This article presents a new framework in Robotics. It is called "android science". In the new framework, robotics is tightly integrated with science. In order to develop robots, knowledge from cognitive science is important and the developed robots can be used as test beds for verifying hypotheses in cognitive science.

View full abstract

Adaptive resonance theory neural network (ART) is an unsupervised learning system that can generate and grow the recognition categories based on the similarity between inputs and memories. By this feature, ART can solve the Stability-Plasticity Dilemma. In this report, we propose a learning system for two player games that actions or strategies of opponents change constantly. In the proposed system, an input state space is segmented adaptively by Fuzzy ART neural networks and then a player learns an input state-action pairs by the reinforcement learning. We applied the proposed system to a fighting action game that two players fight while selecting actions. As results of experiments, we show that the player acquired proper actions against opponents.

View full abstract

Grammatical Evolution (GE) is a method for optimizing the program generated by a one-dimensional chromosome and grammatical rules. GE has an advantage that invalid individuals are not generated by the genetic operations. When a certain gene changes, however, the successive genes might be used for the production rule different from the rule applied before even if they are not changed. Therefore, it is difficult to preserve the character of parents. To solve this problem, we propose GE using multiple chromosomes. In this method, multiple chromosomes as many as the non-terminal symbols in the grammatical rules are prepared. A chromosome according to the expanded non-terminal symbol is selected and used for mapping. Moreover, the technique of the wrapping are improved so that the grammatical rules which increase the number of non-terminal symbols cannot be applied when the wrapping happens. We performed some experiments, and showed the effectiveness of our proposed method.

View full abstract

This paper proposes a music/lyrics composing system consisting of two sections, a lyric composing section and a music composing section, which reflects user's impressions of theme of songs to lyrics/music. Love and nature are considered as theme in this paper. First of all, a user has a theme and image of lyrics to compose. The lyric composing section presents initial lyrics selected at random from database that is generated using existent lyrics and Markov Chain. If presented lyrics do not fit user's image, a part of lyrics not fitting user's image is changed by some other lyrics. When satisfied four lines lyrics are obtained, the music composition section starts. This section composes music fitting lyrics generated by the lyric composing section with the music composition system. The section presents combinations of four lines lyrics and 16 measures music. A subject evaluates each combination of lyrics and music whether they fit his/her image of a song. According to subject's evaluation music melody is changed by Genetic Algorithms and a part of lyrics are changed. These procedures are repeated until satisfied combination of lyrics and music melody is generated. In order to verify the validity of the presented system, subject experiments are performed.

View full abstract

This paper confirms usefulness of the situation-dependent membership function identification method based on analogical reasoning experimentally. In the first experiment, properties of analogies, which are relations between membership functions in different situations, are discussed. The result shows no significant difference between contribution rates of 4-degree polynomial expression as analogy and those of more than 4-degree polynomial expression. In the second experiment, membership functions using analogical reasoning in some situation are compared with those already obtained directly in other situations. The result shows that more subjective membership function can be obtained using proposed membership function identification method.

View full abstract

Particle Swarm Optimization(PSO) is now widely used, due to its powerful search ability and simple implementation. Recently, multi-objective optimization methods by PSO are receiving much attention. There are two differences from the single objective PSO in the multi-objective PSO. The first is an use of archive to reserve Pareto optimal candidates and the other is a selection strategy of appropriate guides for multi-objective optimization. In this paper, we propose a PSO model that introduces a topology-based guide selection scheme for multi-objective optimization. The numerical simulation results show that the proposed guide selection method is effective in the multi-objective optimization benchmark problems.

View full abstract

The nurse scheduling is very complex task. Even veteran director requires one or two weeks to create the schedule. To improve such the complexity, we have proposed a technique to optimize the nurse schedule by using the cooperative genetic algorithm (CGA). In this report, we extend the nurse schedule to permit the change of the schedule. Such the nurse scheduling requires computational time of ten times in the case of the conventional nurse scheduling. In this report, we propose a technique to execute the nurse scheduling in a parallel processing.

View full abstract

This paper presents an interactive framework for the design of bipedal walking gaits. Evolutionary computation is used in this framework. An individual in the evolutionary algorithm represents a cyclic trajectory of three joint angles. A human evaluator subjectively evaluates the robot behavior. Evaluation results by the human evaluator are incorporated in fitness calculation. New individuals are generated through evolutionary operations using the calculated fitness. We show some experimental results where walking gaits are successfully obtained by the interactive framework.

View full abstract

In these decades, research on Evolutionary Multi-objective Optimization (EMO) focused on two or three objective optimization. But, multi-objective optimization with more than three objects called Many-objective Optimization is actively researched in recent years. It is reported that the performance of well-established EMO algorithms such as NSGA-II and SPEA-II rapidly degrade with increasing the number of objectives. In this research, we propose a NSGA-II-based approach that merges objectives into some groups, and compare to existent NSGA-II.

View full abstract

It is generally important for the high performance of search in GA to use the coding which has similar neighboring relationship between in genotype space and in evaluation space, because GA searches the neighborhood of the genes with high evaluation values. However, there is no criterion for the similarity of relationship between them, then each coding is done by trial and error. This paper proposes the visualization method to grasp the relationship between genes and their evaluation values. This paper applies the proposed method to a benchmark function of multi-objective optimization problem and shows that it enables us to grasp the similarity of genes between in the genotype space and evaluation space. It also shows that the visualization result can support us to feed back into genetic operations for more efficient search.

View full abstract

In this paper, we employ a genetic algorithm to increase the vote turnout rates and to reduce the number of polling places without being lower than the actual rates by reallocating polling places. In order to estimate the decision making process of each voter, we adjust the parameters of voters in each area of the city. Through computer simulation, we could minimize the difference between the estimated voter turnout rates and the actual rates. Using the estimated parameters in each area, we increase the rates using a genetic algorithm. The simulation result shows that we can increase the voter turnout rates and reduce the number of polling places without being lower than the actual rates.

View full abstract

Recently, multi-agent simulation (MAS) attract a attention from some domain. Modeling human decision-making is a challenge in modeling and simulating social phenomenon by MAS. Then, we generate If-Then rules that show human decision-making based on the questionnaire concerning selection of medical institution and the coordinate data of GIS in this paper. We compare distinction coefficient of generating If-Then rules that fixed attribute of antecedent part with automatic generating rules that by ID3 proposed by J.R. Quinlan. As well as we decrease number of rules and advance the analysis of the questionnaire data by clustering attribute of antecedent part.

View full abstract

This paper proposes a method for emotion recognition in unintentional speech. Speech data labeled to certain one consists of several kind of emotion utterance. Even though speech data is assigned to same label, corresponding prosodic features are dissimilar with each other. So, it is hard to distinguish one class to the others. We assume the reason that several subclasses are distributed and overlapped in the feature space. In this paper, we propose the technique to improve recognition rate by detecting multiple hidden series from emotional speech. Emotional speech data with the same label are divided into multiple hidden subclasses according to prosodic feature by k-means. A set of similar hidden subclasses with various label are grouped. Grouped speech data in one hidden emotion category train one SVM, so that several SVMs for several hidden emotion categories. Experimental results show that proposed technique raised recognition rate better than the traditional one.

View full abstract

In this paper, we propose a visualization method for the distribution of population in Multi-Objective Genetic Algorithm (MOGA) based on island model. In the proposed method, each island is regarded as a cluster, then, searching solutions belonging to each island are projected together onto a two-dimensional space formed by the axes acquired by Multiple Discriminant Analysis (MDA). The visualization result enables us to grasp the distribution of population in each island and the relationship of evolution among them. This paper investigates the effects of migration of island model with the influence to each island through the experiment by a benchmark function and effective migration can be done through the visualization result.

View full abstract

In this paper we propose the exponential fuzzy numbers in Analytical Hierarchy Process (AHP), employing the fuzzy concept. The objective was determined a method of deriving the weights of criteria and their alternatives. Next, the influence of fuzzy is examined by using the membership function. Specifically we use the exponential fuzzy numbers for the pairwise comparison scale. Same experiments with AHP were conducted using the pairwise comparison scale and the new pairwise comparison scale of the method was examined.

View full abstract

The classical AHP method requires the decision-maker (DM) to express his or her preferences in the form of a precise ratio matrix encoding a valued preference relation. However, it can often be difficult for the DM to express exact estimates of the ratios of importance. One of the most natural ways to cope with this kind of problem is employing a reciprocal matrix with fuzzy-valued entries. Then some of the present authors proposed the fuzzy constraint-based approach before. This allows for a more flexible specification of pairwise preference intensities accounting for the incomplete knowledge of the DM. In this paper, we consider about weights for the fuzzy constraint-based approach and propose an extension of representation for the weights. It employs not only fuzzy concepts but also results of sensitivity analysis for weights and consistency. Moreover it is very useful for investigating data structure and realizing results of AHP. We also show it through numerical examples.

View full abstract

Though various contents are provided through the internet recently, it is not easy to collect favorite contents among huge amounts of contents in terms of user's preference. In this paper, we focus on the collaborative filtering algorithm in the recommender system. We propose a modeling approach based on Modular fuzzy model for preference similarity model in collaborative filtering. In our approach, the model is constructed through optimization of MAE(Mean Absolute Error). The model decides the weights of preference similarity from the value of correlation coefficient, the number of items and variance values of evaluation ratings of each person. Through numerical experiments compared with conventional correlation coefficient based approach using Movie Lens data, we discuss validness of the model for collaborative filtering.

View full abstract

In this paper, we focus on fuzzy random two-level linear programming problems. First, we reformulate the given problem as an alpha-two-level stochastic linear programming problem where the degree of realization of the problem is guaranteed to be greater than or equal to alpha. Introducing fuzzy goals to quantify the ambiguity in the judgement of decision makers (DMs), we transform the reformulated problem into a problem to maximize the satisfactory degree of each fuzzy goal. Next, since the satisfactory degree of each fuzzy goal in the transformed problem is a random variable, it is reduced to a deterministic two-level programming problem based on fractile criterion optimization. Then, for the reduced problem, under the assumption of the cooperative relationship between DMs, we propose an interactive decision making method to derive a satisfactory solution through interactions such that the DM updates the degree of realization of the problem and the minimal satisfactory degree.

View full abstract

The measurement and reconstruction of three-dimensional (3D) object shape with motion are difficult for the conventional time-consuming equipments. In this paper, we propose a system for the measurement and reconstruction of the 3D shape and motion of the object easily. This method is based on the Shape-from-silhouette. The Shape-from-silhouette is a category of methods to reconstruct 3D shape from images taken from various directions. Conventional shape-from-silhouette required multiple cameras or turntable for object. The proposed system takes the picture including object images from multiple directions at once. This system can reconstruct 3D mesh model with motion from image series taken by this system. The reconstructed model with motion is shown in VRML. Some experimental results showed good feasibility of the proposed system.

View full abstract

In a digital still camera (DSC) system, in most cases, a dynamic range is compressed linearly in logarithmic space. Therefore, images taken by a DSC tend to be low contrast, and give us flat impression without liveliness in comparison with silver halide photographs. In this report, a new digital image enhancement method based on linear smoothing is proposed. In the proposed method, at first, the structures of an input image are extracted by subtraction between an input image and the smoothed one. Then, the image enhancement is achieved by adding the differential image to the input image. Further, multiscale image enhancement is carried out by setting some strength of smoothing, and the resulting image, that is high contrast and well-sharpened, is provided by synthesizing enhanced images obtained in the previous step. The effectiveness of the proposed method is verified through the experiments.

View full abstract

We propose analysis system for ultrasound images based on text information of sonographic observation. The proposed system classifies 600000 ultrasound images without meta-information in Kochi medical school hospital based on image characteristics. The classified images are extracted for abnormal area based on structured data generated from sonographic observation. Using the proposed system, it is possible to reuse medical treatment information for analysis and study.

View full abstract

In a digital image processing, the methods changing colors of an image have been researched. In this report, we propose a new color changing method based on the histogram of hue-angle. The color information is transferred into the input image from the referenced image through the hue histogram in the proposed method. Such a method proposed in this report, it could be applied to the field of color coordinates. The effectiveness of the proposed method is revealed through the experiments.

View full abstract

This paper proposes a functional assessment system of autonomic nervous system using an air pressure sensor. Heart rate variability is a recognized measure of autonomic nervous system. In general, heart rate variability is measured by an electrocardiograph. However, this method has to constrain human body to fix plural sensors directly in a surface of a body. The air pressure sensor can unconstraintly detect the vital information in bed by placing it to the under of the mattress. The proposal method detects heart rate variability with this system, and evaluates an autonomic nervous system. We employ a detection method of heartbeat point using fuzzy membership functions. The experimental results show that we detect RR intervals with the correlation coefficient of 0.851 with comparison to that of electrocardiograph. Then, the errors of the HF (index of parasympathetic system) and the LF/HF (index of sympathetic system) are 11.98% and 22.18%, respectively.

View full abstract

This paper describes an ultrasonic intestine thickness determination system for low anterior resection. At the operation of rectum cancer with a circular stapler, the intestine is pulled by the circular stapler. When the thickness becomes too thin, the corresponding cells necrotize. To solve this problem, it is necessary to measure the change of the thickness of the intestine. We propose a thickness determination method of the intestine with the ultrasonic probe of 15MHz. In our method, we determine the surface point by calculating the correlation coefficient between the surface echo and acquisition waveform. Next, we determine the bottom point by calculating three fuzzy degrees: amplitude of the bottom echo, correlation coefficient between the surface and bottom echo, and interval distance between the surface and bottom echo. Finally, we calculate the thickness between the surface and bottom points. As the result, we could measure the thickness within an error rate of 5.09%.

View full abstract

Heart beat brings valuable information about human health condition. In these days,electrocardiograph has been considered the standard method of heart rate measurement. However, it is not suitable for daily healthcare because of its high cost and unease of use. In response to the problems, many studies of heart rate measurement have been done with considering unconsciousness and non-invasion. The authors have also researched and developed method and equipment of heart rate measurement by mainly using an air pressure sensor. The method employs fuzzy logic to extract heart beat point from the sensory data. In this paper, the characteristics of fuzzy logic are discussed from the viewpoints of theoretical aspects with comparative studies. The experiments were conducted to prove that fuzzy logic is suitable for measuring heart rate with using air pressure sensor.

View full abstract

In this report, we propose an error diffusion-based three-level halftoning method with arbitrary middle quantization level. In the proposed method, two images are generated through the nonlinear tone-mapping, at first. Then, those continuous-tone images are converted to bi-levels using the conventional error diffusion method, and an output image represented only by three levels is obtained by averaging the bi-levels images. The effectiveness of the proposed method is verified by applying it to gradational and natural images.

View full abstract

A fuzzy classifier based on the fuzzy c-means (FCM) clustering and parameter tuning by the particle swarm optimization (PSO) has shown decisive generalization ability in classification. The FCM classifier uses covariance structures to represent flexible shapes of clusters. Despite its effectiveness, the intense computation of covariance matrices is an impediment for classifying a set of high-dimensional data. This paper proposes a further modification for the sets of large number of high-dimensional training samples. The third type of the FCM classifier is the relational classifier. The classifier treats relational data instead of object data. All these three classifiers are equivalent when the dimensionality of feature vectors is not very large, and the relational data are obtained based on the dissimilarity measure of Euclidean distances.

View full abstract

In this paper, we propose a new clustering algorithm based on probabilistic similarity. The probabilistic similarity is formed by introducing the concept of probability into conventional similarity. First, we define the probabilistic similarity. Next, we show some examples of probabilistic similarity functions. Third, we consider a objective function with the probabilistic similarity. Furthermore, we construct a new clustering algorithm based on the probabilistic similarity by using the optimal solutions which maximize the objective function. Last, we show some numerical examples to verify the effectiveness of the proposed algorithm.

View full abstract

In this paper, two new clustering algorithms are proposed for the data with some errors. In any of these algorithms, the error is interpreted as one of decision variables - called "tolerance" - of a certain optimization problem like the previously proposed algotithm, but the tolerance is determined based on the opposite criterion to its corresponding previously proposed one.

View full abstract

This paper discusses a method of semi-supervised fuzzy clustering using target clusters. This method aims at classifying complicated data more precisely than ordinary fuzzy c-means. For this purpose the objective function of the FCM is modified by adding a term having target clusters. The solution of the modifed fuzzy c-means is shown. Numerical examples show how the proposed method works.

View full abstract

Adaptive Fuzzy c-Regression (AFCR) is a Fuzzy c-Means (FCM)-type switching regression technique that simultaneously performs adaptive data clustering and local regression model estimation by using a combined clustering criterion of regression errors for switching regression and within cluster errors for FCM. The alternating least squares (ALS) method handles mixed measurement level data by iteratively quantifying nominal variables into numerical scores so that the scores suit the current model. This paper considers an ALS technique for dealing with mixed measurement level data in AFCR. Several experimental results demonstrate the role of the tradeoff parameter in the optimal scaling and adaptive clustering phases.

View full abstract

Fuzzy c-means-based classifier optimized by particle swarm optimization (PSO) is proposed. The procedure consists of two phases. The first phase is an unsupervised clustering, which is not initialized with random numbers, hence being deterministic. The second phase is a supervised classification. The parameters of membership functions and the locations of cluster centers are optimized by PSO and cross validation (CV) procedures. Since different types of classifiers work best for different types of data, our approach is to parameterize the classifier and tailor it to individual data set. The FCM classifier outperforms well established methods such as k-nearest neighbor classifier (k-NN), support vector machine (SVM) and Gaussian mixture classifier (GMC) in terms of 10-fold CV and three-way data splits.

View full abstract

This paper presents new types of clustering algorithms by using tolerance vector. The tolerance vector is considered from a new viewpoint that the vector shows a correlation between each data and cluster centers in proposed algorithm. First, a new concept of tolerance is introduced into optimization problem. These optimization problems are based on standard fuzzy c-means or entropy regularized fuzzy c-means. Second, the optimization problems with the tolerance are solved by using the Karush-Kuhn-Tucker conditions. Next, new clustering algorithms are constructed based on the unique and explicit optimal solutions of the optimization problems. Finally, the effectiveness of proposed algorithms are verified through some numerical examples.

View full abstract

An additive clustering model has been proposed that considers the overlapping of clusters and whose target data is similarity data. In addition, an additive fuzzy clustering model that is an extension in which the degree of belongingness of objects to clusters is represented by aggregation operators has been proposed. In this paper, we propose a fuzzy clustering model in which the similarity structure is captured in a mapped space by using kernel functions and then its features are investigated.

View full abstract

In this paper we study cluster validity measures from different viewpoints. A family of cluster validity measures are investigated and compared using many numerical examples. Kernelized validity measures are also studied. Moreover an application of cluster validity measures are considered, i.e., a clustering algorithm using a cluster validity measure.

View full abstract

Fuzzy c-Medoids (FCMdd) is an FCM-type clustering method, in which the prototypes of clusters are selected from data samples, and is used for robust clustering of relational data. This paper proposes a linear fuzzy clustering algorithm based on extended FCMdd approach that extract local sub-structures from relational data by estimating 2-D linear prototypes spanned by three representative objects (medoids). Clustering criterion is calculated by using only mutual distances between medoids and objects, and new prototype is given by solving a combinatorial optimization problem. The information summarization approach is based on a similar concept with multi-dimensional scaling of relational data and can be regarded as a multi-cluster-type multi-dimensional scaling for summarizing data into several 2-D planes.

View full abstract

k-means is a basic method for non-hierarchical hard clustering and a similar result is given by a PCA guided subspace learning mechanism in an analytical batch process. This paper proposes a new robust k-means algorithm based on a fuzzy PCA-guided clustering procedure, in which a responsibility weight of each sample in k-means process is estimated based on Dave's noise fuzzy clustering mechanism. Cluster cores are extracted by estimating the connectivity indices among samples in k-means from fuzzy principal component scores. A spectral ordering technique, which is useful for visual assessment of clustering result, is applied to the connectivity matrix in order to represent cluster structure in the diagonal block structure.

View full abstract

Four algorithms for sequential clustering using fuzzy cluster memberships are developed and their effectiveness and efficiency are compared. First is a known technique of the mountain clustering, and the second is a new method of mountain medoid clustering which uses individuals instead of grid points in the mountain clustering. The third is an algorithm using possibilistic clustering which the authors have developed, and the fourth is an sequential algorithm using the noise clustering with one cluster. In these algorithms, the number of clusters need not be specified beforehand. These four algorithms are described; their performances are discussed and tested using different numerical examples.

View full abstract

In order to investigate the structure and unique point of brain waves data, new analysis methods are expected. However, brain waves data is not a constant value, the extraction of a peculiar part is very difficult. Based on this point of view, we propose a way of Peculiarity Oriented Mining (POM) for peculiar part discovery in multiple brain waves data. The proposed approach provides a new way for automatic analysis. The main task of POM is the identification of peculiar data. We will take human computation as an example. In this paper, we evaluate the POM technique by using the simulation data first. And, we show the effectiveness of the proposal technique. Finally, we analyze how peculiar part of the brain waves in a human computation process.

View full abstract

In this paper, we consider the optimal flow assignment problem of finding a set of link flows that satisfy the requirements and minimize the average end-to-end network delay for a given topology characterized by the capacities and costs of links. This problem can be formulated as a nonlinear programming problem. In recent years, a growing body of literature suggests the use of genetic algorithm as one of powerful heuristic search methods to deal with many hard-solving problems. For highly constrained problems, conventional genetic operators often yield illegal solutions in the sense of violation of system constraints. In this study, we present an implementation of genetic algorithms to the optimal flow assignment problem. Non-uniform mutation and arithmetical crossover are adopted to guarantee the legitimation of offspring. Simulation results show that the proposed approach performs well for this problem.

View full abstract

The purpose of a job-shop scheduling problem (JSP) is to find a schedule with the minimum makespan. In the scheduling problem in the real world, in addition, it is necessary to consider the delivery time and the holding cost. In JSP with the delivery time (JSPD) proposed by Asano et al., the holding cost of the product completed before the delivery time is not considered. In this paper, we propose a new model (JSPDH) from which the concept of the holding cost is added to JSPD, and we propose a method for solving JSPDH by a genetic algorithm. The effectiveness of the proposed method is clarified by numerical experiments using benchmark problems for a job-shop scheduling.

View full abstract

The Transportation Planning (TP) is well-known basic network problem which can be defined as the problem that calculates the optimal amount of deliveries. But for some real-world applications, it is often that the TP model is extended to satisfy other additional constraints or performed in several stages. In addition, the concept of inventory and time are not included. Moreover, today's distribution channel becomes a flexible form. In this study, we formulate a flexible logistics network model with concept of inventory and time. The purpose of this model is minimization of the total cost that includes inventory cost. Moreover, the network form has flexible connection and a certain period divided into some terms. To solve the problem, we propose the hybrid Genetic Algorithms (hGA) approach by using a priority-based encoding method. Finally, numerical experiments with various scales of logistics network problems are used to show the effectiveness and the efficiency of our approach.

View full abstract

It is necessary to develop car navigation technology that provides dynamic information about environmental condition and road traffic as an element of Intelligent Transport System (ITS). The Dijkstra method can solve single-objective shortest path problem in polynomial time. However, to allow tourists to travel to destination efficiently, we need a navigation system which specify multiple and near optimal choices according to multiple different and conflicting criteria, for example, on total time, distance, and total expenses. We defined three objective network design problem. To solve this problem which has multiple criteria, we applied a multiobjective genetic algorithm, aiming at simplification of parameter tuning and efficiency improvement of maintaining diversity. The traffic forecasting is also necessary to avoid traffic congestions dynamically. We use this genetic algorithm to calculate the near Pareto-optimal solutions.

View full abstract

The authors have recorded EEGs from subjects in viewing four types of arrows presented on the CRT. Each of four symbols have direction of upward, downward, leftward and rightward, respectively. Subjects were asked to make read them silently. The equivalent current dipole source localization (ECDL) method had been applied to these ERPs. With arrow symbols, ECDs had been localized to areas related to the shape recognition. Taking account of these facts, the authors investigated a single trial EEGs of the subject precisely and determined effective sampling latencies for the discriminant analysis to four types of arrow. The number of the variates is twenty one by three channels, therefore the sum is sixty three variates. Results of the discriminant analysis for four type objective variates were mostly more than 80 %. By four type code of infrared rays according to the discrimination results from a PC, the authors have controlled a micro robot the e-puck.

View full abstract

Rat hippocampal neurons were cultured on a dish with 64 micro planer electrodes. A complex network of neurons was formed and the network was able to distinguish patterns of action potentials evoked by different electrical current inputs. We integrated a living neuronal network and a Khepera II robot or robot made by LEGO mindstorm NX kit as a body for contacting to outside world. Using self-tuning fuzzy reasoning, we associated a distinct spatial pattern of evoked action potentials with a particular phenomenon in the outside of the culture dish. We succeeded in performing collision avoidance behaviour with premised control rule sets. During collision avoidance, the responding pattern of evoked action potentials was stable and robust against perturbation to spontaneous network activity. These results suggest that a cultured neuronal network can represent particular states as symbols corresponding to outside world.

View full abstract

We propose a vector conversion method of words and a language processing neural network (LPNN) with additional learning. A vector conversion method is a technique of generating word vector. The LPNN receives Japanese texts, and outputs or infers the knowledge that related to the inputs. The LPNN can learn inputs, even if the first learning process is completed. We carried out several kinds of computer simulations to confirm validity and effectiveness of the network.

View full abstract

Recently, a lot of researches on Brain Computer Interface (BCI) have been reported. It is expected that BCI will help patients like those with ALS to control a wheel chair or to communicate with other people just by the thoughts. It is necessary for the application of BCI to recognize the thoughts from the Electroencephalogram (EEG), and it is an approach to employ the change of amplitude on EEG. Event-Related Desynchronization (ERD) or Event-Related Synchronization (ERS) related to body's movement is one of them and well known in this research area. This paper tries to recognize two thought, "Relax" and "Image of movement", using ERD. In this paper, some experiments are done and the feedbacks are shown for higher accuracy of the recognition.

View full abstract

A fuzzy classifier based on the fuzzy c-means (FCM) clustering has shown decisive generalization ability in classification. For classifying the blood oxygen level dependent (BOLD) responses of the brain, a way of directly handling high-dimensional fMRI signals is adopted. Our goal is to distinguish the BOLD responses to recalling tasks from those to resting. We use the signals from wide areas of the brain, which form a set of high dimensional data vectors. The evolutionary algorithms are introduced for parameter optimization of the classifier. Relatively low classification error rate was obtained by both the two optimization methods. The error rate on the test set surpassed the support vector machine (SVM), which is a high performance classifier and well suited for the set of high dimensional data.

View full abstract

Development of operation skills to supervise a dynamic mechanical system with a large time constant is analyzed from the perspective of emerging regularities between "proximal," i.e., directly perceptible or manipulatable, variables and "distal" properties of the human-machine system. Our skill analysis aims to confirm increasing ability to accurately identify those situations requiring the human operator of adjusting the system behavior as his experience enlarges, as well as differentiated use of maneuvering options in accordance with adjustment purposes. This paper proposes an evaluation method of those consistencies' formation processes through a couple of entropies with respect to the observed distributions in proximal and distal variables to specify the task situation.

View full abstract