Novel on-line learning algorithms with self adaptive rates (parameters) for blind separation of signals are proposed. The main motivation development new rules is to improve convergence speed and reduce cross-talk, especially non-stationary signals. Furthermore, we have discovered that under some conditions the proposed neural network models associated exhibit a random switch attention, i.e. they ability chaotic or switching cross-over output in such way specified separated signal may appear...

Bifurcation-diagram reconstruction estimates various attractors of a system without observing all them but only from several with different parameter values. Therefore, the bifurcation-diagram can be used to investigate how change values, especially for real-world engineering and physical systems which limited number observed. Although bifurcation diagrams have been reconstructed time-series data generated in numerical experiments, that targeted reconstructing time series measured phenomena...

We describe a method for reconstructing bifurcation diagrams with Lyapunov exponents chaotic systems using only time-series data. The reconstruction of is problem prediction and predicts oscillatory patterns data when parameters change. Therefore, we expect the diagram could be used real-world that have variable environmental factors, such as temperature, pressure, concentration. In conventional method, accuracy can evaluated qualitatively. this paper, estimate reconstructed so...

A dynamical associative network is constructed with chaotic neuron models interconnected through an auto-correlation synaptic matrix. Dynamics of the external inputs are analysed by output pattern sequences, largest Lyapunov exponent and response characteristic to perturbations.

In this paper, the synchronization characteristics in response to external inputs are investigated for chaotic neural networks with coupled lattices based on Newman-Watts model. The model was originally proposed a ring-coupled lattice as its initial structure. However, not suitable number of applications, including image processing. Therefore, network. As result, we find that synchronized clusters generated spatially distributed inputs, and recombination neurons into occurs case parameter...

We propose an algorithm to estimate parameter spaces by using a pruned extreme learning machine, but without principal component analysis, and we plot bifurcation diagrams in the estimated visualize changes system patterns. The estimation of can predict behavior when its parameters are changed. It be very helpful adjust optimal unknown systems. In this paper, dimension singular values trained synaptic weights based on method proposed Bagarinao et al., machine. motivate use machine through...

In this paper, we describe a chaotic time series prediction using combination of an echo state network (ESN) and radial basis function (RBFN). The ESN is neural consisting three layers, where the hidden layer (the "reservoir") composed many neurons. RBFN (RBF) for its output function. We propose model which RBFN. Time predictions Mackey-Glass equation laser are examined. Numerical experiments to examine efficiency proposed reveal that combined shows higher ability than conventional model.

We describe a method for tracking bifurcation curves from only time-series datasets. apply algorithm to unknown systems based on the reconstruction of diagrams that can estimate parameter space and oscillatory patterns when parameters change. Therefore, this track measured datasets, whereas target in previous studies are known. By curves, we obtain points with increased accuracy as compared plotted by brute-force methods. In paper, present results numerical experiments which Hénon map an...

We describe a method for reconstructing bifurcation diagrams by using extreme learning machines (ELM). Principal component analysis (PCA) is performed the coefficient vector obtained training time-series predictor. From results of PCA, we estimate number significant parameters target system, reconstruct diagram, and compare it with original one. The show that computation time required ELM considerably shorter than conventional methods. In addition, quantitatively evaluate accuracy...

The authors introduce chaos into simple mathematical neuron models which are deterministic rather than probabilistic. apply chaotic dynamics to artificial neural networks, using a model based on electrophysiological experiments with squid giant axons and numerical the Hodgkin-Huxley equations. First, explain its dynamics. also demonstrate spatio-temporal pattern of networks nearest-neighbor couplings. It is shown that couplings have abundant possible applicability dynamical memory.< <ETX...

Associative chaotic neural networks with weighted pattern storage are studied. Values of the synaptic weights conventional associative determined by an auto-associative matrix. On other hand, in this paper, we use a matrix order to store that is stronger than stored patterns. Retrieval characteristics and dynamical properties numerically analysed. As result, network retrieves strongly more frequently patterns, even case where dynamics chaotic.

We describe the reconstruction of bifurcation diagrams using an extreme learning machine with a pruning algorithm. can reconstruct diagram from only some time-series data by neural network. However, accuracy is influenced structure To improve we apply algorithm to network used for diagrams. In this study, use pruned (ELM) based on sensitivity analysis. numerical experiments, first compare predictions ELM and without Then, show effectiveness

We analyze instantaneous stability of a chaotic neural network which shows nonperiodic associative dynamics. The is composed discrete-time neuron models individuals show dynamics with certain parameter values. synaptic weights the are determined by an auto-associative matrix so that four binary patterns stored as basal memory network. It has been reported retrieves nonperiodically. However, dynamical property in each step not clarified. In this paper, during retrieval analyzed calculating...

DESPITE several notions on the gustatory code proposed over three decades, investigators have not yet reached a consensus. This paper describes new approach to analyse neural activities. Three-layer networks were trained by back-propagation learning algorithm, classify response patterns four basic taste qualities. The discrimination qualities in of rat chorda tympani fibres (CT) and cortical neurons (CN) was consistent both with correlation analysis behavioural experiments. By examining...

We reconstruct bifurcation diagrams of all components in the Rössler equations only from time-series data sets, thereby estimating attractors when parameter values are changed. In this study, we show that can be reconstructed components. addition, estimate Lyapunov spectrum diagrams. expect reconstruction requires a shorter length training using sets compared with one component. Accordingly, numerical experiments, whose is than diagram

Complex phenomena are observed in various situations, and might be generated by deterministic dynamical systems or stochastic systems. Clarifying analyzing complex is an important issue the development of technologies, such as control prediction. In this study, we propose a method for quantifying complexity graph structures obtained from chaotic time series data based on Campanharo's method. Our results show that it possible to quantify periodic, quasi-periodic, states structure, numerical...

NIRS (Near infra-red spectroscopy) is a spectroscopic device to assess the dynamic changes in hemoglobin concentration evoked by brain activity non-invasively. It has been said that data reflects activities of cortical surface. Recently, it reported not only cortex blood flow but also scalp flow. To discuss about this matter, we applied Granger causality for detect relationship between and flows motor execution. Five healthy subjects took part experiment. We measured conventional...

Spontaneous electroencephalogram (EEG) and evoked potentials measured from the scalp are considered to be attenuated by impedance of mediated biological tissues such as skull, biomembrane cortex. Voltage loss at these may deteriorate signal. In this study, we explored a possibility that an electroencephalograph bearing enhanced input-impedance could detect more sensitively short latency somatosensory-evoked potential (SEPs) high frequency oscillations (HFOs) in SEPs. We introduced negative...

In the biomagnetic measurement, signal is extremely weak compared with environmental magnetic noise. Therefore, it important to reduce noise component. There are many noise-reduction studies for MEG using Independent Component Analysis (ICA). The ICA method expectable extract and remove components from brain field measurement data. However, in these researches, each obtained independent artificially distinguished signal. We propose a of distinguishing automatically by subspace vector field....