Though the complex-valued self-organizing map (CSOM) is powerful in distinction between plastic landmines and other objects in landmine visualization systems, the distinction sometimes fails in a portable low-resolution system. In this paper, we propose two techniques to enhance the visualization ability. One is the utilization of SOM-space topology in the CSOM adaptive classification. The other is a novel feature extraction method paying attention to local correlation in the frequency domain. In experimental results, we find that these two techniques significantly improve the visualization performance. The localcorrelation method contributes also make the system free from choosing an optimal "base frequency".

View full abstract

Clustering is one of the most useful methods to understand similarity among data. However, most conventional clustering methods do not pay sufficient attention to geometric properties of data. Geometric algebra (GA) is a generalization of complex numbers and of quaternions, and it is able to describe spatial objects and relations between them. In this study we introduce GA to systematically extract geometric features from data. We propose a new clustering method by using various geometric features extracted with GA. We apply the proposed method to clarification of human impressions of a product. In the field of marketing, companies often carry out a questionnaire on consumers for grasping their impressions. Analyzing consumers through the obtained evaluation data enables us to know the tendency of the market and to find problems and/or to make hypotheses that are useful for the development of products. Finally, we discuss clustering results of a questionnaire with/without GA.

View full abstract

Many quantum computation algorithms utilizing quantum mechanical behavior such as superposition of quantum states have been proposed. Recently, Farhi et al. have proposed an adiabatic quantum computation (AQC), which can be applied to various problems including non-deterministic polynomial time problems if one can know an appropriate Hamiltonian for a target problem. We have proposed a neuromorphic adiabatic quantum computation (NAQC) as the AQC with energy dissipation and an efficient method for designing a final Hamiltonian in consideration of the analogy with a neural network. The NAQC can be applied to optimization problems if its cost function can be expressed in a quadratic form. And successful operations have been confirmed by numerical simulations. In addition, we have proposed a new learning method for the NAQC inspired by Hebb learning and have shown its successful results for the network with four quantum bits. In this paper, we show preliminary results for a quantum bit network with NAQC and the proposed learning by numerical simulations. And we discuss its capability as an associative memory.

View full abstract

This study proposes the multistate associative memory scheme of high-dimensional Hopfield-type neural networks based on the quaternion algebra, that is a class of hypercomplex numbers. The presented model is an extension of the complex-valued multistate neural network in which the state of a neuron is represented as one of the points on a unit circle, and the state of its neuron is represented as one of the polar coordinates on a three-dimensional unit hyper-sphere, thus expressed by three kinds of phase variables. The quaternionic signum function, the energy function, and the method for embedding patterns to the network are introduced, and the properties and stability of the network are explored, such as the monotonically decrease of the energy with respect to the change of the neuron state and the basins of the attractors around the local minima in the network. This extended multistate network will be appropriate for processing the color image patterns, because the three multistate phase variables in the quaternionic neuron state can be regarded as the tone levels of three basic colors, i.e., red, blue, and green.

View full abstract

We study the optimization of neural networks with Clifford geometric algebra versor and spinor nodes. For that purpose important multivector calculus results are introduced. Such nodes are generalizations of real, complex and quaternion spinor nodes. In particular we consider nodes that can learn all proper and improper Euclidean transformations with so-called conformal versors. Thus a single node can correctly compute full 3D screws and rotoinversions with off-origin axis and offorigin points of inversion. The latter is a unique property of our proposed versor neuron. Computing inversions by ordinary real-valued networks is not easily possible due to its nonlinear nature. Simulation on learning inversions illustrating these facts are provided in the paper.

View full abstract

This paper discusses the learning mechanisms of cognitive development for partner robots through interaction with people. First of all, we discuss the cognitive capabilities from various points of view. We proposed several learning methods for cognitive development of partner robots based on the relevance theory. Next, we explain a prediction-based perceptual system to learn the human behaviors and gestures related with the background environmental information. Next, we explain the associative learning method of the relationship among linguistic information and perceptual information to realize natural communication with people. Finally, we discuss the learning capabilities of cognitive development of partner robots through these several experimental results.

View full abstract

In neuro-fuzzy learning by on-line first-order backpropagation, fuzzy rules' interpretability may be lost during the training phase. This is a so-called interpretability-precision dilemma. We show that ``hidden-node teaching'' is a simple and effective remedy for this dilemma. To verify hidden-node teaching effects, we first analyzed learning results in two small-scale regression problems. Since the posed scheme is well suited to a practical large-scale setting, we then applied hidden-node teaching to a CANFIS neuro-fuzzy modular network with two local-expert multilayer perceptrons for attacking the letter recognition problem, a large-scale UCI machine learning benchmark. Although our current approach is still ``ad hoc'' (and thus more extensive investigation is required), our preliminary experiment shows encouraging results.

View full abstract

We here introduce a novel biologically inspired adaptive controller for autonomous robot. The proposed controller binds N number of Aplysia-like spiking neural network each of which could interact with particular sensory information and produce various motors output. The post-synaptic weights in each model are gradually updated by the property of spike timing-dependent plasticity (STDP) and that of the presynaptic modulation signal (synapse-on-synapse contact) from the sensory neurons. Information from different types of sensors is bound at the motor neurons. Experimental results show that a physical robot Khepera with the proposed controller quickly adapted into an open environment by evolving obstacle avoidance behavior while tracking a target object. We believe that this approach could be applicable to an autonomous robot with various sensory and motor modalities.

View full abstract

The paper presents a simple, efficient approach of formulating and optimizing the global Airline Crew Scheduling Problem (ACSP). Both the tasks of plane and crew assignment are addressed simultaneously using Discrete Particle Swarm Optimization (DPSO). Realistic constraints are conceptualized and incorporated. Simulation is performed for a simple test case and a more complex one. Importance of adopting explicit diversity preservation techniques, advantages of constraining the search space as well as the effects of DPSO on different aspects of ACSP are explored. Analysis and discussions provide a more holistic understanding of the complexity involved in ACSP and new insights into effective ways of improving the results. The overall findings indicate that DPSO is a potential candidate for solving more complex extensions of the ACSP.

View full abstract

This paper presents a new Particle Swarm Optimization (PSO) technique with velocity control. In PSO, when a particle finds a local optimal solution, all of the particles gather around the one, and cannot escape from it. In the proposed method, we lead the particles from intensification to diversification by adding random numbers to the velocity of the particles depending on the distance from gbest, and thereby the particles can search widely in the search space. Then, the proposed method is validated through numerical simulations with several functions which are well known as optimization benchmark problems comparing to the conventional PSO methods.

View full abstract

A hybrid optimization based on (μ,λ)-ES and particle swarm intelligence is proposed. In this new hybrid algorithm, individuals of each population are divided into two groups by fitness. The first (better) group is dealt with (μ,λ) mutation, and the second group is dealt with particle swarm intelligence. Experiments are done on a set of standard benchmark functions. In order to find out the performance of the hybrid, we test different population divide ratios. Experimental results show that hybrid optimization performs better than (μ,λ)-ES on all the benchmark functions and better than particle swarm optimization on part of the function set.

View full abstract

We propose a method utilizing a semi-supervised Laplacian eigenmap technique for transferring the colors of an image to another image,i.e. changing the color configuration in an image to that in another one. A user draws some marks in two images for specifying desired correspondence between colors in them. The colors in two images are then mapped into a common subspace where the colors are transferred by their nearest neighbor matching. We apply this method to a general purpose non-photorealistic rendering technique called the image analogies. We also extend the method to videos where the marks are drawn in the first frame from which the color is transferred to every successive frame.

View full abstract

This paper presents a new strategy to estimate pitches in mixed musical tones in which harmonic components of each tone overlap. Our system based on the temporal approach can estimate pitches with a high time resolusion by using comb filters, which elimainate corresopnding harmonic components, and auto-correlation functions, which reveals the periodicity of a harmonical signal. When a comb filter corresponding to one tone which is contained in the singal eliminates its harmonic compnents, the residue signal consists of harmonic components of another tone. Then another pich can be estimated by detecting a periodicity from auto-correlation functions of the residue signal. Therefore, this system means an attempt to find comb filters which eliminate one of each tone by observing a behavior of auto-correlation functions showing a single tone's one, and estimates pitches. Pich estimation results with this sytem for polyphony are presented and accuraacy is investigated.

View full abstract

In this research, we developed software that makes an image from music. We call it music visualization. Our music visualization is based on pitch detection, so it visualizes true musical notes, and this differs from simple tempo detection. We have two interests. One is whether our music visualization has some different effects from tempo detection's. Another one is whether a pattern of coloring of musical notes affects people subconsciously. So, we designed the software to be able to answer these questions. The sound has the direction (right and left) and formless. Therefore, it is necessary to define the shape of the sound to visualize it. We proposed the hue circle model and the Orb-model as a model of visualization. This Orb-model can handle an input signal of a stereo sound, and calculates the frequency spectrum of each right and left channel. A horizontal location of the screen is decided by the spectra ratio of the right and left channels. A vertical location is decided by the pitch of the sound, i.e. the high-pitched sound is displayed in the upper part and low-pitched sound is in the lower one. The size of the orb is decided by the size of the composite vector of the right and left channel spectra. The orb with the color corresponding to the pitch name is displayed. To calculate the frequency spectrum corresponding to each pitch name, we adopted the pitch discrimination algorithm by the synchronous addition and subtraction method. In order to examine how our music visualization affects people, we carried out two evaluation tests. Evaluation 1 is comparing our music visualization with tempo detection's. Evaluation 2 is a evaluation test of our music visualization on two different colorings. Evaluation 1 suggested that our music visualization has some unique characteristics. Evaluation 2 suggested that a pattern of coloring of musical notes does not have definite effect to people.

View full abstract

This paper addresses the problem of on-line Independent Component Analysis (ICA). This area has been explosively investigated, and nowadays, there have already been a great many algorithms. However, a serious problem is that most of the existing algorithms can only work as batch processing, since they usually include some ensemble expectation operations. In this paper, we propose an on-line ICA algorithm that is performed as recursive processing on a sequential finite lengthened input blocks, for learning the mixing parameters and for separating the sources based on a maximization of non-Gaussianity of the outputs. Numerical experiments show that the proposed algorithm can converge efficiently and can separate sources appropriately.

View full abstract

The Mental Image Directed Semantic Theory (MIDST) has been proposed an omnisensory mental image model and its description language Lmd. This language can provide multimedia expressions with intermediate semantic descriptions in predicate logic. This paper presents a brief sketch of Lmd and its application to cross-media operations between linguistic and pictorial expressions of space and time.

View full abstract

A fast image processing is necessary technique for many computer vision applications.In this paper, we study about fast processing based on a parallelization. There are some parallelization by using some CPU or PC clusters are proposed, but they have some problems about coding complexity and installation costs. In this paper, we utilize Graphics Processing Unit (GPU) that is generally used for PC. And we build computer vision library works on GPU. As a result of evaluation, our proposed library has been marked x16 faster than working on CPU in the best scene.

View full abstract

An extraction of a specific figure in image has basic problems in intelligent image sensing. The generalized Hough transform (GHT) is the representative method to extract arbitrary figures which are rotated and enlarged or reduced. It has been numerous improvements. However, for extraction of arbitrary figures, it takes much processing time and needs much memory space. It is impossible to apply to figures including intersections. For an improvement of the problems, a new method to extract arbitrary figure using one-dimensional histogram is proposed in this study. In comparison with conventional method, it is understood that memory space is very small, processing time is very short and figures including intersections can be extracted. In addition, this method is effective for the target figure with different aspect ratio.

View full abstract

It is often that we count the number of objects in various situations. In the case of several objects, we can recognize the number at a glance. On the other hand, in the case of a dozen of objects, the task to count the number might become troublesome. Thus, simple and easy way to enumerate the objects automatically has been expected. In this study, we propose a method to recognize the number of objects by image. In general, the target object to count varies according to user's request. In order to accept the user's various requests, a desired object in the taken picture is manually designated as a region, which is used as a template. Main process of the proposed method is to search and count regions which are resemble to the template. To achieve a robust searching for spatial transformation, such as translation, rotation, and scaling, scale-invariant feature transform (SIFT) is employed as a feature. To show the effectiveness, the proposed method is applied to some images taking ordinary things, e.g. binders, cans, etc.

View full abstract

This paper relates to a practical headline area extraction method employing a half-cosine function wavelet network. The half-cosine function wavelet network has multi-level structure. In each level, basis functions with same support width corresponding to a space frequency are arranged. Generally basis functions corresponding to a headline area are wide support width. Therefore, after decomposing an image to weighted basis functions, headline area is obtained by combining support areas of basis functions having high weights in the lower level which has the highest weight. In this paper, we demonstrate the adaptive headline area extractions from Japanese and English documents and video in which the font size of headline changes dynamically. Extraction speed more than 10 fps for video images is achieved. The validity of the proposed method is discussed through experimental results.

View full abstract

3D pose recognition is important as the intelligent function of autonomous robot. The pose recognition is achieved by two stages called an initial registration and a detailed registration. In the initial registration, a corresponding between the model and the scene is made roughly. On the other hand, in the detailed registration a highly accurate corresponding is made. In this paper two new invariant features called HSI_S (Horizontal Spin Image of Surface) and HSI_E (Horizontal Spin Image of Edge) are proposed, and parallel use of them is shown experimentally to improve generality of pose recognition. In addition to that, two different ICP algorithms called F-ICP and I-ICP are also proposed and parallel execution of these two ICP's is shown experimentally to improve robustness to clutter and occlusion. The reason of that is that since focus regions of F-ICP and I-ICP are mutually complementary, the parallel execution of them (hybrid method) leads to the accuracy improvement. In this paper, several non-hybrid methods are compared to the hybrid method through pose recognition (registration) experiment of objects like auto parts and bird models. As a result, the proposed hybrid method with HSI_S and HIS_E has been shown to be better than single method (non-hybrid method) in its performance. That is, the proposed hybrids method is dominant in both generality and robustness.

View full abstract

This work concerns practical issues surrounding the application of learning and memory in a robot toward optimal navigation in dynamic environments. A novel hierarchical adaptive controller that contains two-level units was developed and trained in a physical mobile robot e-Puck. In the low-level unit, the robot holds N numbers of biologically inspired Aplysia-like spiking neural networks that have the property of spike time-dependent plasticity. Each of these networks is trained to become an expert in a particular local environment(s). All the trained networks are stored in a tree-type memory structure in the high-level unit. These stored networks are used as experiences for the robot to enhance its navigation ability in both new and previously trained environments. Forgetting and dynamic clustering techniques are also developed to control the memory size. Experimental results show that the proposed model can produce a robot with learning and memorizing capabilities that enable it to survive in complex and highly dynamic environments.

View full abstract

Adapting a real mobile robot in complex and changeable environments is one of the known challenging tasks in robotic researches. Complex environments are usually a combination of patterns, each of which requires a unique behavior from the robot to deal with. It is very difficult to develop a single network that can cope with such complex environments using the commonly known evolutionary algorithms. The network would be confused between adapting and readapting into each pattern and an optimal network would never occur. In this short paper, we proposed a simple structured adaptive controller with learning and memorizing ability that can cope with such complexity. The proposed controller works to simplify the complex environment into simple patterns that each of which could be independently trained. A memorizing mechanism is introduced to the controller to enhance the robot ability in tracking its own experiences and use it to cope with the upcoming events. Experiment results show that the proposed controller helps the stability of the robot performance in a complex environment

View full abstract

The indirect adaptive regulation of unknown nonlinear dynamical systems under the presence of dynamic and parameter uncertainties, is considered in this paper. The method is based on a new Neuro-Fuzzy Dynamical Systems definition, which uses the concept of Fuzzy Dynamical Systems (FDS) operating in conjunction with High Order Neural Network Functions (F-HONNFs). Since the plant is considered unknown, we first propose its approximation by a special form of a fuzzy dynamical system (FDS) and in the sequel the fuzzy rules are approximated by appropriate HONNFs. Thus the identification scheme leads up to a Recurrent High Order Neural Network, which however takes into account the fuzzy output partitions of the initial FDS. The proposed scheme does not require a-priori experts' information on the number and type of input variable membership functions making it less vulnerable to initial design assumptions. Once the system is identified around an operation point, it is regulated to zero adaptively. Weight updating laws for the involved HONNFs are provided, which guarantee that under the presence of 'small' dynamic uncertainties both the identification error and the system states reach zero exponentially fast, while keeping all signals in the closed loop bounded. The existence of the control signal is always assured by introducing a method of parameter hopping, which is incorporated in the weight updating law. The applicability is tested on a DC Motor system where it is shown that by following the proposed procedure one can obtain asymptotic regulation quite well in the presence of unmodeled dynamics with small values.

View full abstract

Neural network ensemble (NNE) is a collection of a finite number of networks that are trained for a task to achieve better performance. A number of NNE methods are proposed and is shown effective over single network. The proposed methods apply different techniques for NNE construction and a method is shown effective for a subset of problems. In another word, no method is found that is superior to others for all the problems. Keeping it in front, in this study we investigated an approach to construct an NNE that might effective for all the problems. For NNE construction first we trained sets of networks with different NNE methods and then select networks for final NNE. The proposed technique has been tested extensively on several benchmark problems of machine learning and neural networks. Experimental results show that this method able to construct concise NNE and outperformed other individual methods.

View full abstract

In this paper, we discuss the influence of a discrete time computation in SpikeProp, which is a kind of training algorithm for spiking neural networks. It adjusts timing of spikes according to the gradient of their error. Digital computation, especially computation in discrete time, causes some quantization errors. We discuss the influence of the quantization error based on the shape of error surfaces, which represent error depending on parameters. Through some experiments, we show learning processes degraded by a digital computation, and a typical shape of error surfaces that cause the degradation. Digital computation brings rough error surfaces, which have many false local minima. These local minima block the effective acceleration brought by sophisticated optimization algorithms.

View full abstract

This paper proposes an improvement method of classification efficiency in a classification model. The model is used in a risk search system and extracts specific labels from articles posted at bulletin board sites. The system can analyze the important discussions composed of the articles. The improvement method introduces ensemble learning methods that use multiple classification models. Also, it introduces expressions related to the specific labels into generation of word vectors. The paper applies the improvement method to articles collected from three bulletin board sites selected by users and verifies the effectiveness of the improvement method.

View full abstract

Document map construction is a useful approach to intuitive text mining, in which mutual relations among text documents composed of many keywords are characterized in a 2-D map. Usually, text documents are first preprocessed into numerical weights such as tf-idf weights by considering term frequency and inverse document frequency, and then, dimension reduction techniques, such as principal component analysis (PCA), are performed for constructing low dimensional plots of multivariate data. This paper considers using a linear fuzzy clustering-based variable selection mechanism for selecting keywords that are useful for characterizing documents, in conjunction with applying document clustering for extracting multiple linear sub-structures. In the approach, meaningful keywords are selected in each cluster (linear sub-structure) and mutual relations among documents are represented in simple linear sub-spaces.

View full abstract

In this study, we compare several variations of Fuzzy Clustering for Cooccurrence Matrix (FCCM) in applications to text analysis. The FCCM was proposed to partition individuals and items of the cooccurrence matrix by maximizing the degree of aggregation of each cluster. The total amount of products of cooccurrence variables and memberships for individuals and items is regarded as the degree of aggregation. Several variations of FCCM which employ two types of constraints for memberships i.e. probabilistic and possibilistic and two types of regularizations to obtain fuzzy clusters, entropy maximization and K-L information, exist. In the experiments, we apply our methods to a data set which represents frequency of keywords appearing in text documents and compare the results of each clustering method. They are used to find mutual relation (or co-occurrence structure) among text documents and keywords in the applications. Those tasks are known as text mining.

View full abstract

Many studies have been performed on query expansion recently. We propose a query expansion system using conceptual fuzzy sets (CFSs) based on context. CFSs support an meaning expression by changing fuzzy sets within a context, and we use fuzzy clustering for conceptualizing fuzzy sets. We performed an experiment comparing our system with pseudo relevance feedback, which is usually used in information retrieval, and our system was found to be more precise than pseudo relevance feedback. Additionally, the results from this experiment show that our system has the possibility of being more precise by combining conceptual fuzzy sets and pseudo relevance feedback.

View full abstract

The amount of image data has recently been rapidly increasing because of the spread of digital cameras and the increasing capacity of data storage. Users' needs to manage these images have therefore been growing explosively. This paper proposes a method of automatically annotating images that automatically attaches keywords that express these images based on the recognition of blobs within them. The system selects the most appropriate keywords from ones attached to similar images applying context-based image retrieval. We demonstrated the effectiveness of the new method by evaluating the accuracy of a system for retrieving natural images employing the proposed system of annotation.

View full abstract

In a business, it is an important task to evaluate efficiency between each branches or each industries. In evaluation of efficiency, DEA (Data Envelopment Analysis) has been proposed. In particular, CCR Model and BCC Model are frequently used. However it may cause confusion owing to differences in method of evaluation. First, we introduce a notion of fuzzy production possibility area and crisp production possibility area. Then, we construct a fuzzy goal from each evaluation. We also fuse CCR Model and BCC Model and then construct a fuzzy DEA model. Furthermore, we apply the proposed fuzzy DEA model to the problem of major banks. Then, we are going to show its effectiveness and usefulness.

View full abstract

In order to deal with multiobjective programming problems, the concept of domination structures based on convex cones were introduced, which can be regarded as a generalization of Pareto optimal concept. Since domination structures are deeply related to the decision maker's preference in objective space, it seems to be very difficult for the decision maker to supply precise information that makes it possible to find a sharp borderline of a domination structure. From such a point of view, Takeda and Nishida proposed the concept of fuzzy domination structures based on fuzzy convex cones. In this paper, we first consider multiobjective programming problems with fuzzy domination structures and define the solution concept using the alpha-level set for the fuzzy convex cone. After that, we focus on multilevel multiobjective programming problems with fuzzy domination structures where multiple decision makers in a hierarchical organization have their own multiple objective functions and their own fuzzy domination structures. After introducing the solution concept using the alpha-level sets for the fuzzy convex cones of multiple decision makers, we propose an interactive decision making method to obtain the satisfactory solution which reflects not only the hierarchical relationships between multiple decision makers but also their own preferences for their objective functions. An interactive process is demonstrated by means of an illustrative numerical example.

View full abstract

This paper considers a minimum spanning tree problem under the situation where costs for constructing edges in a network includes both fuzziness and randomness. In particular, this article focuses on the case that the edge costs are expressed by random fuzzy variables. A new decision making model is proposed in order to find a solution which fully reflects random and fuzzy information. It is shown that an optimal solution of the proposed model is obtained by a polynomial-time algorithm.

View full abstract

In recent studies of network formation, it is assumed that a player directly receives utility from the network, namely the player receives utility from any other players by forming links. Jackson and Wolinsky (1996) proposed a mathematical model such that the complete, the empty, and the star network can be stable. Hayashida et al. (2005) proposed a mathematical model including social reputation of the players such that the multiple disjoint components connecting some star networks can be also stable. In these mathematical models, it is assumed that the players are rational, and can discriminate between two payoffs with a minute difference and maximize their utility. Such optimization approaches are not always appropriate for analyzing human behaviors and social phenomena, however, a model basing on adaptive behavior can alternative to such optimization models. In this paper, for a complement of the mathematical model, we construct an adaptive behavioral model and analyze the structure of social networks through simulations.

View full abstract

This paper considers a possibility maximization model for the probability satisfying the total future return in each scenario including random fuzzy variables is more than or equal to a target value, based on possibilistic programming and stochastic programming. The main multi-criteria random fuzzy programming problem is not a well-defined problem due to including random variables and fuzzy numbers. Therefore, in order to solve it analytically, some criterions of probabilities for all objective functions are set and chance constraints are introduced. Furthermore, considering decision maker's subjectivity and flexibility of the original plan, a fuzzy goal for each objective function is introduced. Then, main problem is transformed into the deterministic equivalent problem. Since this problem is a nonlinear programming problem, the analytical solution method extending previous solution approaches is constructed.

View full abstract

In this paper, we propose a way of solving conflict resolution in agent control based on rough-set-based granularity. The solution is obtained as the lower approximation of possible actions selected from limited knowledge. Then, we explain how the lower approximation can be generated. In case that the lower approximation is an empty set, variable precision rough set model enables us to obtain the solution of conflict resolution.

View full abstract

Granular reasoning proposed by Murai et al. is a mechanism for reasoning using granular computing, and the concept of ``focus" has been proposed as a key concept of granular reasoning. On the other hand, the authors have proposed another concept of granularity, called ``visibility". In this paper, we try to capture the concepts of visibility and focus as modalities of modal logics by introducing Scott-Montague models that illustrate the visibility and focus by modal operators, respectively.

View full abstract

In the field of rough sets,methods for inducing minimal rules from a given decision table have been proposed. When the decision attribute is ordinal, we may induce rules w.r.t.upper and lower unions of decision classes. This approach would be better in simplicity of obtained rules than inducing rules w.r.t.decision classes directly. However, because of independent applications of rule induction methods, inclusion relations among upper/lower unions in conclusion parts are not inherited to the condition parts of obtained rules. This non-inheritance may debase the quality of obtained rules. In this paper, we propose a few approaches to inherit the inclusion relations among conclusions to the conditions of rules. The performacnes of the proposed approaches are examined by a numerical experiment.

View full abstract

A classification algorithm of decision tables has been proposed to investigate major opinions and the distribution of opinions in the given group of decision tables. In the approach, an agglomerative hierarchical clustering algorithm is applied using a similarity between clusters of decision tables. Therefore, the algorithm terminates after all decision tables compose a cluster so that the proper number of clusters cannot be determined automatically. In this paper, a few clustering methods based on the rational choice theory are proposed. It is shown that the algorithm can terminate before all decision tables compose a cluster so that the number of clusters is automatically determined.

View full abstract

Recently, a robotic hand with tactile sensors is important for manipulating objects dexterously. We have developed a universal robot hand with tactile sensors. However, computer resources are needed for processing tactile sensor information because tactile sensor has numerous measurement points in general. For resolving this problem, this paper proposes an adaptive tactile measurement method with the genetic algorithm. The validity of the proposed method shows in some experiments.

View full abstract

Genetic Algorithms (GAs) are well known as adaptive heuristic search algorithm to find approximate solutions. However for GAs, the problem of the premature convergence and falling in the local solution also need to be solved. Although we have proposed Fuzzy Adaptive Search method for Parallel Genetic Algorithm based on diversity measure as the improvement method, its performance depended on what is the type of applied problem. Therefore, in this paper we use history of evolution to improve the robustness of proposed method. Simulation results are also further presented to show the effectiveness and performance of method we proposed in this paper.

View full abstract

Integration of existing information accumulated in individuals, such as experience and knowledge, is effective for turning private judgments of various individuals into the collective decision. Such effective use of existing information will be a good reference for the development of dependable intelligent information systems that can cope with unfamiliar situation. This paper considers a representation model for the information integration from the view point of information theory. The representation model regards the manner of information integration as a mixture of probability distributions under the assumption that they characterize the structures of existing information sources. The mixture of probability distributions is designed through information divergence measures such as Kullback-divergnence and alpha-divergence. The designed mixture of probability distributions includes not only conventional linear type but also exponential and power types. The present paper also considers the relationship between the above mentioned representation model of information integration and so called ensemble learning.

View full abstract

This paper presents a receding horizon Nash game approach to formation control of mobile robots. The formation control is formulated as a linear-quadratic Nash differential game through the use of graph theory. Finite horizon cost function is discussed under the open-loop information structure. An openloop Nash equilibrium is investigated by the solutions of coupled (asymmetrical) Riccati differential equations. Based on finite horizon open-loop Nash equilibrium solution, a receding horizon approach is adopted to synthesize a state-feedback controller for the formation control. Mobile robots with double integrator dynamics are used in the formation control simulation. Simulation results are provided to justify the models and solutions.

View full abstract

This paper discusses the visual tracking of a partner robots through interaction with people based on computational intelligence. Steady-state genetic algorithms are applied for detecting human face and objects in image processing, but the computational cost is relatively high, because a genetic algorithm is one of stochastic search. Therefore, in this paper, we apply growing neural gas for the preprocessing of image processing. The reference vectors are used in the update of population of candidate solutions in human detection and object detection. We conducted several experiments of the partner robot on the interaction. The experimental results show that the proposed method can improve the search performance of people tracking and object tracking.

View full abstract

Particle Swarm Optimization(PSO) is widely used in optimization problems, due to its powerful search ability and easy implementation. Recently, multi-objective PSO is introduced to deal with multi-objective optimization problems. There are two differences from single objective PSO. One is an archive that preserves Pareto optimal candidates, and the other is selection strategies for the guide particles, such as the personal best and the global best. In this paper, based on particle topology we propose a novel strategy in guide selection in multi-objective PSO. Numerical simulation results show the availability of the proposed method in several benchmark problems.

View full abstract

In this work, we analyze the functionality transition in the evolution process of NSGA-II and an enhanced NSGA-II with the method of controling dominance area of solutions (CDAS) from the viewpoint of front distribution. We examine the relationship between the population of the 1st front consisting of non-dominated solutions and the values of two metrics, NORM and ANGLE, which measure diversity and convergence of Pareto-optimal solutions (POS), respectively. We also suggest potentials to further improve the search performance of the enhanced NSGA-II with CDAS by emphasizing the parameter S, whcih controls the degree of dominance by contracting or expanding the dominance area of solutions, before and after the boundary generation of functionality transition. Furthermore, we analyze the behavior of the evolution for the best parameters combination.

View full abstract

This paper proposes a new design support approach, which efficiently utilizes the information of many non-dominated solutions obtained from evolutionary multi-criterion optimization (EMO). The proposed approach consists of four mechanisms: grouping (clustering), reducing the number of candidates (selecting the representative solutions), dimensionality reduction, and estimation. In this paper, we examine the characteristics and effectiveness of the proposed approach through computational experiments on a design problem of a counter rotating axial fan turbojet engine. The design target of this problem is two fans and two turbines of the engine. We handle this task as a seven-objective design problem.

View full abstract

In this research, we applied an interactive Genetic Algorithm (iGA) to a product recommendation system. Products that suit a user's preference can be presented by applying iGA to the system and learning the user's preference. However, if the user's preference is biased, the dependency among design variables should be considered. Therefore, we proposed an offspring generation mechanism taking this dependency into consideration. In the proposed method, we first apply a clustering technique to the archived individuals selected by a user, and then construct a Probabilistic Model for crossover based on the clustering results. We discussed the effectiveness of the proposed mechanisms by experimenting with iGA for selecting colors and figures of symbols.

View full abstract

In this paper, we propose an interactive genetic algorithm that obtains valuations of individuals from multiple people. This technique is effective for consensus building and in bringing together many opinions for cases such as allowing multiple people to design clothes together. With an interactive genetic algorithm in which multiple people participate, a simple and easy evaluation interface is required. This technique determines the solution group of the genetic algorithm by the tournament method. In this technique, two solution candidates are presented to the evaluator. The evaluator then votes between the two solution candidates. A large number of evaluators judge the solution candidate's superiority or inferiority, and the relative superiority or inferiority of the two solutions is decided. The evaluation value of all solution candidates is decided depending on a relative value of the superiority or inferiority of the two solutions based on the results of this tournament competition. The obtained solution reflects the opinion of all evaluators well. In this paper, the effectiveness of the proposed technique was verified by a simulation using multiple evaluation agents in place of real evaluators. In each tournament competition, the result when 60% of all evaluation agents participated in a vote almost equaled the result when all the evaluation agents participated. This means the load to which the evaluator evaluates the solution candidate can be reduced by 40%. Moreover, we compared the proposed technique with a general voting method that uses a large number of evaluators voting for the first good individual from among all solution candidates in a displayed list. We confirmed that the proposed technique was more effective than a general voting method from the viewpoint that many participants are satisfied.

View full abstract