction of this algorithm is not too much and the computation burden is light. Simulation results confirm that the algorithm is statistically efficient for all practical purpose and the separation effect is very feasible. [ABSTRACT FROM AUTHOR] more...

The natural gradient blind source separation (BSS) algorithm with unknown source number proposed by Cichocki in 1999 is justified in this paper. An new method to detect the redundant separated signals based on structure of separating matrix is proposed, by embedding it into the natural gradient algorithm, an novel BSS algorithm with an unknown source number is developed. The novel algorithm can successfully separate source signals and converge stably, while the Cichocki's algorithm would diverge inevitably. The new method embedded in novel algorithm can detect and cancel the redundant separated signals within 320 iteration, which is far quicker than the method based on the decorrelation, if some parameters are chosen properly. [ABSTRACT FROM AUTHOR] more...

The purpose of this paper is to develop two novel unified parametric and non-parametric Independent Component Analysis (ICA) algorithms, which enable to separate arbitrary sources including symmetric and asymmetric distributions with self-adaptive score functions. They are derived from the parameterized asymmetric generalized Gaussian density (AGGD) model and GGD kernel based k-nearest neighbor (KNN) non-parametric estimation. The parameters of the score function in the algorithms are been chosen adaptively by estimating the high order statistics of the observed signals and GGD kernel estimation based non-parametric method. Compared with conventional ICA algorithms, the two given methods can separate a wide range of source signals using only one unified density model. Simulations confirm the effectiveness and performance of the proposed algorithm. [ABSTRACT FROM AUTHOR] more...

A novel algorithm is proposed for the nonnegative independent component analysis. In the algorithm, we employ the gradient algorithm with some modifications to separate nonnegative independent sources from mixtures. Since the local convergence of the gradient algorithm is already proved, the result in this paper will be considered one of the convergent nonnegative ICA algorithms. Simulation shows the proposed algorithm can separate the mixtures of nonnegative signals very successfully. [ABSTRACT FROM AUTHOR] more...

Hyävrinen and Oja have proposed an offline Fast-ICA algorithm. But it converge slowly in online form. By using the online whitening algorithm, and applying nature Riemannian gradient in Stiefel manifold, we present in this paper an extended online Fast-ICA algorithm, which can perform online blind source separation (BSS) directly using unwhitened observations. Computer simulation resluts are given to demonstrate the effectiveness and validity of our algorithm. [ABSTRACT FROM AUTHOR] more...

This paper presents a novel independent component analysis algorithm that separates mixtures using serially updating geodesic method. The geodesic method is derived from the Stiefel manifold, and an on-line version of this method that can directly treat with the unwhitened observations is obtained. Simulation of artificial data as well as real biological data reveals that our proposed method has fast convergence. [ABSTRACT FROM AUTHOR] more...