The research focuses on approach for vision system. We propose the selective visual attention system which realizes high-speed processing flexible to environmental change. In order to obtain visual attention point, proposed system extracts three kinds of visual features; still image feature, blinking feature and motion feature. Proposed system introduced intention maps to attend interested area. Intention maps are updated dynamically to represent the spatial concepts which express the remarkable regions for each visual feature according to recognition purpose. The value on the intention map represents the gain of corresponding visual feature. Using intention map, proposed system can select visual attention points adapting to surroundings condition and tasks. As the result of experiments, the proposed system can confirm visual transition to the interest object.

View full abstract

We introduce a making method of "Edge image data like rough sketch" for the smooth outline tracing using fuzzy reasoning and examine it. This method was based on the sketch technique that man draws the rough sketch. At first, the edge of the car is detected from the input image using Sobel operator. Next, the edge is divided in the section where curvature is nearly. Finally, the edge was drawn many short and straight approximation lines by the least square method. Thus image data with values of degree that seems to be outline is made. In experimental results, most output images were nearly equal to the outline of input images.

View full abstract

In this paper, the digital images of cracks are classified into different damage levels based on the extracted characteristics through the Non-Linear SVM system.

View full abstract

In this paper, we propose fuzzy one-pair comparison values in place of one-pair comparison values used with Analytic Hierarchy Process(AHP). And we compute the weight of the evaluation item and alternatives by used fuzzy one-pair comparison values in AHP on experiments. This fuzzy one-pair comparison values were proposed by our paper presented in FSS2004. These membership function was presumed by the experiment that compared objects. Therefore, we use these fuzzy one-pair comparison values in AHP of new qualitative experiment. And the adjustment level and the comprehensive evaluation are compared. Moreover, error margins with a true value are compared. It will be clear that our method is useful.

View full abstract

This paper is an application of fuzzy theory to an analysis of appearances of houses. Four houses are investigated with a questionnaire, and the questionnaire has seven questions on houses. We analyze the degrees of agreements using the results from the sensitivity analysis by means of Lamda-fuzzy measure and Choquet integral, and then evaluate appearances of the houses synthetically. The most preference of items in the questionnaire is choiced using the degrees of agreements, finally.

View full abstract

We propose a method to calculate the attribute's intensity of an object from multiple opinions. The intensity is calculated by min-max inference and the membership value of each degree group is obtained from the rate of collected opinions. The rate is calculated considering to the reliability of each opinion which are extracted from evaluative sentences on the WWW. Extracting the opinions from sentences, classifying the opinions into three degree groups, and applying the reliabilities of the opinions are done based on the grammatical features. Furthermore, in order to deal with such degree expression and reliability expression, we define a quintuplet dataset which consists of "evaluative subject," "focused attribute," "orientation expression," "degree expression," and "reliability expression."

View full abstract

AHP is one of major methods for a domain of decision making. However indeed, a comparison matrix does not always have good consistency to enough rely upon the data. In these cases, we suppose that answers from a decision-maker have ambiguous or fuzziness and then weights also have that kind of property. Therefore, it is natural to represent the weights by use of fuzzy sets. We define a kind of representation of weights by using results from the sensitivity analysis. And also propose an approach to composition the weights. At last the study shows how weights of alternatives include fuzziness when a comparison matrix has relatively not so good consistency.

View full abstract

Classification procedures using normal populations predominate in statistical practice, since they have simplicity and reasonably high efficiency across a wide variety of population models. Although computing the posterior probabilities of the populations of each class after observing an object's feature vector is frequently useful for the purposes of identifying the less clear-cut assignment, the assignments to outlying observations tend to be near zero or one. A broad class of membership functions can be used in fuzzy c-means (FCM) clustering from the viewpoint of iteratively reweighted least-squares (IRLS) techniques. This paper clarifies the clustering characteristics of regular FCM, entropy regularized FCM and the proposed IRLS approaches. A new membership function and the IRLS approach enhance the classification procedure by truly less clear-cut assignments of memberships to the outlying observations.

View full abstract

This paper describes algorithms for term clustering using the concept of a fuzzy neighborhood. Similarity measures between two terms in a document set and the center of cluster are considered. A nonhierarchical method of c-means using the concept of the fuzzy neighborhood is proposed. Effectiveness of this method is shown by numerical examples.

View full abstract

Clustering by competitive learning is one of the most popular techniques in cluster analysis. However, it is well-known that this method cannot produce nonlinear cluster boundaries. To obtain nonlinear cluster boundaries, the use of the kernel method can be considered. This paper aims at expanding the competitive learning clustering to nonlinear clustering algorithm to handle nonlinear data set by using the kernel method. Moreover document information is modeled by using fuzzy multisets and the effectiveness of the proposed method is shown in numerical examples.

View full abstract

This paper aims at discussing at the on-line learning based LVQ using akernel function in support vector machines. Furthermore, its extension with fuzzy membership functions is considered.Numerical examples are compared to other kernel-based clustering algorithms and effects of proposed algorithms are discussed.

View full abstract

This paper proposes a simultaneous application of homogeneity analysis and fuzzy clustering which simultaneously partitions individuals and items in categorical multivariate data sets. In the proposed method, the graded possibilistic approach is applied to estimation of memberships of items for deriving the absolute responsibilities of them. The memberships can be regarded as the probability that an experimental outcome coincides with one of mutually independent events. Then, soft transition of memberships from probabilistic to possibilistic constraint is performed by using the graded possibilistic constraint in the approach.

View full abstract

Fuzzy c-Varieties (FCV) clustering is a linear fuzzy clustering algorithm that partitions a data set into several linear clusters using linear varieties as prototypes of clusters. Although the goal of the FCV clustering is unsupervised classification of non-labeled data sets, it can be regarded as a simultaneous approach to principal component analysis (PCA) and fuzzy clustering since the basis vectors of prototypical linear varieties are often identified with local principal component vectors. This paper reviews four PCA models and discusses the connections between linear fuzzy clustering algorithm and local PCA models based on different concepts. While the standard FCV algorithm is the modified version of the PCA model based on fitting low-dimensional sub-space, the same clustering result can be derived by considering estimation of latent variables that keep the original information as well as possible. In this sense, linear fuzzy clustering is a new approach to local multivariate analysis, in which fuzzy partitioning plays a role for classification of samples. It is expected that the new concept for local multivariate analysis stimulates future studies with similar ideas.

View full abstract

Knowledge discovery in databases (KDD) or data mining are the fundamental issues in many application fields and the task consist of two major processes: classification of data and analysis of correlation structure. Recently, several data mining tools, which can be regarded as the hybrid of clustering techniques and multivariate data analysis, have been proposed. In the non-linear approaches, simple linear models are used in conjunction with some suitable clustering algorithms. Fuzzy c-varieties (FCV) is a linear fuzzy clustering technique that captures the local linear structures of data sets, and is often identified with a technique for local principal component analysis because the vectors spanning prototypes form the orthonormal basis of principal subspaces. It is, however, difficult to define the clustering criterion when data sets include not only numerical variables but also nominal variables.In this paper, we propose a clustering technique that performs the FCV clustering of data sets including categorical data. The proposed algorithm iterates quantification of categorical data in the FCV clustering process so that quantified scores suit the FCV clustering. Because the quantified category scores are effectively assigned considering the relationship among categories, they are useful for interpreting the cluster structure.

View full abstract

This article presents two new clustering algorithms which are based on the entropy regularized fuzzy c-means and can treat data with some errors. The first, the tolerance which means the permissible range of the error is introduced into optimization problems which relate with clustering, and the tolerance is formulated. The next, the problems are solved using Kuhn-Tucker conditions. The last, the algorithms are constructed based on the results of solving the problems.

View full abstract

The kernel functions are useful into the field of classification and give the value of the inner product of two vectors in a high dimensional feature space by a mapping which one-to-one corresponds to the kernel function.Kernel hierarchical clustering is a technique of clustering by mapping the data of pattern space to feature space using a kernel function. Kernel hierarchical clustering can obtain a good result, but it is unknown how the data of pattern space is mapped in the feature space. This paper will propose how to visualize the distribution of the data in the feature space by using the mapping which maps the data in the feature space into the lower dimensional space with supporting the structure of clusters in the feature space. Moreover, the paper will discuss the relation with the visualization results and the clustering results of kernel hierarchical clustering.

View full abstract

Sequential pattern mining methods efficiently discover all frequent sequential patterns by using apriori property. However, analysts are not always interested in frequent patterns, because the patterns are common and the analysts cannot get new knowledge from the patterns. The paper proposes a new criterion which discovers interesting patterns for the analysts. The paper shows that the criterion satisfies the apriori property and how the criterion is related to existing criteria: support and confidence. Also, the paper proposes an efficient sequential pattern mining method based on the proposed criterion. Moreover, the paper shows its effect by applying it to daily business reports given from a sales force automation system.

View full abstract

This paper proposes an image search interface employing a directory structure to limit the search in an image data base by considering color and composition. In the proposed interface, an image is divided into blocks, in each of which the color difference between images is compared. By changing the block size depending on the number of the images to which clustering is applied, the interface makes image clusters that constitute a hierarchical directory structure. An abstract image, which is made based on the average color of each block, is expected to be a clue for selecting an appropriate directory. A user can limit the search for images from the directory tree based on color and composition of the abstract image.

View full abstract

In this paper, we propose a method to project products' review pages onto two-dimensional space by counting the number of evaluating expressions and the number of pictures contained in the page. We apply our method to review pages for mobile phones, and show the effectiveness of the proposed method by experimentation with human subjects. The experiments using a crossover design show the effectiveness of using the number of pictures in our products' review page projection method.

View full abstract

Web contents are not mere data but should be regarded as a consequence of user's behaviour which takes part in creating the contents. To extract useful knoledge from such contents, not only how to manipulate them (i.e., Web Intelligence) but also how to interact with them (i.e., Web Interaction) should be considered. Intelligent Web Interaction is a methodology to meet the requirement by taking both aspects into account. Upon this point of view, this paper discusses how soft computing technologies relate/contribute to the Intelligent Web Interaction.

View full abstract

In this paper, we proposed improvement method that automatically classifies Web pages retrieved by a search engine. In addition to apply the vector space model method, proposed method has index word selection by fuzzy reasoning, group name creation by the frequency of index words co-occurrence, and index page selection group name appearance. In experiments, retrieval result using a double meanings key word was classified by using proposed method.

View full abstract

Recently documents of various specific fields exist on the Web, which are updated frequently. As a result, there exist many fields-specific new words that are not listed in dictionaries. When a document including fields-specific new words is processed with computers for the purposes of indexing and information extraction, the treatment of new words becomes a problem. This paper proposes a method for extracting new words from category names in a large-scale Web dictionary service. The method is based on several characteristics of a word, such as the number of hits in Google search, the number of categories containing the word in the directory service, and the part-of-speech pattern. The experimental results show their effectiveness for extracting new words.

View full abstract

This paper considers the influence given by the importance of keywords to the arrangement pattern by the user. We conduct the experiments in which subjects are asked to arranges the keywords on the same map, when the keywords are classified into several groups in terms of their importance. From the results, the relative position between keywords with different importance is investigated. It is confirmed that there exist typical arrangement patterns in terms of the user's intentions. The result will be used to extract the user's intention from the keyword map for realiziy interactive information retrieval.

View full abstract

This paper describes the development of an intelligent corpus system for English composition support. When a non-English speaker composing an English sentence uses a web search engine in order to refer to good English sentence samples, finding appropriate English sentences out of numerous search results that include low-quality sentences is a troublesome task. The corpus system proposed in this paper infers the quality of each English document in a similar manner that humans do, and it provides the information of the quality to corpus users. This enables users to find good English sentence samples with less time and effort.

View full abstract

This paper presents a mutation-based evolutionary algorithm that evolves genotypic genes for regulating developmental timing of phenotypic values. The genotype sequentially generates a given number of entire phenotypes and then finishes its life at each generation. Each genotypic gene represents a cycle time of changing probability to determine its corresponding phenotypic value in a life span of the genotype. This cycle time can be considered to be a sort of information on developmental timing. Furthermore, the algorithm has a learning mechanism for genotypic genes representing a long cycle time to change the probability more adaptively than those representing a short cycle time. Therefore, it can be expected that the algorithm brings different evolution speed to each phenotypic value. The experimental results show that the algorithm can identify building blocks of uniformly-scaled problems sequentially and also that a population size required for solving the problems is quite small but the number of function evaluations required is sub-exponential scale-up with the problem size.

View full abstract

In this paper, we propose a novel approach to inference system with a DNA-based semantic knowledge representation model. DNA Computing-inspired Semantic Network (DCSN) is theoretically proposed and constructed with DNA molecules. It has the network structure of a graph formed by the set of all attribute-attribute values pairs contained in the set of represented objects, plus two tag nodes for each object. Each path in the network, from an initial tag node to a terminal tag node, represents the object named on the tag. To verify our theory, the preliminary experiment on implementing of inference system with a small model was successfully done by using very simple techniques, Parallel Overlap Assembly (POA) method, Polymerase Chain Reaction (PCR), and gel electrophoresis. The proposed model is very suitable for DNA-based knowledge representation in order to store vast amount of information with high density.

View full abstract

In this paper a new processing algorithm is proposed from an idea from DNA computing. Human brain is a typical information machine along with one of the points of views. The purpose of the human brain is to select information, and to acquire the algorithm of the selected information processing voluntarily. Everything cannot be expressed digitally though a part of the function until present gave an external input, and no acquired one. Then, passive was assumed to be an unconcious state, and active was assumed to be a state of consideration. Version Space Method of the Fuzzy theory to have treated the function of the processing of infromation on the brain was applied to DNA computing. It proposes an algorithm advanced from DNA computing.

View full abstract

To Study with e-learning that uses various learning materials, it is difficult for the learner to obtain required information from among them. In such a case, glossary as an index for the materials helps the learner search required information. However, existance of many materials makes it difficult to update and expand the glossary. To solve the problem, this paper proposes the method of automatic glossary generation based on meta data appended to the materials. In the method, meta data appended to the learning materials can assist the glossary generation.

View full abstract

The information retrieval using the metadata is paid attention to as a new, intelligent retrieval method. But it is difficult to give metadata to various information on the Web automatically. This paper proposes the system employing hybrid search engine which can make use of metadata while keeping recall with full-text search engine. This paper also proposes an adaptive retrieval method based on user's retrieval history.

View full abstract

In this experiment, we paid attention to the interlocution with friendly robot as an agent. We propose the method that makes the friendly robot study the Kansei expression using the study of mimicry. The chaos parallel evolution was used for the study of mimicry. Each Kansei expression of a friendly robot is made by the qualitative parallel evolution that is parallelization of functional level. Moreover, Kansei expressions that made by different users are acquired by the quantitative parallel evolution that is parallelization of role-sharing level. By constructing qualitative and quantitative parallelization with the same mechanism, we used the system that can adjust to both parallelizations. We propose a useful application for human support network system that uses acquired Kansei expressions. In particular, we conducted driver's support system for elderly driver and achieved good result.

View full abstract

In recent years we have various systems based on electronics, including cars, computers, mobile phones, and even robots, in our daily life. However, an unsymmetrical relationship exists between human and systems. There are various problems to be solved for balancing asymmetry of the relationship between human and systems. We named the electronics for solving these problems "Humatronics". Humatronics has two stages. The first stage is to establish the symmetric interaction between human and electronic systems. The systems have capability of understanding humans. The second stage is computer networks, which can share knowledge, informations, and experiences. Intelligent systems connected via networks will bring us boundless support throughout our daily life. We paid attention to the second stage of Humatronics and aimed to build human vehicle system that uses network intelligence for human support. Especially we build systems to display danger for drivers and platooning system use driver's intention (turn left, right and straight).

View full abstract

In this paper,we propose an IF-THEN- rule based inference method which is necessary to construct a natural language communication system and an expert system,and so on. The method is referred to how to estimate a truth qualifier tB1 when an input proposition is QA are F is t and IF-THEN- rule IF Q1A1 are F1 is tA THEN Q2A2 are F2 is tB is given and an inference result is Q2A2 are F2 is tB1(Q,Q1,Q2:Fuzzy quantifiers,A,A1,A2:Fuzzy subjects, F,F1,F2:Fuzzy predicates,t,tA,tB,tB1:Truth qualifiers). We propose a method which infers a result proposition Q2A2 are F2 is tB1 for monotone Q's and show concrete application examples of the method.

View full abstract

The aim of this paper is to propose a morphological error measure in order to automatically detect morphological errors such as part-of-speech tagging errors when analyzing. In particular, we focused on three similarities as the measure and experimented using them. From our experimental results, it was found that the precision of one of measures was over 70% and it functioned well.

View full abstract

The analysis is done by expert who analyzes data while combining various algorithms with the trial and error on text mining. In this paper, two or more text classification algorithms are combined by using the genetic programming, and it proposes the system that classifies the text. The tuning of the parameter of the algorithm at the same time constructing the best use of the feature of each algorithm by learning the combination by the genetic programming becomes possible. We discuss combined text mining with genetic programming for mail classification task.

View full abstract

We have many kinds of data of time series such as daily stock prices. We understand it via its natural language expression. We proposed a method to express trends of time series in a natural language. In this paper, we propose a concept of view, which has 3 components, target data, focused variables and terms in a natural language, to faciliate to manage knowledge for expressing the target data with the terms via the focused variables. We consider some combination methods of views to define a new view.

View full abstract

Recently, Evolutionary Algorithms (EAs), which are based on the biological evolutionary process, are widely studied in a lot of fields. However, their probabilistic behavior makes the theoretical investigation difficult. As a result, the search performances of EAs have been tested through simulations in most of the studies. The parameters of genetic operations have also been decided experientially or by trial and error. This paper tries to mathematically analyze the search performance of EAs using the parameters of the operations for theoretical investigation of them. This paper defines the number of searchable solutions (NSS) and the times of improved solutions (TIS) to indicate search performance of EAs and attempts to formulate them. This paper employs hill climbing method to a monotone increasing function as a basic study of this research, and it derives the formulas for these indicators, NSS and TIS, that consist of the parameters of the genetic operation. This paper shows that appropriate formulas were derived and they enable us to compare and optimize the performance of the parameters for the operations in EAs without actual trials.

View full abstract

In this paper, we examine the performance of genetic network programming (GNP) for learning agents on perceptual aliasing problems (PAPs). In order to cope with this problem, a genetic programming approach called Adaptive Genetic-Programming Automata has been already proposed. While it effectively tackled to PAPs, too many rules are generated that are not used to control the agent. Simulation results clearly show that the number of rules is reduced by GNP in a maze problem in which a learning agent tries to reach a goal.

View full abstract

In this paper, we propose an idea to identify the gene linkage in permutation problems where a chromosome is represented by a permutation of different alleles in genetic algorithms. We apply the proposed method to a standard local search and genetic local search. We show the effectiveness of the proposed method by computer simulations on flowshop scheduling problems.

View full abstract

We focuses on emotion recognition using prosodic features in speech.There are some differences in prosodic features among indivisuals.Therefore, the accuracy of emotion recognition goes down.This paper proposes evaluation method of feature selection.We apply KL-divergence to measure a distribution of prosodic features in emotional speech.It can quantify a effectiveness of individuals and of speaker's emotion on prosodic features.So, we can choose the most suitable features for emotion recognition system.Furthermore, we propose a method of feature selection system, using Genetic Algorithm(GA)making use of benchmark based on KL-divergence.We can make more effective features choices from a variety of options using this method.

View full abstract

Lots of Genetic Algorithms (GA), such as distributed or parallel methods, have been proposed. These GA methods classified into two types. One type is a coarse grained GA. The other type is a fine grained GA. The individuals of both methods are classified by crisp clustering. Therefore, we can see the GA method using fuzzy clustering as one of the new approaches for distributed or parallel GA methods. In this paper, we propose the GA method using fuzzy clustering. We also discuss about behaviors of our GA via experiments of a function optimization problem.

View full abstract

We examine three methods for improving the ability of evolutionary multiobjective optimization (EMO) algorithms to find a variety of fuzzy rule-based classification systems with different tradeoffs with respect to their accuracy and complexity. The accuracy of each fuzzy rule-based classification system is measured by the number of correctly classified training patterns while its complexity is measured by the number of fuzzy rules and the total number of antecedent conditions. One method for improving the search ability of EMO algorithms is to remove overlapping rule sets in the three-dimensional objective space. Another method is to choose similar rule sets as parents for crossover operations. The other method is to bias the selection probability of parents toward rule sets with high accuracy. The effectiveness of each method is examined through computational experiments on benchmark data sets.

View full abstract

In this study, we try to find a tipping-point mechanism in social diffusion phenomenon. Recently, various attempts of elucidating such a social diffusion phenomenon using the multi-agent system are being done. Fujii discovered the social diffusion phenomenon to be generated by the interaction of agents. Murata and Yagi paid their attention to the point for which the spread of the adoption action depended on initial placement method of agents. However, there was an uncertain point in the mechanism. Therefore, by computer simulations, we try to see conditions where the tipping-point can be found.

View full abstract

In this paper, we propose a method to reward a robot in Q-learning in order to follow a moving target. In the previous study, the robot moves toward the fixed target, but we try to follow the moving target in this paper. We modify a rewarding method in a GA-based Q-learning called "Q-learning with Dynamic Structuring of Exploration Space Based on Genetic Algorithm (QDSEGA)". Simulation results clearly show the effectiveness of the proposed method.

View full abstract

This paper proposes a learning algorithm of fuzzy if-then rules for pattern classification problems. In this paper we assume that each training pattern has a weight that can be viewed as the degree of importance in its classification. Fuzzy if-then rules are generated from the weighted training patterns. The antecedent part of fuzzy if-then rules involves fuzzy sets for each attribute value in pattern space. The class and the grade of certainty are used in the consequent part of the fuzzy if-then rules in this paper. The consequent class and the grade of certainty of fuzzy if-then rules are determined by a heuristics using given training patterns. The proposed method adjusts the grade of certainty in an error-correction manner. That is, when a training pattern is misclassified by the generated fuzzy if-then rules, the grade of certainty of the responsible fuzzy if-then rule for the misclassification is decreased while we increase that of the fuzzy rule that should correctly classify the training pattern. Through a series of computational experiments we show the effectiveness of the proposed method for several real-world pattern classification problems.

View full abstract

In this paper, we examine the performance of Genetic Network Programming (GNP) using Automatically Defined Groups (ADG) for a multi-agent control problem. GNP is a kind of evolutionary methods inspired from Genetic Programming (GP). We consider two types of problems in this paper: one problem is to assign an appropriate role to each agent according to its ability, and the other is to assign a proper role to each agent with the same ability. We show the effectiveness of GNP with ADG through computer simulations on the two types of load transportation problems.

View full abstract

There are a number of conventional studies for accelerating the convergence speed and reducing IEC user's fatigue through the learning of IEC user's evaluation characteristics. Since IEC users can evaluate a limited number of individuals, the number of evaluations is often not enough for learning the characteristics of user evaluation. To address this problem, this paper proposes a method in which the characteristics of multi-IEC users are used as a substitution for a single IEC user. The evaluation values of IEC users are observed in each generation to mimic subsequent evaluations. For the mimicking process, the distance between the characteristics of an individual IEC user and other users are calculated and the most similar users' characteristics are taken for evaluating many individuals. We evaluate the method using functions instead of IEC users. Distinct groups of functions, which have different intervals from the pseudo-IEC user, are used to mimic characteristics of the further users. The speed of IEC convergence for each group is then compared to the conventional method. The result shows that the proposed method can accelerate the convergence of IEC when the characteristics of other user evaluations are similar to the pseudo-IEC user.

View full abstract

This paper presents a method to aquire concepts of behavior modification rule.First, behaviors are represented in frequency domain by signal processing.Next, we introduce gain parameters for modulating each frequency band of behaviors.We can have various patterns of behaviors by adjusting gain parameters.Their parameters are systemized as concepts of behavior modification. Robots can generate various modified behaviors by using acquired concepts. So the variation of behaviors can be realized by modifying a few basic behaviors.In addition, other robot's modifed behaviors can recognized by acquired concepts.

View full abstract

This paper proposes a fuzzy rule-generation method from weighted training patterns. We assume that each training pattern has a weight that corresponds to its importance. The weight is treated as a cost of a misclassified/rejected pattern. The weights of the training patterns are used in the generation process of fuzzy if-then rules. We formulate the problem of constructing classification system as minimization of a total cost and an error rate. The proposed method handles training patterns with large weights more importantly than those with small weights. Our objective is to construct a fuzzy classification system that decreases a total cost and error rate. In the classification of an unseen pattern, we take the maximum value of the product of the compatibility and the grade of certainty. In computational experiments, we compare the conventional construction method of fuzzy classification system with the proposed method. We show that a total cost and an error rate are reduced by the proposed method. The performance of the proposed method is also compared with that of nearest neighbor classifiers.

View full abstract

We propose a fuzzy rule extraction method for behavior analysis of agents in an artificial futures market. Our goal is to linguistically explain the behavior of highly profitable agents using fuzzy If-Then rules. Fuzzy rules are extracted from a trading history of such an agent. The antecedent part of each fuzzy rule is specified by previous spot prices while its consequent part is one of the three alternative actions of agents:Buy, Sell, and No Action. That is, fuzzy rules explain how each agent determines its action based on previous spot prices in the futures market. In computational experiments, we extract fuzzy rules from a trading history of a software agent with an unknown strategy. It is demonstrated that the behavior of the agent is linguistically explained through the visualization of the extracted fuzzy rules. While the proposed method is applied to the software agent in our computational experiments, it is also applicable to behavior analysis of human agents.

View full abstract

This paper mentions knowledge acquisition for car driving by means of fuzzy neural networks and Q-learning. Acquired knowledge is expressed by linguistic terms in the if-then form. If-then form rules are simplified by the combination of some rules. The paper also performs simulation experiments using the proposed algorithm and discusses the control results and acquired rules.

View full abstract