We have been proposing a framework Rough Non-deterministic Information Analysis (RNIA), which handles rough sets based concepts in not only Deterministic Information Systems (DISs) but also Non- deterministic Information Systems (NISs). We have recently developed some algorithms and software tools for rule generation from a NIS. Obtained rules characterize the tendencies in a NIS, and they are often applied to a decision making. However, if the condition parts in such rules are not satisfied, obtained rules are not applied to a decision making. In this case, we need to examine the NIS, directly. In this paper, we add a question-answering function with criterion values to RNIA. For a set CON of condition attribute values, this function picks up decision attribute values DEC with maximum and minimum values of support(CON=>DEC) and accuracy (CON=>DEC). According to these criterion values, we define the validity of the decision DEC.

View full abstract

In this paper, we investigate rule induction from decision tables with imprecise decision attribute values. In such decision tables, we assume that decision attribute values are specified imprecisely. Rough sets are defined for the imprecise decision tables with some proper modifications. Based on the proposed rough sets, several rule induction schemes are considered. In each rule induction scheme, the conventional decision matrix method is extended to the case of imprecise decision tables. A simple numerical example is given to show the differences of rule induction schemes. Some future topics are described with concluding remarks.

View full abstract

In this paper, we study rough sets in incomplete information tables, which are data sets containing some missing values. We proposed a rough set approach to incomplete information tables. In the proposed approach, object sets are represented by pessimistic and optimistic approximations. These approximations deal with uncertainty caused by not only indiscernibility of objects but also missing values. Moreover, we introduce the concept of the variable precision rough set approach into the proposed approach. As a result, the membership degrees of objects to a set of object become intervals.

View full abstract

Bags alias multisets have important generalizations, among which R-bags (real-valued bags), F-bags (fuzzy bags), and G-bags (generalized bags)are studied. In particular R-bags have interesting theoretical properties when contrasted with fuzzy sets. In this paper a class of complement operations and $s$-norms are studied and generalizations are mentioned. Moreover bag relations and their compositions using max-s algebra and max-t algebra are discussed.

View full abstract

In this paper, we propose a model which enables an agent to select his actions based on variable accessibility relations. We formulate in the model relationships among variable neighborhoods, the agent's observations, and the agent's behaviors in a framework of rough set theory and topological spaces. The main task is to explore a method by which we can select sizes of neighborhoods under given contexts. We also show simulation results of the proposed method.

View full abstract

In this paper, we propose a new evaluation method of relative reducts based on classification ability of condition attributes. Extraction of relative reducts and decision rules are important elements of the data analysis by rough set theory. As an evaluation method of relative reducts, the authors have proposed an evaluation criterion of relative reducts based on partitions by equivalence classes. However, this method needs to construct equivalence classes for evaluating relative reducts, and moreover, does not evaluate each condition attribute. The new method proposed in this paper evaluates relative reducts by combining evaluation results of classification ability of condition attributes.

View full abstract

In this paper, we introduce an approximate calculation method of a relative reduct based on an evaluation method of classification ability of condition attributes with respect to decision classes proposed by the authors. Because it has been proved that computational complexity of calculating all relative reducts of a given decision table is NP-hard, many algorithms have been proposed to calculate relative reducts approximately. By our proposal, we intend to calculate a relative reduct with as better evaluation by the evaluation method of relative reducts as possible. We applied the proposed method to Zoo data set in UCI machine learning repository, and obtained the best relative reduct among 33 relative reducts of Zoo data set.

View full abstract

We consider introducing logic minimization techniques into rough set theory. Computational complexity of calculating all relative reducts is NP-hard and therefore some heuristic methods are needed for real-time calculation of decision rules from huge size data. Logic minimization is a useful heuristic method for generating minimal decision rules with respect to minimizing the number of rules.

View full abstract

A new method for deciding on the class borders is also proposed to improve the accuracy of clustering using the spherical SOM. The method approximates the border on the SOM from nearest neighbor data of different classes.In this paper, the advantages of both methods using the spherical SOM, the clustering analysis and the decision on the class borders, are successfully demonstrated with two examples.

View full abstract

The GTM (Generative Topographic Mapping) algorithm was introduced by Bishop et al. as a probabilistic re-formulation of the self-organizing map (SOM). The GTM algorithm captures the structure of data by modeling the data distribution in terms of nonlinear transformation from latent variables to data space, and which is used as a data visualization tool. The object of this paper is to visualize time series data using GTM. The standard GTM algorithm assumes that the data are independent and identically distributed samples. For time series, however, the i.i.d. assumption is a poor approximation. In this paper we propose the extension of the GTM to handle time series, which we call the GTM-ARHMM algorithm, by assuming that the time series is generated by an Auto-Regressive Hidden Markov Model (ARHMM).

View full abstract

We propose a method to build topological maps from vision sensor information for mobile robots navigations. The method is based on Evolving Self-Organizing Maps (ESOM) proposed by Deng. Nodes and paths represent essential visual information in local environments and similarity relationships between environments respectively in Topological maps built by the ESOM. Moreover, the ESOM can online build the maps in real-time without pre-defined landmarks and metrical information of environments. In this presentation we show the validity of this method.

View full abstract

A bone conduction microphone detects a vibration of a bone, e.g., jawbone. It is useful for communication in extremely noisy places such as in an engine room in a ship or on an airfield, because it has a strong resistance to noise in the air. However, its tone quality is bad. The aim of this research is to improve the quality of a bone conduction voice as much as possible to the quality of an air conduction voice. In this report, we propose a novel voice conversion method from a bone conduction voice to an air conduction voice. The proposed method employs a neural gas network to perform a uniform quantization, and uses local linear conversion models from bone to air conduction voices, to realize a stable conversion performance. The validity and the effectiveness of the proposed method have been verified by applying it to the tone quality improvement of the real bone conduction voice.

View full abstract

An Elman-type feedback SOM (EFSOM) is a revised version for dealing with time-variant information. In the preceding studies, the EFSOM is applied to an on-line character recognition task, and it shows good performance on adaptability for both spatial displacement and temporal elasticity. In spite of such success through increasing the number of winner neurons, its working mechanism has not been clear yet. Then, in order to solve this problem, some further analyses are tried from the viewpoint of behavior in the state space. As a result, it is found that deviation of spatially-displaced patterns from the normal one is small enough.

View full abstract

The infrared optical source was shed on the fingertip and the intensity of the transmitted light or the reflection light was measured. The volumetric change of the bloodstream was estimated from these measurements. The acceleration plethysmogram wave data were obtained by the 2nd differentiation of the original measured waves. The results of the waves were classified by the Self-Organizing Maps (SOM) and then, the hardness of the vein was estimated. Further, the looseness of the wave position on the SOM map, the pulse period, the stability of the pulse wave were totally analyzed. Thus, the integrated analysis was successful from the acceleration plethysmogram wave data The display of the clinical examples was also successful related to the Acceleration Plethysmogram Analysis.

View full abstract

The on-line data collected at the time of operation of a factory was visualized. As for the usual pre-processing, it is ideal to perform normalization based on the amount of standards statistically from the information of the data-set. However, from a demand at the spot and statistical processing not being necessarily in agreement, the simple pre-processing was performed using the rule considered to be appropriate with it being commonsense, and the spherical self-organization map was used for realizing visualization.

View full abstract

An estimation method of physical and psychosocialstate from pulse wave is proposed, and a result of estimation is visualized in 2-D map. It is known that various features such as amplitudes of inflection points, heart rates, chaotic indexes can be detected. Some of these features depend on differences of individual or measurement environment, then it is needed to find the adequate features to improve the estimation ability. To realize this, we adopt Stochastic Neighbor Embedding and Multiple Relational Embedding.

View full abstract

The purpose of the paper is to verify whether or not the following conjecture is valid: it is necessary for some respondents to change colors for three-period sequences of warm-colored and cool-colored facades until the second time period. The examination suggests that the conjecture is valid to warm-colored facades if a small change is made to brightness and saturation. In order to make a suitable change to colors, an evaluation function is required to be estimated at each time period because the represented preference orders over stimuli of periods are in general different.

View full abstract

Assume that some impressions of landscape image are estimated by Choquet Integral of some color feature values using a two-additive Fuzzy measure. According to the least mean square approximation of some image impressions, which are quantified by questionnaires, we obtain a two-additive measure. Using this approach, we obtained some effects of combinations of two color features for image impressions.

View full abstract

In this paper we investigate the relation between gaze-parameters preference patterns when paired comparisons of fashion stimuli are conducted. First, subjects are divided into two cases according to gaze-parameters. In the experiment, the stimuli have set down 'hesitant case' and 'not hesitant case'. Second, the preference pattern of each case is studied by the conjoint analysis.

View full abstract

The present study considers a fuzzy color system in which three membership functions are constructed on the RGB color triangle. This system can process a fuzzy input (as the membership values of subjects) to an RGB system and output the center of gravity of three weights associated with respective grades. Three membership functions are applied to the RGB color triangle relationship. By treating three membership functions of redness, greenness, and blueness on the RGB color triangle, an average color value can be easily obtained as the center of gravity of the output fuzzy set. In the present paper, the differences among fuzzy input, inference output, and chromaticity are described, and the relationship between the centers of fuzzy inputs and inference outputs for fuzzy inputs are shown on the chromaticity diagram.

View full abstract

When considering viable colors for a design, such as a web page design, human designers may rely their own visual intelligence, cultural backgrounds, and other processes of human cognition which are a challenge to duplicate in intelligent systems. This paper proposes employing a neural network trained to recognize human-like qualities of color selection. The proposed system is validated by experiments in which a portion of a set of web page designs, including over 800 individual color selections, is used to train the network to recognize human selections versus random selections. The system shows the ability to learn non-specific qualities of color selection similar to that of a human designer.

View full abstract

We have proposed emotion color combination model using the quantification theory type I, and applied to the uniform color combination system. In this system, however, there was bias in the questionnaire data to relate image words of sport and emotions. And so, in this paper, we improved the questionnaire method, and constructed the image emotion model of sport with the quantification theory type II. A uniform color combination system was constructed using the emotion color combination model and the image emotion model. Also, to reflect the preference of the individual, the function to change tone of the displayed color combination was added. As a result of evaluation experiments, effectiveness of the proposed system was confirmed.

View full abstract

In this paper, we propose a concept of the virtual stylist system. This system supports men in their fashion coordination for suits, shirts and ties when they desire to find suitable shirt or tie designs for their suits. In order to find such clothing designs we should consider color harmony between colors of suit, shirt and tie we intend to put on. In the case of buying shirts or ties through the Internet, it is important for us to make sure of our appearances with them, because we can not try on them. So we propose the system which automatically searches clothing and enables to virtually try on it.

View full abstract

This paper proposes a method for quantitative evaluation of color harmony for men's fashion coordination support system. The system supports men in their fashion coordination for suits, shirts and ties when they desire to find suitable shirt or tie designs for their suits. In order to find such clothing designs we should consider color harmony between colors of suit, shirt and tie we intend to put on. This paper shows new approach to evaluate color harmony automatically based on the harmonious relationship between hue distribution type and tone distribution type.

View full abstract

PID control methods are still popular in industrial applications. However, due to their poor ability to react appropriately in a dynamic operational environment, they sometimes fail to control the actual plant, including operations such as feed oil changing and process line changing. On the other hand, well-experienced operators can easily manage these operations to appropriately manipulate the set points or to adjust the output of the PID controller using their experience and knowledge. In an ethylene plant, the decoking operation of the decomposition furnace is highly unstable. To mimic the well-experienced operator's procedures, we have developed a combined system that consists of an intelligent control and a PID control. The intelligent control system plays the role of the well-experienced operator. We applied three hierarchical systems to the different ethylene plants in order to demonstrate its effectiveness for the decoking process.

View full abstract

An expert has carried out the fuel cost mode test run. The test run carries out so frequent that a load of the expert is large. Then if we try to automate the test run using conventional control, it is very difficult to carry out the test run like the expert. In this paper, therefore, we applied an intellectual control to the fuel consumption examination, and evaluated the effectiveness.

View full abstract

We introduce to the summary of IPCC (Intergovernmental Panel on Climate Change) 's fourth assessment report (AR4). In this report, they said that "Warming of the climate system is unequivocal. There is very high confidence that the net effect of human activities since 1750 has been one of warming". It is a matter of life and death for us. We expect a speedy solution of the Global warming problem. For this problem, it is important things the reduction of greenhouse gases.

View full abstract

This paper reports on the new system called ParkLotD for detecting vacancy/occupancy in parking lots. ParkLotD uses a classifier based on fuzzy c-means (FCM) clustering and hyper-parameter tuning by particle swarm optimization (PSO). ParkLotD demonstrates high detection performances and enables the camera-based system to apply to outdoor parking lots.

View full abstract

In a petroleum refining plant with a large number of high-pressure facilities, high-pressure gas leaks resulting from equipment failures can engender disasters. By electro-chemical gas leak detectors, however, monitoring of gas concentration can identify leaks only long after they have accrued, sometimes 10 minutes or more. For immediate detection of high-pressure gas leak, we have developed a gas leak detection system based on the acoustic diagnosis. This paper shows; (1) it is examined chaotic behavior of background noise and leak sounds with various mixing ratios, (2) the CIC and the method of gas leak detection is described, (3) the actual results of our proof experiment for gas leak detection in Idemitsu Chiba Refinery, (4) it result determinism of residual data obtained by inverse filter is estimated using the CIC, and (5) the usefulness of CIC estimated the row data and residual data is discussed.

View full abstract

Needless to say, failure analysis and maintenance (Kaizen) are important in any industry field. However, it requires great deal of experience and long time for operators to analyze the failures efficiently and effectively. This is the reasons why the reports could not be used as knowledge for problem resolution even though the troubleshooting reports have been stored sufficiently. In this article, cause-effect structure is employed as knowledge representation. The structure is acquired from the text documents of the trouble shooting reports. Experiments were conducted to evaluate the method.

View full abstract

The binarization of document image plays an important role in many applications such as character recognition, computerization of the book which became old. However, it is difficult in degraded document image to distinguish the background and the character. In this paper, we propose the method that can do binarization well in such degraded document image. First, we use a median filter to get a rough background image. The image from which the background was removed is obtained by taking the difference of the background image and source image. Furthermore, the feature in the character part is enlarged by linear converting the difference image in the local, and clear sentence image to remove the deterioration part by making the image two values is obtained. Finally, we show experimental results of the proposal method and the four existing methods, and show the effectiveness.

View full abstract

This paper proposes a Bayesian super-resolution method for reconstructing a high resolution image from multiple lower resolution ones. The main contribution is to introduce an automatic method for determining regularization parameters for balancing a prior distribution and a likelihood function in the Bayesian model. The regularization parameters are determined by minimizing the Akaike Bayesian information criterion (ABIC). In order to calculate the ABIC, a second-order approximation is used to evaluate the multiple integral of the posterior distribution, because it is difficult to analytically evaluate the integral. Experimental results show the effectiveness of the proposed super-resolution method.

View full abstract

Decorative characters are widely used in various scenes. Practical optical character reader is required to deal with not only commonfonts but also complex designed fonts. However, since the appearances of decorated characters are complicated, most general character recognition systems cannot give good performances on decorated characters. In this paper, an algorithm that can extract character's essential structure from a decorative character is proposed. This algorithm is applied in preprocessing of character recognition. Experimentalresults show character skeletons are clearly extracted from very complex decorative characters.

View full abstract

We propose a method which improves the contrast of undiscriminatable colors of the input image by the hue and chroma modification of the image. Through the experiment, it is confirmed that the effectiveness of the proposed method.

View full abstract

In order to understand the trend of social bookmarking with respect to the huge number of web pages in the same category, "Bookmark Visualizer," is proposed. The Bookmark Visualizer merges the three dimensional data sets of huge number of web pages to suitable size by using hierarchical clustering, and plot the obtained data as the single graph. In experiments using social bookmarking data extracted from eight categories on HatenaBookmark (May 2nd, 2009), it is confirmed the effectiveness of the proposed system.

View full abstract

To analyze agents' decision-making process or rules about their actions, a long period of time is required for researchers to observe the target agents' actions. For example, in order to use such agents' actions as training data for artificial agents, numerical data and its linguistic expressions are needed. In this research, first, neural network agents are trained using numerical data from the log of RoboCup. Next, decision-making rules are extracted from the input-output relations of the trained neural network agents. Experimental results show the effectiveness of the proposed technique.

View full abstract

In this paper, we examined methods to classify the texts by using internet search engine's results of "and search" between keyword and category. We setup two functions which determines the keywords classification. First function simply uses the number of "and search" results. Second function uses the number of "and search" results as a numerator and uses the number of category's search results as a denominator. They both classify the keyword to the highest valued category. For the text classification, first we use these functions to classify the keywords which are in the text, then classify the text to the most classified category of the keywords. We also sort the keywords by its frequency or TF-IDF and use top 10,30, or 50\% of the keywords to classify the text. We compared precision of all of the combination of the text classification. We used internet news for test data set and the result of the test shows a future possibilities to be used in text classification.

View full abstract

Although optimization of vehicle routing problems(VRPs) is indispensable for cost reduction, experts need to decide it under very short time restrictions. Therefore, in this paper, a VRPs optimization system which can satisfy the demand is developed using the evolutionary algorithms on a small parallel computer. Results of numerical experiments show the effectiveness of our proposed optimization system for some real VRPs.

View full abstract

Currently, the domain of scientific research covers a wide range, from traditional sciences to new compound areas, and vast amounts of scientific knowledge are produced rapidly. Therefore, it is difficult to comprehend the whole structure of science from a large number of papers. In this research, we analyzed grant application data to learn about the structure of science and governing mechanisms which guides the time series behavior of the collaborative research field.

View full abstract

We have been developing The Researcher Name Resolver (alpha) that provides a web resource linking service of Japanese researchers registered in KAKEN (Grants-in-aid for Scientific Research under the Ministry of Education, Sports, Culture, Science and Technology). This linking service requires linking technologies for researcher name identification, tackling for the same family name / last name problem and different Chinese characters problem. We analyzed statistically the nature of the same family name / last name researchers in between KAKEN database and researcher databases of 34 universities. Then, we defined an identification method build in our system and conducted a sampling survey. This report describes the research results.

View full abstract

As an application for data mining from a large scale database manually built, we have constructed and analyzed academic researcher networks. We have used Kakenhi database as a master database for this purpose. And then we have picked two dataset as a first target: One is researchers from members of three academic societies in Statistics related field; and the other is from members of National Institute for Materials Science. An analysis and comparison on these types of researcher networks shows their different characteristics.

View full abstract

The present research has two objectives. The first objective is to introduce a method of data preprocessing particularly applicable to large-scale bibliographic databases in order to reduce the number of misidentified entries. Our method utilizes authors as points of reference in order to link error-prone affiliation IDs with theoretically error-free IDs by using carefully constructed inclusive and exclusive search terms. The result showed that our method effectively reduced the number of improper affiliation IDs. The second objective is to explore research productivities in various Japanese research organizations using a corrected database. Our preprocessing and article-classification methods offer fine and representative descriptions of the productivities of research organizations in Japan.

View full abstract

We study on research network formation including university-industry research collaboration based on Citation database for Japanese Papers (CJP), produced by National Institute of Informatics. The organization name identification is the most important pre-process in this research. There are 4.4 million unnormalized records of authors information in this database. In preceding research, the organization names are filterd out using the techniqus with stop words etc.. And we have identified 86% of organization records using the master table produced with visual check. In this study, we discuss about the available condition and problem for using Levenshtein distance in the name identification.

View full abstract

Statistical data analysis using legacy databases often requires grouping of mentions that refer to the same real world entity. This type of pre-processing becomes particularly important when dealing with large-scale databases since there exist much variation of names that makes the cost for generating dictionaries or normalization rules infeasible high. Based on this, we investigate, in this paper, methods for automatic name matching and discuss the advantages and disadvantages of (i) a binary classifier which determines whether two mentions refer to the same entity or not and also (ii) a graph-based clustering algorithm which disambiguates two similar mentions using their global features.

View full abstract

It is useful to transform continuous data to descrete one for condensing data amaount if information loss is small. In multivariate analysis like PCA correlation between variables plays an important roll. Therefore the transformation continuous to discrete must be one that keeps relationship between variables. In this paper effects of the transformation from continuous to discrete varables and reciprocal transfromation in multivariate analysis will be discussed. PCA and Hayashi's quantification method will be shown and compared by data before and after transformation.

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 extended to linear fuzzy clustering techniques by estimating linear prototypes spanned by multiple 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. In this research, the clustering method is interpreted as a multi-cluster-type multi-dimensional scaling for summarizing data, and a cluster validation approach is considered.

View full abstract

Text document classification is a fundamental technique for text analysis such as e-mail filtering and patent retrieval tasks. In this research, fuzzy PCA-based robust k-Means is applied to extraction of document clusters so that each cluster core includes mutually related documents ignoring the effect of noise documents. After quantification of documents by calculating tf-idf weights of frequently used words, fuzzy PCA is performed for constructing connectivity matrix composed of connectivity degrees among documents, and then, cluster structures are intuitively recognized by re-ordering the documents considering the responsibility of documents (degree of non-noise level).

View full abstract

FCM-type linear fuzzy clustering is also useful for local PCA because the clustering algorithm partitions data sets by calculating linear prototypes that can be identified with local principal sub-spaces. However, FCM-type algorithms often suffer from the initialization problem where we have multiple results with different initializations, and we must also pre-define the cluster number. In this research, several validation indices are compared in linear fuzzy clustering tasks and then applied to pareto optimality analysis considering both of minimization of objective function and maximization of cluster validity.

View full abstract

A technique for analytically estimating a relaxed solution for k-means clustering was proposed based on a PCA-guided manner. In the technique, however, the derived cluster indicator is a rotated solution, in which the rotation matrix cannot be explicitly estimated. Then, such an approach as visualization by ordering of samples in connectivity matrices is applied for visually access the cluster structures. This paper introduces a technique estimating rotation matrix by Procrustean transformation of principal component scores in order to select the optimal solution from multiple solutions derived by k-Means, and proposes a cluster validation method considering the deviation between k-Means solution and re-constructed membership indicator matrix.

View full abstract

Semi-supervised Clustering with two types of additional functions are considerd and compared using typical numerical examples.

View full abstract