We present maximum likelihood (ML) methods for estimating evoked dipole responses using electroencephalography (EEG) and magnetoencephalography (MEG) arrays, which allow for spatially correlated noise between sensors with unknown covariance. The electric source is modeled as a collection of current dipoles at fixed locations and the head as a spherical conductor.We permit correlation between the dipoles' moments. The dipoles' locations and moments are estimated. We also propose ML-based methods for scanning the brain response data, which can be used for imaging the brain's electromagnetic activity.

View full abstract

In this paper, we analyze two subspace identification methods. The one is an N4SID-like algorithm which performs poorly in certain conditions where the past signal and future input spaces are nearly parallel. The other method, based on a preliminary orthogonal decomposition of output data space, is more robust and reliable than the first method in critical cases. Numerical results demonstrate a substantial improvement of performance in such a parallel case.

View full abstract

In this paper we derive a subspace-based state-space identification method for continuous-time systems. By using δ-operator [15], we transform the continuous-time system to the discrete-time δ-operator state-space model which converges to the original continuous-time model as the sampling period goes to zero. Then we obtain the estimates of system matrices by applying some well-known subspace identification methods such as MOESP [2] to the discrete-time δ-operator state-space model. Numerical examples are included to show the effectiveness of the proposed method.

View full abstract

Subspace-based State Space System IDentification (4SID) methods have attracted much attention because of being essentially suitable for multivariable system identification. The methods have been demonstrated to perform well in a number of applications, but the properties of these have not been fully analyzed or understood yet. For applying the methods, no assumptions on structure of realization are needed and any coordinate transformation is allowed for the estimates. This is one reason why many kinds of properties expected for identification procedures have not been clarified yet. We illustrate, by using Schur complement, another interpretation of the 4SID method. And we will propose a recursive formula for the error covariance matrix in the 4SID procedure. The results in this paper can be useful for 4SID methods.

View full abstract

Identification of transfer functions in a tri-feedback system is discussed from the viewpoint of stochastic inverse problem. Information is contracted due to observations, and we have an innovation model equivalent to the tri-feedback system corresponding to a way of observation. In an underdetermined case, the identification problem of transfer functions can be discussed via the numerical innovation model, which is obtained from a stable solution of a Riccati equation.

View full abstract

For the decision Support under a multi-objective environment, it is effective to offer Pareto optimal solution subset with the uniform distribution to decision-maker. In this paper, a new optimization method in which the concept of the cannibalism is introduced in BUGS (A Bug Based Search Strategy using Genetic Algorithms) is proposed. The introduction of the concept of the cannibalism realizes the uniform distribution of Pareto optimal solutions. The numerical experiment using typical continuous and discrete multi-objective optimization problems clarifies the usefulness of the proposed method.

View full abstract

This paper discusses optimal dynamic investment policies for investors, who make the investment decisions in each of the asset categories over time. We construct the framework integrating stochastic optimization and Monte Carlo simulation for dynamic asset allocation, and we propose the linear programming models using simulated paths to solve a large-scale problem in practice. Linear programming models can be formulated to adopt either a fixed-value rule or a fixed-amount rule instead of the general fixed-proportion rule. These formulations can be simply implemented and solved very fast. Some numerical examples are tested to illustrate the characteristics of the models. These models can be used to improve the trade-off between risk and expected wealth, and we can get interesting results for the dynamic asset allocation policies.

View full abstract

In this article, we develop a computation algorithm to generate the optimal block replacement schedule which minimizes the expected cost per unit time in the steady-state. Since the renewal function of the inter-failure time distribution is involved in the expected cost representation, one has to evaluate the renewal function numerically and calculate the optimal preventive replacement schedule. Then, the radial basis function neural network (RBFNN) is applied to calculate the renewal function effectively. Finally, in numerical examples, we show that the RBFNN approach can generate the optimal block replacement schedule with higher precision.

View full abstract

From 1980s mean field approximation algorithm has been applied by many researchers to solve combinatorial optimization problems. It has been well known that critical phenomenon shows up during the process of annealing in the mean field approximation and critical temperature has also been studied.In this paper, an adaptive annealing method is introduced to the mean field approximation algorithm, where the annealing schedule is changed according to the prediction of critical temperature. This Adaptive mean field approximation algorithm is applied to graph partitioning problem and experimental results show that the computing speed is faster than mean field approximation algorithm with the same solution quality.

View full abstract

In this paper, we consider an M/G/1 queueing system with removable server subject to failure interruptions, where the failure occurs only during operating busy period. More precisely, jobs arrive according to an identical and independent Poisson stream, and are processed with identically and independently distributed general service times. Two kinds of recovery functions for a system failure are analyzed. One is the retry function, the other is the repair function. In the retry model, after a system failure the system undergoes instantly the corrective repair and the interrupted job is processed again after the system becomes as good as new. On the other hand, in the repair model, the minimal-repair is executed for each failure. In the both models, the preventive maintenance is carried out at the termination of the process of a job. In the framework of the N-policy for the M/G/1 queue, we formulate the relevant expected costs under two recovery functions, respectively, and derive the optimal control-limit policies minimizing them.

View full abstract

Since most real processes are nonlinear systems, nonlinear models are required for accurate characterization of the systems. However, nonlinear identification techniques have been less developed. In this paper, we proposed an iterative method to identify nonlinear systems that can be described by a nonlinear Wiener model. The result shows that the model parameters converge quickly to the asymptote, and prediction error is greatly reduced by the model parameter optimization. For the future control of Wiener systems, an Extended Kalman Filter-based state estimator is designed, which allows the linear control technique to be applied to nonlinear Wiener systems directly. The numerical simulations show that the proposed algorithm is effective and has a low computational cost for identification, and the states of the Wiener model are estimated successfully.

View full abstract

This paper is concerned with a fault detection scheme to nonlinear black-box systems based on Quasi-ARMAX model [4], in which the system non-linearity is incorporated into model parameters by using nonlinear non-parametric models (NNMs). The model can be considered as a multi-ARMAX model which consists of interpolated multi local linear models weighted with basis functions used in the NNMs. The fault detection is then executed by model discrimination approach using KDI (Kallback discrimination information) as the detection index. The effectiveness of the method has been confirmed through numerical simulation studies.

View full abstract

Identifying a neural network model is equivalent to multidimensional, nonlinear optimization. This paper presents a modified adaptive random search scheme for the optimization. The idea is to introduce a sophisticated probability density function (PDF) into a usual random search scheme for generating search vector. The new PDF provides two parameters that are used respectively to control local search range and search direction based on the past success-failure information so as to improve the searching efficiency. Computer simulations show that the new adaptive random search algorithm is a good alternative for the case where it is difficult to apply the well-known backpropagation algorithm.

View full abstract

In this paper, deterministic chaotic systems with random perturbations are investigated. The stationary distribution of the chaotic system with additive random perturbations is theoretically obtained, where the invariant measure of the chaotic system is described by a linear function and random perturbations are given by independent random variables with the identical and uniform distribution. An example with numerical experiment shows the validity of results obtained in this investigation.

View full abstract

Ergodic type Bellman equations of risk-sensitive control in general cases are considered. We study the existence of solutions of the equations relating to the eigenvalue problem of Schödinger operators. Furthermore, by taking their singular limits, we obtain particular viscosity solutions of Hamilton-Jacobi-Isaacs equations of differential games which correspond to nolinear H_∞-control problems.

View full abstract

The main aim of the present paper is to give an explicit formula of the price of the currency put-option with knock-out. In the real market, the purpose of holding the option is to hedge the currency-risk of a foreign asset. To lessen the hedging cost, the option is knocked-out if the price of the asset goes up enough at the expiration date of the option. So the knock-out of the option is measured by the level of the price at the expiration date of the foreign asset. The formula is given in section 2 and in section 3 we compare it with the pricing formula of the plain option by Black and Scholes[1]. The model must be essentially a multi-factor one since the exchange rate and the asset price are not perfectly correlated in general. We use an extension of Black-Scholes economy in which the risk-free rate is constant and the asset price is modeled by a geometric Brownian motion. The model is given in section 1.

View full abstract

We construct a forward LIBOR-rates model, which is calibrated to all “implied volatility-curves of caplets” (which is assumed to be given) under a framework of Jamshidian and Musiela-Rutkowski (cf., [2-3]). Further, we suggest an algorithm of the construction starting with market datas of implied volatilities of caps.

View full abstract

The aim of this paper is to calculate the replicating portfolio of two geometric average options in the Black-Sholes model. Seeing this calculation, we can observe that the delta hedge of these options have simple forms. So, it is easy to create the hedging scheme of these options.

View full abstract

The problem of evaluating the goodness of statistical models is important in various fields of statistical science. Akaike's information criterion [1] provides a useful tool for evaluating models estimated by the method of maximum likelihood. By extending Akaike's basic idea, several attempts have been made to relax the assumptions imposed on AIC and obtained information criteria which may be applied to various types of statistical models. In this paper, we briefly review the definition of the information criteria GIC and EIC, and then show some of their modifications which can yield more refined results than previously proposed criteria.

View full abstract

This paper tries to explain mathematical foundation for statistical inference using hierarchical parametric models. By using Sato-Bernstein's b-function, we show that the asymptotic form of the average Kullback distance between the true distribution and the Bayesian estimated one is equal to λ₁/n － (m₁ － 1)/(nlogn), where n is the number of empirical samples. We also show that the constant values λ₁ and m₁, which are invariant under the bi-rational transforms, can be calculated by resolution of singularities in algebraic geometry, and that λ₁ is smaller than the number of parameters. Even in the case when the true distribution is not contained in the parametric model, hierarchical models with Bayesian estimation are better learning machines than regular statistical models.

View full abstract

Based on the theory of fuzzy recognition and fault diagnosis, this method only depends on experience and statistical data to set up fuzzy query relationship between the outside phenomena (fault characters) and the fault sources (fault patterns). From this relationship we can obtain the most possible fault sources, so that we can reach the goal of quick diagnosis.

View full abstract

In this paper, the author proposes the vague estimators of unknown system parameters using some kind of fuzzy random data(FRD). The FRD described in this paper have intrinsically both properties of vagueness and randomness and they are considered as the realizations of output processes of fuzzy stochastic ones.

First, using the set representation approach, the reasonable definitions of fuzzy stochastic processes(FSPs) and their statistical moments up to second ones are proposed. Secondly, the estimators of statistical moments of FSPs are proposed using only FRD. Finally, the fuzzy estimators of unknown system parameters in the simple time series models are proposed and examined numerically.

View full abstract

Three methods are used for restoring environmental noise-corrupted speeches. The methods use a conventional Wavelet Transform, a Wavelet Packet Transform, and low-pass filters based on moving-average. The prominent result is obtained by using a conventional Wavelet Transform. In this paper, we proposed a new method of optimum decomposition of the speech using the Wavelet Transform.

View full abstract

An application of a new multi-scale stochastic theory called the Multiple Tree (MT) theory, will be presented in this paper. The MT construction itself is an aggregation of several single trees where these trees are connected each other through a Gaussian random vector. The nature of the MT is suitable to th computer network environment. In this paper such advantage will be exploited in estimating 1/f signal given ill-posed observation over the network.

View full abstract

In this paper, a probabilistic signal processing method which is possible to give any methodological suggestion to the measurement method of compound and/or accumulation effect in electromagnetic environment is discussed. In order to extract latent various interrelation characteristics between waved environmental factors (sound, light and electromagnetic wave) leaked from the real operation VDT, an extended regression system model reflecting hierarchically not only linear correlation information of the lower order but also nonlinear correlation information of the higher order is firstly introduced. Then, through identifying each regression parameter of this model under two different environments (with and without a background noise), two evaluation methods for predicting the fluctuation distribution from the one to the other between above-mentioned waved environmental factors are newly proposed. Finally, the validity and effectiveness of this proposed methods are experimentally confirmed too by applying it to the actually observed data leaked by a VDT with some television games in the room of an actual working environment.

View full abstract

Probabilistic principal component analysis is a multivariate Gaussian model where the principal components are used. The mixture of such kernel functions can be a general tool for expressing probability density functions. However, it has not been extensively discussed how to decide the number of kernels nor where to fix the initial points. In this paper we propose to use the genetic algorithm to overcome those problems.

View full abstract

In this paper, a stabilization method for the fast QRD inverse-updates adaptive filtering algorithm by leakage is proposed and its stability analysis is presented using the averaging principle. It is shown that the proposed algorithm is stable if the variance of the one-step ahead linear predictor of the input signal is smaller than that of the prediction error with the forgetting factor and the leakage factor sufficiently close to 1.

View full abstract

A lot of algorithms have been proposed to solve the least squares (LS) problem for transversal adaptive filters, that is, finding the tap-weights of the filter which minimizes the exponentially weighted sum of the squared errors. However, some are computationally consuming as the recursive LS (RLS) algorithm and others are numerically unstable as the fast RLS(FLS) and RLS algorithms. The predictor-based LS (PLS) algorithm [9] gives the exact solution of the LS problem and its numerical stability has been proven by the linear time-variant state-space method. its computational complexity comparable to the RLS algorithm can be reduced to the linear order when the input signal is sufficiently modeled by an autoregressive of rather small order, since it is based on the predictors which can be truncated. We call it the fast PLS algorithm. Its good numerical properties have been confirmed, however, its convergence properties are not elucidated yet. This paper shows that the fast PLS algorithm has almost the same convergence properties as the PLS and RLS algorithms though its computational complexity is much smaller than them.

View full abstract

We consider the blind deconvolution problem of FIR multiple-input multiple-output channel systems with or without transmission zeros at the origin. The number of the outputs (of a system) is assumed to be greater than the number of the inputs. A channel system is said to be equalizable if there exists an FIR equalizer which equalizes it. It is shown that any equalizable FIR system can be expressed as a product of an irreducible system and a paraunitary system. Based on this fact, it is shown that any equalizable FIR system can be converted to a paraunitary system through the whitening of the observations using the second-order statistics. When the original FIR system has no transmission zero at the origin, the paraunitary system obtained through the whitening of the observations becomes a constant (static) system.

View full abstract

An algorithm for the subspace approach to blind identification of multichannel (or single-input and multiple-output) FIR systems is proposed. The subspace approach requires the so-called noise subspace spanned by some eigenvectors of the correlation matrix of observations. In this paper, it is shown that a subspace of the noise subspace can be obtained by one-step scalar-valued linear prediction and is sufficient for blind identification. In place of eigenvalue decomposition, the proposed algorithm utilizes the linear prediction and hence is computationally effective. Computer simulations are presented to compare the proposed algorithm with the original one.

View full abstract

Criterion for the factor selection of on-off decision making by using conditional probability was introduced. The criterion is based on the divergence between the probability density functions of Go events and Nogo events. Numerical examples showed the usefulness of the proposed criterion.

View full abstract

For the evaluation of actual sound insulation systems, a new method is proposed by introducing a stochastic mixed model on an intensity scale. The change of the system characteristic due to internal factors within a long period is functionally described by introducing a multiplicative system model composed of input and the random parameter. This model is identified based on entropy and Kullback's information criteria available for the non-Gaussian fluctuation. By using the identified model, the statistical moments of the random parameter are estimated recursively. Thus, after expressing the probability distribution of the output on an intensity scale to an arbitrary random input in the frame of the statistical type Laguerre series expansion form, its statistical moments can be predicted and its expansion coefficients can be predicted. Finally, the proposed method is experimentally confirmed by applying it to actual sound insulation systems.

View full abstract

Learning Automaton (LA) is a representative model with the properties of outstanding learning ability, autonomy and theoretically guaranteed convergence in learning process[1]. But for function optimization problems, the problematic point is with the increase of solution space, the output number of stochastic automaton also increases, therefore convergence is possibly very time-consuming.

For alleviating the problem, in present paper, we introduce Genetic Algorithm (GA) to existing LA. According to GA, a searching space is constructed to look for the optimal output from the entire output space and the way of searching for the optimal output from the smaller-sized searching space is observed. To verify the efficacy, for multi-variable function optimization problem under parallel environment, the parallel picture of this method is drawn.

View full abstract

Learning and evolving in artificial agent is an extremely difficult problem, but on the other hand, a challenging task. At present the studies mainly centered on single agent learning problem. In our case, we use simulated soccer to investigate multi-agent cooperative learning. Consider the fundamental differences in learning mechanism, existing reinforcement learning algorithms can be roughly classified into two classes-that based on evaluation functions and that of searching through policy space in direct. Genetic Programming developed from Genetic Algorithms is one of the most well known approaches that belong to the latter. In this paper, we give detailed algorithm as well as data construction description that are necessary for learning single agent strategies at first. In the following sections, we extend developed methods into multiple robot domains moreover. We investigate and contrast three different solutions-single agent learning, simple team learning and sub-group learning and conclude the paper with some actual experiments and result analyses.

View full abstract

In this paper, we have presented the learning automata solution to the NP-hard problem, the graph partitioining problems (GPP), which involves partitioning the nodes of a graph G into K sets of equal size so as to minimize the sum of the costs of the edges having endpoints in different sets. We proposed a method for this problem, a parallel learning automata (LA) method. This method aimed at the parallel and solution accuracy of the GPP and applied local evaluation function. The simulation results show that, our learning automata based method has good performance on computing time and has some superiority with the increase in the number of nodes.

View full abstract

In network computing system, load balance is a job distribution among multi-processors. Round Robin and random job allocation are the two simplest algorithms. The problem with these two approaches is they don't weight the loads when sending them out to the addresses. Therefore, for efficiently making use of the network, a job scheduling mechanism is required to make reasonable job assignment. This mechanism should be able to respond dynamically to the job transfer address and at the same time properly allocate jobs when the changes in traffic or in individual computer processing speed or in the length of wait-queue are not predictable.

In this paper, on the load distribution problem in parallel distribution computing system, learning automaton is introduced to the scheduling mechanism, which is able to dynamically respond to job addresses and make proper job assignment.

View full abstract

We propose a model of multi agents with multi-teacher learning automata for cooperative pursuit problems. Each agent must have a communication mechanism to realize cooperative properties. However, since the communication mechanism makes the theoretical analysis of the total system difficult, we make its effects on the completion of the task clear using computer simulation. We show that our method has some efficient properties of pursuing with cooperation.

View full abstract

In recent years, soft computing techniques such as neural networks, GAs, and etc. have been successfully applied for constructing various intelligent decision support systems. In this paper, we shall try to utilize neural networks to construct an intelligent decision support system for dealing TOPIX. Further, we shall suggest that the proposed decision support system can be improved by another soft computing techniques such as GAs, TD-Learning, and etc.

View full abstract

Optimal filters for extracting time-varying harmonics such as a voiced sound from the noise-corrupted observation are proposed. They are derived through the Kalman-Bucy filter analysis in which the dynamics of amplitude and pitch fluctuations are introduced into the state-space signal model. The Laplace analysis to the filter equation leads to three types of comb filters, i.e., the uniform-BW (-bandwidth) type, the uniform-Q type and those mixture type which have robustness to the amplitude fluctuation, the pitch fluctuation and both of them, respectively. All-pole digital filters can be also realized for real-time processing. Examples of filter design are presented, and the performance of harmonics extraction is examined by comparison between the uniform-BW type and the uniform-Q type.

View full abstract

This paper is concerned with the application of the H_∞ and the Kalman filtering techniques to GPS positioning, specially to a recursive estimation algorithm for the static carrier phase differential method. In order to speed up the algorithm and to keep the positioning accuracy, we propose an algorithm utilizing the H_∞ filter in the early stage of the estimation, and then switching to the Kalman filter. Finally, the experimental results by using real receiver data obtained at static points are shown.

View full abstract

In this paper, an effective tracking method for constant speed maneuvering targets which move on the 2-dimensional plane is proposed utilizing kinematic constraints of the target effectively. First, dynamics of maneuvering target are derived. Kinematic constraints which express target's characteristics are incorporated into not dynamics but measurement equation as pseudo-measurements. Based on the extended Kalman filtering theory, the tracking filter is derived from the linear dynamics and the augmented nonlinear measurement equation. The effectiveness of the proposed method is examined by numerical simulations.

View full abstract

A generalized Kalman filter is proposed, which can be applied to any state space model with exogenous variables or inputs.

View full abstract

For evaluating the acoustic systems in the actual measurement environment, it is inevitable to consider the effects of additive background noise. Nevertheless, the non-Gaussian characteristics of energy acoustic signals and the nonlinear observation mechanism often bias the estimates of unknown acoustic signals, as well as the randomly fluctuating background noise. In this paper, a new type of wide sense digital filer has been proposed for detecting acoustics signals with use of decibel observations, based on the expanded characteristic functions of Mellin and Lapace transform types and the orthogonal projection theorem. The proposed algorithm will be experimentally confirmed.

View full abstract

The authors previously proposed a length measurement system for straight pipes using stationary waves, which were generated in the pipe. The system which modeled the stationary waves as output of a linear dynamic system enabled to realize an on-line high-accurate pipe length measurement. This paper presents a length measurement system for pipes with curvatures and branches by extending the preproposed measurement system and experimental results show that accurate pipe length measurement is realized by the proposed method.

View full abstract

In this paper, system identification of anti-vibration units in semiconductor exposure apparatus is discussed. Since the semiconductor exposure apparatus has multi-degrees-of-freedom (multi-DOF) motional mechanism, the system must be treated as a Multi-Input-Multi-Output (MIMO) system. As a modeling method, the system identification methods are utilized to semiconductor exposure apparatus, in particular subspace based state-space system identification (4SID) method is applied. Effectiveness of the 4SID method to MIMO system identification is examined through experimental data. Moreover, an estimation method of physical parameters of semiconductor exposure apparatus based on identified state-space model is presented. By identifying physical parameters of semiconductor exposure apparatus, it will be expected that fault detections and diagnosis of the apparatus can be done easily.

View full abstract

We propose a mathematical method to analyze the transient change in the probability of gene regulation protein particle ( a repressor protein ) for the first time arrival at the target operator gene region of the DNA chain. The differential equation for the probabilistic behavior of the repressor molecular particle was diffusion type. The probability was expressed by the modified Bessel functions. By applying the Allen's approximation for the modified Bessel function, we obtained an approximation form of the first time arrival probability of the gene regulation protein particle immediately after the onset of the reaction. The impulse response of the probability showed considerable oscillation. The present result indicates the genetic expression particularly at its initial phase is unstable. The present method will be available for evaluating the temporal change in the probability of the gene regulation.

View full abstract

In this paper, we study about restoring the signal which is corrupted by additive noise. In this category, we need the smoothing filter which can utilize both rank- and temporal-order information of input data. It is well known the LWOS-filter is one of nonlinear filters which can utilize two information. However, the LWOS filter suffer from a rapid growth in complexity as the window size increase. It is impossible to design the LWOS filters with 5 × 5 window. This limitation of window size is fatal defect for image processing. This paper presents a novel form of LWOS filter called the modified LWOS (M-LWOS) filter. M-LWOS filters defined in the threshold decomposition domain. In the case of the M-LWOS filter, arbitrary window size of this filter can be designed. We study a design method of the M-LWOS filter and show the effectiveness of the proposed filter.

View full abstract