In the current Internet system, there are many problems using anonymity of the network communication such as personal information leaks and crimes using the Internet system. This is why TCP/IP protocol used in Internet system does not have the user identification information on the communication data, and it is difficult to supervise the user performing the above acts immediately. As a study for solving the above problem, there is the study of Policy Based Network Management (PBNM). This is the scheme for managing a whole Local Area Network (LAN) through communication control for every user. In this PBNM, two types of schemes exist. As one scheme, we have studied theoretically about the Destination Addressing Control System (DACS) Scheme with affinity with existing internet. By applying this DACS Scheme to Internet system management, we will realize the policy-based Internet system management. In this paper, concept of AI-Based automatic unauthorized communication prevention function in DACS-based PBNM Scheme is described.

Machine learning, deep learning and neural networks are extensively developed in many fields, with neural networks playing an important role in a wide variety of applications. However, a sufficient explanation of the structure and functionality of complex and deep neural networks is still needed. In this paper, it is shown that bio-inspired networks are useful for the explanation of network functions. First, the asymmetric network is created based on the biological retinal networks. Second, the classification performance of the asymmetric network is compared to that of the symmetric networks. The directional vectors in the asymmetric networks are generated on the adjacent neurons caused by movement stimulus, which create independent subspaces. Vectors for the movement stimulus are reported experimentally to be generated in the layered cortex in the brain. In this paper, it is shown computationally that many directional movement vectors are generated in the layered asymmetric networks, which create also independent subspaces. Further, when the correlational activities of the adjacent cells are represented in the directed vectors, they create independent subspaces than the direct inputs in the networks. These asymmetric subnetworks will facilitate the transmission of sensory information to higher-level processes such as efficient feature extraction, classification, and learning in the layered networks.

Models for artificial intelligence, machine learning, and neural networks are implemented on digital computers with a von Neumann architecture. Few studies have considered analog neural networks. Our previous study used multipliers to represent connecting weights in a neural network. The multipliers calculate the product of input signals and their corresponding connecting weights. However, using MOSFET multipliers, their operating range is limited by semiconductor characteristics. The input and output ranges for networks that use these multipliers are thus limited. Furthermore, the circuit operation becomes unstable. Here, we propose a logarithmic four-quadrant multiplier for representing connecting weights. The output of this multiple circuit is a more accurate value compared to the previous circuit. Experiments show that this multiplier exhibits stable operation over a wide range. Therefore, this model can be used directly for input/output of an analog control unit. A model that uses only analog electronic circuits is presented. Its learning time is quite short compared to that for models implemented on a digital computer. We increased the number of units and network layers. We suggest the possibility of a hardware implementation of a deep learning model. Furthermore, this model expects the elucidation of the biomedical learning mechanism.

In the current Internet system, there are many problems using anonymity of the network communication such as personal information leaks and crimes using the Internet system. This is why TCP/IP protocol used in Internet system does not have the user identification information on the communication data, and it is difficult to supervise the user performing the above acts immediately. As a study for solving the above problem, there is the study of Policy Based Network Management (PBNM). This is the scheme for managing a whole Local Area Network (LAN) through communication control for every user. In this PBNM, two types of schemes exist. As one scheme, we have studied theoretically about the Destination Addressing Control System (DACS) Scheme with affinity with existing internet. By applying this DACS Scheme to Internet system management, we will realize the policy-based Internet system management. In this paper, as a preliminary step, concept of the PBNM Scheme for multiple domains as cyber physical system that utilizes data science and AI is proposed.

Generation of useful variables and features is an important issue throughout the machine learning, artificial intelligence, and applied fields for their efficient computations. In this paper, the nearest neighbor relations are proposed for the minimal generation and the reduced variables of the functions in the threshold networks. First, the nearest neighbor relations are shown to be minimal and inherited for threshold functions and they play an important role in the iterative generation of the Chow parameters. Further, they give a solution for the Chow parameters problem. Second, convex cones are made of the nearest neighbor relations for the generation of the reduced variables. Then the edges of convex cones are compared for the discrimination of variables. Finally, the reduced variables based on the nearest neighbor relations are shown to be useful for documents classification.

Artificial intelligence (AI) and big data processing are extensively developed in recent information technologies. Deep learning in AI has been tried to apply to robotics, social activities and home appliances. Also, big data processing has been used in the networks and social services. There were fundamental problems in these recent technologies in AI and big data processing. One of these was “surface problems” in information technologies. The word “surface problem” is well known in physics, material sciences, and nanosciences as in the surface phenomena of the semiconductor, metals and crystal structures. As the surface problem in the information science, threshold element of neuron model is discussed, which are also fundamental elements used in deep learning neural networks. The surface is characterized by the surface extreme data, which discriminate classes of data. Threshold logic of their elements with much variables has not fully solved yet. We mention how to deal with such problem with much variables using surface data. As another problem of the surface problem, we have the reduction of higher dimensional data used in the big data processing. As the reduction method, reducts in the rough set are well known in the data processing. We mention how to generate reducts using the surface border data among classes.

Dimension reduction of data is an important issue in the data processing and iti is needed for the analysis of higher dimensional data in the application domain. Rough set is fundamental and useful to reduce higher dimensional data to lower one. Reduct in the rough set is a minimal subset of features, which has the same discernible power as the entire features in the higher dimensional scheme. The nearest neighbor relation with minimal distance proposed here has a fundamental information for classification. The nearest neighbor relation plays a fundamental role for generation of reducts and threshold functions using the Boolean reasoning on the discernibility and in discernibility matrices, in which the indiscernibility matrix is proposed here to test the sufficient condition for reduct and threshold function. Then, genetation methods for the reducts and threshold functions based on the nearest neighbor relation are proposed here using Boolean operations on the discernibility and the indiscernibility matrices.

In the current Internet system, there are many problems using anonymity of the network communication such as personal information leaks and crimes using the Internet system. This is why TCP/IP protocol used in Internet system does not have the user identification information on the communication data, and it is difficult to supervise the user performing the above acts immediately. As a study for solving the above problem, there is the study of Policy Based Network Management (PBNM). This is the scheme for managing a whole Local Area Network (LAN) through communication control for every user. In this PBNM, two types of schemes exist. The first is the scheme for managing the whole LAN by locating the communication control mechanisms on the course between network servers and clients. The second is the scheme of managing the whole LAN by locating the communication control mechanisms on clients. As the second scheme, we have studied theoretically about the Destination Addressing Control System (DACS) Scheme. By applying this DACS Scheme to Internet system management, we will realize the policy-based Internet system management. As the second step of the second phase for the last goal, we examine the implementation method of the cloud type virtual PBNM which can be used by plural organizations in this paper.

Summary form only given. Extensive developments are carried out in the fields of Applied Informatics. Among these fields, the fundamental problem is to develop the specific data processing and to provide each technology for the discipline of the Informatics. First, I would like to talk about an example of Neuro Informatics and Biomedical Informatics, which I was engaged in the research of the School of Medicine. In this talk, I will show the data processing problem in the experimental neurophysiology study and how to solve the problem in cooperation with Shannon information theory. Second, text classification problem in the Data Informatics is taken. In this discipline, since the text processing needs multiple distance measurements, the problem is how to cope with these situations for the higher classification accuracy. In the rough set theory, the data reduction is carried out by reducts, which have minimal discernibility classification ability. In this study, the multiple reducts corresponding to different distance measures are shown to be useful for the classification.

Dimension reduction of data is an important theme in the data processing and on the web to represent and manipulate higher dimensional data. Rough set is fundamental and useful to process higher dimensional data. Reduct in the rough set is a minimal subset of features, which has the same discernible power as the entire features in the higher dimensional scheme. It is shown that nearest neighbor relation with minimal distance introduced here has a basic information for classification. In this paper, a new reduct generation method based on the nearest neighbor relation with minimal distance is proposed. To improve the classification accuracy of reducts, we develop a nonlinear mapping method on the nearest neighbor relation, which makes vector data relation among neighbor data and preserves data ordering.

In this study, we examine the relationship between review processes on embedded software development projects. We use the Shapiro-Wilk test and Spearman's rank correlation coefficient to analyze them. The results indicate that the data selected was not from a normally distributed population. The results of Spearman's rank correlation coefficient exhibite the following: (1) the number of errors identified in conceptual design review (CDR) has a slightly effect on the number of errors identified in subsequent review processes except for module design review (MDR), (2) the same holds for MDR and programming review (PGR), (3) the number of errors identified in functional design review (FDR) influences the number of errors identified in subsequent review processes except for MDR, (4) the precision of structural design (SD) has an effect on the precision of programming (PG).

In the current Internet-based systems, there are many problems using anonymity of the network communication such as personal information leak and crimes using the Internet systems. This is because the TCP/IP protocol used in Internet systems does not have the user identification information on the communication data, and it is difficult to supervise the user performing the above acts immediately. As a solution for solving the above problem, there is the approach of Policy-based Network Management (PBNM). This is the scheme for managing a whole Local Area Network (LAN) through communication control of every user. In this PBNM, two types of schemes exist. The first is the scheme for managing the whole LAN by locating the communication control mechanisms on the course between network servers and clients. The second is the scheme of managing the whole LAN by locating the communication control mechanisms on clients. As the second scheme, we have been studied theoretically about the Destination Addressing Control System (DACS) Scheme. By applying this DACS Scheme to Internet system management, we intend to realize the policy-based Internet system management finally. In the DACS Scheme, inspection is not done about compatibility to cloud environment with virtualization technology that spreads explosively. As the result, the coverage of the DACS Scheme is limited only in physical environment now. In this study, we inspect compatibility of the DACS Scheme for the cloud environment with virtualization technology, and enlarge coverage of this scheme. With it, the Virtual DACS Scheme (vDACS Scheme) is established.

In the neural network field, many application models have been proposed. Previous analog neural network models were composed of the operational amplifier and fixed resistance. It is difficult to change the connecting weight of a network. In this study, we used analog electronic multiple and sample hold circuits. The connecting weights describe the input voltage. It is easy to change the connection coefficient. This model works only on analog electronic circuits. It can finish the learning process in a very short time and this model will enable more flexible learning. However, the structure of this model includes only one input and one output network. We improved the number of unit and network layers. Moreover, we suggest the possibility of the realization the hardware implementation of the deep learning model.

Japan has problems with the declining birth rate and growing of aging population. The number of solitary aged persons has been on the increase. In social situation like this, a solitary death of aged person becomes a serious problem and to establish the system to prevent it is urgent necessity. Some safety equipments for aged people are already sold in Japan, but they have not been widely used because most of them are considerably expensive and users need to pay contract price or usage fee. By taking into consideration an increase in solitary aged persons toward an aged person society, we studied and developed the application, “Anshin”, for mobile devices to prevent solitary death and the network system, “Otassha Tushin”, to monitor and support conditions of solitary aged persons on a server. We will distribute these systems for solitary aged persons and their family or caregivers for free.

We try to develop a support application for physically handicapped children to communicate with others by blinks. Because of limited body movements and mental disorders, many of them cannot communicate with their families or caregivers. If they can use applications in smart phones by blinks, they will be able to tell what they really need or want. Fist, we try to detect an eye area by using OpenCv. Then we develop the way to detect opening and closing of eyes. We combine the method using saturation and using complexity of image to get more accurate results to detect blinks. Then we develop the technique into a communication application that has the accurate and high-precision blink determination system to detect letters and put them into sound.

The incremental learning is a method to compose an associate memory using a chaotic neural network and provides larger capacity than correlative learning in compensation for a large amount of computation. A chaotic neuron has spatiotemporal sum in it and the temporal sum makes the learning stable to input noise. When there is no noise in input, the neuron may not need temporal sum. In this paper, to reduce the computations, a simplified network without temporal sum are introduced and investigated through the computer simulations comparing with the network as in the past. It turns out that the simplified network has the same capacity to and can learn faster than the usual network.

Dimension reduction of data is an important theme in the data processing and on the web to represent and manipulate higher dimensional data. Rough set is fundamental and useful to process higher dimensional data. Reduct in the rough set is a minimal subset of features, which has the same discernible power as the entire features in the higher dimensional scheme. Nearest neighbor relation between different classes has a basic information for classification. We propose here a reduct generation method based on the nearest neighbor relation. To characterize the classification ability of reducts, we develop a new graph mapping method of the nearest neighbor based on reducts and weighted modified reducts for the classification with higher accuracy.

Dimension reduction of data is an important theme in the data processing and on the web to represent and manipulate higher dimensional data. Reduct in the rough set is a minimal subset of features, which has almost the same discernible power as the entire features in the higher dimensional scheme. But, there are problems in the application of reducts for classification. Here, we develop a method which connects reducts and the nearest neighbor method to classify data with higher classification accuracy. To improve the classification ability of reducts, we develop a new graph mapping method of the nearest neighbor based on reducts and weighted modified reducts for the classification with higher accuracy. Then, the mapping method is useful and the weighted modified reduct classifies with higher accuracy.

In this study, we examine the effect of reviewing processes for embedded software development projects. We analyze the relationship between review processes and debugging process using the Shapiro-Wilk test and Spearman's rank correlation coefficient. The results indicate that the data selected was not from a normally distributed population and that review processes can reduce the number of development errors.

In this study, we try to develop a new application for physically handicapped children to communicate with others by a blink. Because of limited body movements and mental disorders, many of them cannot communicate with their families or caregivers. We think if they can use application in smart phones by a blink, it will be big help for them to tell caregivers what they really need or want to tell. First, we try to detect an eye area by using OpenCv. Then we develop the way to detect opening and closing of eyes. We combine the method using situation and using complexity of image to get more accurate results to detect a blink. The level of handicapped is very varied in children. So we will try to develop the application to be able to customize depends on the situation of users. And also, we will try to reduce the error to detect a blink and pursue the high precision of the eye chased program.

To make clear the mechanism of the visual movement is important in the visual system. The problem is how to perceive vectors of the optic flow in the network. First, the biological asymmetric network with nonlinear ties is analyzed for generating the vector from the point of the network computations. The results are applicable to the V1 and MT model of the neural networks in the cortex. The stimulus with a mixture distribution is applied to evaluate their network processing ability for the movement direction and its velocity, which generate the vector. Second, it is shown that the vector is emphasized in the MT than the V1. The characterized equation is derived in the network computations, which evaluates the vector properties of processing ability of the network. The movement velocity is derived, which is represented in Wiener kernels. The operations of vectors are shown in the divisive normalization network, which will create curl or divergence vectors in the higher neural network as MST area.

We developed a new VOCA (Voice Output Communication Aid) application for personal digital assistant (PDA), “Let's Talk!” for autistic children. This application has many particular advantages comparing to existing VOCA. We especially focused on an easy and simple manipulation. By tapping a symbol on a screen of a PDA with this application, a user can show his/her thoughts with pictures and sounds to others easily. There are 2 modes that can be switched depending on different situations of users. It has 120 symbols based on daily life and a user can also create the original page with new icons made by pictures or sound. A user also can customize an original page by arranging icons he made or existing symbols. On the newest version of this application, we added Task Schedule System to stimulate motivations of children to do something by them. On the last part of this study, we show some case studies. We introduced this application to students in a school for handicapped children and collected data.

In the neural network field, many application models have been proposed. A neuro chip and an artificial retina chip are developed to comprise the neural network model and simulate the biomedical vision system. Previous analog neural network models were composed of the operational amplifier and fixed resistance. It is difficult to change the connection coefficient. In this study, we used analog electronic multiple and sample hold circuits. The connecting weights describe the input voltage. It is easy to change the connection coefficient. This model works only on analog electronic circuits. It can finish the learning process in a very short time and this model will enable more flexible learning.

We have been using animation creation to stimulate students' interest in programming. We proposed a method to measure a beginner's programming ability and investigated the animation programs created by these beginners over several years. The function point (FP) and lines of code (LOC) methods were used to quantitatively evaluate the programming ability. In this study, we focused on the relationship between LOC and FP. The LOC/FP indicates the relationship between the number of lines and the number of functions in the program. To investigate the relationship, cluster analysis was used to divide the animations into groups.

In this paper, we create effort prediction models using self-organizing maps (SOMs) for embedded software development projects. SOMs are a type of artificial neural networks that rely on unsupervised learning. They produce a low-dimensional, discretized representation of the input space of training samples, these representations are called maps. SOMs are useful for visualizing low-dimensional views of high-dimensional data a multidimensional scaling technique. The advantages of using SOMs for statistical applications are as follows: (1) enabling reasonable inferences to be made from incomplete information via association and recollection, (2) visualizing data, (3) summarizing large-scale data, and (4) creating nonlinear models. We focus on the first advantage to create effort prediction models. To verify our approach, we perform an evaluation experiment that compares SOM models to feed forward artificial neural network (FANN) models using Welch's t test. The results of the comparison indicate that SOM models are more accurate than FANN models for the mean of absolute errors when predicting the amount of effort, because mean errors of the SOM are statistically significantly lower.

In the computer literacy with a PC at the Japanese university, it was clarified that a touch-typing skill was considerably improved by the touch-typing practice which was given to the first-year students for 15 to 20 minutes. These data was accumulated for seven years from 2002. In addition, it is discussed about how much change in achievement degree was observed for the students who continued this practice to the second-year from the first-year. Furthermore, the authors mentioned the relations between the actual values measured from the touch-typing practices and the self-evaluation of the effect made by the students themselves.

As the work for managing a whole network effectively without a limited purpose, there is the work of a PBNM (Policy-based network management). The PBNM has two structural problems such as communication concentration from many clients to a communication control mechanism called PEP (Policy Enhancement Point) and the necessity of the network system updating at the time of introducing the PBNM into LAN. Moreover, user support problems in campus-like computer networks such as trouble user support at updating a client’s setups and coping with annoying communication cannot be improved by the PBNM. To improve these problems, we have been shown a next generation PBNM which overcomes theses problems and has the function which does not exist in the existing PBNM, and called it a DACS (Destination Addressing Control System) Scheme. By the DACS Scheme, communication concentration from many clients to the PEP is solved, and system updating becomes unnecessary. Moreover, user support at updating the client’s setups and coping with annoying communication by the DACS Scheme becomes very effective. In this study, we implement a DACS system to realize a concept of the DACS Scheme.

Most classification studies are done by using all the objective data. It is expected to classify objects by using some subsets data effectively. A rough set based reduct is a minimal subset of features, which has almost the same discernible power as the entire features. Here, we propose multiple reducts which are followed by the k-nearest neighbor with confidence to classify documents with higher classification accuracy. To select better multiple reducts for the classification, we develop a greedy algorithm for the multiple reducts, which is based on the selection of useful attributes for the documents classification. These proposed methods are verified to be effective in the classification on benchmark datasets from the Reuters 21578 data set.

Today embedded software is widely used. As the complexity of embedded systems grows, the number of tasks performed by embedded software is increasing. These tasks must communicate with each other to achieve synchronization. And therefore, the growing complexity of communications is making it increasingly difficult to program and debug these tasks. To confirm the correctness of embedded software, this paper introduces a formal specification language for embedded systems that can deal with their concurrent nature. After formalizing the APIs of an embedded real-time operating system, we demonstrate how to describe and verify embedded software formally.

Most classification studies are done by using all the objects data. It is expected to classify objects by using some subsets data in the total data. A rough set based reduct is a minimal subset of features, which has almost the same discernible power as the entire conditional features. Here, we propose a greedy algorithm to compute a set of rough set reducts which is followed by the k-nearest neighbor to classify documents. To improve the classification performance, reducts-kNN with confidence was developed. These proposed rough set reduct based methods are compared with the classification by AdaBoost and SVM(Support Vector Machine) methods. Experiments have been conducted on some benchmark datasets from the Reuters 21578 data set.

Neural networks are able to learn more patterns with the incremental learning than with the correlative learning. The incremental learning is a method to compose an associative memory using a chaotic neural network. In the former work, it was found that the capacity of the network increases along with its size, with some threshold value and that it decreases over that size. The threshold value and the capacity varied by the learning parameter. In this paper, the capacity of the networks was investigated by changing the learning parameter. Through the computer simulations, it turned out that the capacity increases in proportion to the network size. Then, the error correction capability is estimated with learned patterns changing to the maximum capacity.

Paintings have some sensibility information to human hearts. But, the sensibility information will be different to person to person. We like some paintings, while other persons like another ones. For appreciation of paintings, grouping of paintings with similar sensitivity will be helpful. In this paper, we developed a distance measure relation among paintings for similarity grouping. Four similarity relations (types) are developed by the distance measure for classification of paintings. Impression words and color words play an important role for the sensibility expression of paintings. To verify the combination of these words, a self-organizing map (SOM) by two layered neural network, was applied for the grouping of paintings. It was shown that the combination of the impression words and color ones make a well similar grouping of paintings. Grouping similar paintings in impression words and color ones will be useful for a recommendation system.

Vehicle routing problem (VRP) is a problem to find the minimum path length for a fleet of vehicles which deliver goods from the depot according to demands. Some specific methods for the VRP include evolutionary approaches such as genetic algorithms. A naive data structure (i.e., chromosome) for the evolutionary approaches is called path representation (PR) which is an ordered list structure which represents a single path. On the other hand, we propose a novel data structure called R-tree based path representation (R-PR). R-PR is a tree structure based on a spatial indexing structure called R-tree. Each node of R-PR represents a sub-path (i.e., a partial solution) for VRP, and its parent node represents a set of the sub-paths, hierarchically. We compare R-PR with PR by using some of VRP instances by Augerat et al, and the results show that R-PR outperforms PR in large scale instances.

The process of creating photorealistic 3-dimensional computer graphic (3DCG) images is divided into two stages, i.e., modeling and rendering. Automatic rendering has gained popularity, and photorealistic rendering is generally used to render different types of images. However, professional artists still model characters manually. Moreover, not much progress has been achieved with regard to 3-D shape data acquisition techniques that can be applied to complex object modeling; this is an important problem hampering the progress of 3DCG. Generally, a laser and a highly accurate camera are used to acquire 3-D shape data. However, this technique is time-consuming and expensive. Further, the eyes may be damaged while measuring by this method. In order to solve these problems, we have proposed a simple method for 3-D shape data acquisition using a projector and a Web camera. This method is economical, and simpler than conventional techniques. In this paper, we describe the setup of the projector and Web camera, shape data acquisition process, image processing, and generation of a photorealistic model. We also propose a method for application to complex objects using several cameras.

As the work for managing a whole network effectively without a limited purpose, there is the work of PBNM (policy-based network management). PBNM has two structural problems such as communication concentration from many clients to a communication control mechanism called PEP (policy enhancement point) and the necessity of the network system updating at the time of introducing PBNM into LAN. Moreover, user support problems in campus-like computer networks such as trouble user support at updating a client's setups and coping with annoying communication can not be improved by PBNM. To improve these problems, we show a next generation PBNM which overcomes theses problems and has the function which does not exist in existing PBNM, and called it DACS (destination addressing control system) scheme. By DACS scheme, communication concentration from many clients to PEP is solved, and system updating becomes unnecessary. Moreover, user support at updating the client's setups and coping with annoying communication by DACS scheme becomes very effective.

In standard IP networks, the control of communication of individual connections is an important aspect of network management. Generally, we distinguish two models of communication control. One is based on the exploitation of standard mechanisms, such as IP filtering, port blocking, routing rules, deployed along the path between the client and the server. We refer to these mechanisms as infrastructure-based management model. On the other hand, the control can be carried out by controlling the client machines of every user. This new model is a client-side management model, and it has been realized in our previous work, by means of a dedicated framework we called DACS. Essentially, DACS distributes the control within the client machines. Therefore, also the network management processing workload is distributed. While in our previous work we investigated functions of DACS scheme and network services, here we are more concerned with analyzing the processing workload due to communication control demanded to the client machines in practical networks. In this regard, we evaluate DACS performance from the point of view of the processing load and demonstrate by some preliminary experiments that DACS framework is better than infrastructure-based management models. From this result, we conclude that, DACS outperforms existing solutions for the communication control problem, because it can control a whole local area network by distributed control of client machines, with an acceptable increase of the processing workload.

The k-nearest neighbor(k-NN) is improved by applying rough set and distance functions with relearning and ensemble computations to classify data with the higher accuracy values. Then, the proposed relearning and combining ensemble computations are an effective technique for improving accuracy. We develop a new approach to combine kNN classifier based on rough set and distance functions with relearning and ensemble computations. The combining algorithm shows higher generalization accuracy, compared to other conventional algorithms. First, to improve classification accuracy, an instance-based learning method with genetic algorithm is developed. Second, additional ensemble computations are followed by the relearning. Then, rough set approach for the classification, is discussed. Experiments have been conducted on some benchmark datasets from the UCI Machine Learning Repository.

In an earlier paper, we established 9 models for estimating errors in a new project. In this paper, we integrate the 9 models into 5 by investigating similarities among the models. In addition, we establish a new model using an artificial neural network (ANN). It is becoming increasingly important for software-development corporations to ascertain how to develop software efficiently, whilst guaranteeing delivery time and quality, and keeping development costs low. Estimating the manpower required by new projects and guaranteeing the quality of software are particularly important, because the estimation relates directly to costs while the quality reflects on the reliability of the corporations. In the field of embedded software, development techniques, management techniques, tools, testing techniques, reuse techniques, real-time operating systems and so on, have already been studied. However, there is little research on the relationship between the scale of the development and the number of errors using data accumulated from past projects. Hence, we integrate the previous models and establish a new model using an artificial neural network (ANN). We also compare the accuracy of the ANN model and the regression analysis models. The results of these comparisons indicate that the ANN model is more accurate than any of the 5 integrated models.

As a communication mean, personal portal, which is an entrance where each user can acquire his/her interested information on the network, is often used. We have problems to realize the personal portal described as follows. First, the program to realize the personal portal, which is located on a specific Web server, becomes large-scaled and complicated. Moreover, because Web pages as personal portal depend on that program, those pages can't always be useful for each user equally. To solve these problem, we develop two kinds of new Web service which are realized on the network introducing DACS Scheme, which is the network management scheme by communication control of client computers. By use of these Web Services, regardless of information starage form, each user can create the customized personal portal to use information on the network easily and freely for oneself. That is, a free information usage system is realized.

As a communication mean, personal portal which is an entrance where each user can acquire his/her interested information on the network, is often used. We have problems on the existing personal portal described as follows. First, the program to realize the personal portal which is located on a specific Web server, becomes large-scaled and complicated. Because Web pages as personal portal depend on that program, those Web pages can't always be useful for each user equally. Therefore, to solve these problems, we developed two kinds of new Web service which are realized on the network introducing DACS(destination addressing control system) scheme, which is the network management scheme by communication control of client computers. By use of these Web services, each user can create the customized personal portal, which displays the information by a user unit, for using in actual network easily and freely for oneself. In this paper, to deal with information not only by a user unit, but also by a group unit and by all users unit, DACS Web service is proposed. DACS Web service is realized by extending two kinds of functions of Web service respectively and integrating these functions. As the result, each user can create the customized Web page as personal portal, which will be often useful in actual network.

In this study, we analyze and evaluate development data obtained in the course of "OMRON software Co.", that develops equipments used in financial markets, as well as data obtained during the development of software. Because of this, it is becoming very important for the software- development corporations to find how to develop software efficiently while guaranteeing delivery time and quality, and holding down developing costs. Especially, estimating manpower of new projects and guaranteeing quality of software are important, because the estimation relates to costs and the quality relates to the reliability of corporations. In the field of embedded software, development techniques, management techniques, tools, testing techniques, reusing techniques, real time operating systems and so on have already been studied. However, there is few studies about the relationship between the development scales and the total errors using the data accumulated by past projects. Hence, we establish a model for the first trial to estimate the errors of a new project by using regression analysis.

Where customers with different membership and position, use computers as in the university network systems, it often takes much time and efforts for them to cope with the change of the system management. This is because the requirements for the respective computer usage are different in the network and security policies. In this paper, destination addressing control system scheme (DACS scheme) for the university network services is shown. The DACS scheme performs network management efficiently through the communication management of a client. As the characteristic of DACS scheme, only the setup modification is required by a system administrator, when configuration change of the network system is performed. Then, the setup modification is unnecessary by a customer, which shows a merit for both a system administrator and a customer. However, there is a problem to be solved. In existing DACS scheme, it is promised that, DACS client for controlling the communication must be implemented on all clients. When the client not installing DACS client is connected to the network, each network server can be accessed from the client. Therefore, in this paper, secure DACS scheme with function to prevent the communication from the client not installing DACS client, is proposed and examined.

When operation and management in the university network are performed, various problems occur. As one of those problems, user support problems in the network are often shown. As an example, support at the time of changing setups of a client computer by a user, is described. In this case, it take trouble very much for a system administrator because it is often necessary to support individual users, respectively. As another example, it is necessary to take some measure for annoying communication to other users. In this case, much time and effort are spent to specify where the client with annoying communication is under the management by DHCP service. To simplify user supports such as those cases, a new user support by DACS (destination addressing control system) scheme, which is an efficient scheme for the network operation and management in university network, is examined.

In a corporation or the division of a corporation where software is being developed, it is becoming very important to develop software efficiently while guaranteeing the quality, limiting the cost, and maintaining the development schedule. Therefore, the corporation and the division of the corporation that develop software are implementing various improvement methods, including process improvement. In this study we have analyzed data from a software development project and studied how to determine which software development project will fail because it takes more manpower that originally estimated. In addition, we have implemented project-management support tools by combining the project-monitoring and managing tools that are already in use. We have also developed a new model that uses the statistical method, by reviewing the tools that are already used to estimate manpower for new projects

Since data is becoming greatly large in the networks, the machine classification of the text data, is not easy under these computing circumstances. Though the k-nearest neighbor (kNN) classification is a simple and effective classification approach, the improving performance of the classifier is still attractive to cope with the high accuracy processing. In this paper, the kNN is improved by applying the different distance functions with weights to measure data from the multi-view points. Then, the weights for the optimization, are computed by the genetic algorithms. After the learning of the trained data, the unknown data is classified by combining the multiple distance functions and ensemble computations of the kNN. In this paper we present a new approach to combine multiple kNN classifiers based on different distance functions, which improve the performance of the k-nearest neighbor method. The proposed combining algorithm shows the higher generalization accuracy when compared to other conventional learning algorithms.

The k-nearest neighbor (KNN) classification is a simple and effective classification approach. However, improving performance of the classifier is still attractive. Combining multiple classifiers is an effective technique for improving accuracy. There are many general combining algorithms, such as Bagging, Boosting, or Error Correcting Output Coding that significantly improve the classifier such as decision trees, rule learners, or neural networks. Unfortunately, these combining methods developed do not improve the nearest neighbor classifiers. In this paper, first, we present a new approach to combine multiple KNN classifiers based on different distance functions, in which we apply multiple distance functions to improve the performance of the k-nearest neighbor classifier. Second, we develop a combining method, in which the weights of the distance function are learnt by genetic algorithm. Finally, combining classifiers in error correcting output coding, are discussed. The proposed algorithms seek to increase generalization accuracy when compared to the basic k-nearest neighbor algorithm. Experiments have been conducted on some benchmark datasets from the UCI machine learning repository. The results show that the proposed algorithms improve the performance of the k-nearest neighbor classification.

In authors programming course, students create animations as programming practice because this type of practice will make students interested in learning programming. This type of animation creation practice is well received by students. For the animation creation, we provide students with material such as fundamentals for creating basic animation, explanations of graphics functions, basic animation creation method, its application and its programming example in printed form (black and white). However, it is difficult for students to see the actual color or motion in printed material, because some of the animation is only shown on the screen and it is difficult for the students to actually reproduce the motion or examine the source codes by themselves. In order to overcome those problems, we have developed teaching material on the Web and actually used them in programming course. As a result it is clear that this material contributes to the increase in LOC, FP and the volume of animation created. In this paper, we give the general description of programming course, the objective of developing teaching material on the Web and guidelines for it, a general description of material, its effectiveness and problems we may see in the future.

Among prominent features of the visual networks, movement detections are carried out in the visual cortex. The visual cortex for the movement detection, consist of two layered networks, called the primary visual cortex (VI), followed by the middle temporal area (MT), in which nonlinear functions play important roles in the visual systems. In this paper, the fundamental characteristics in VI and MT model networks, are discussed by analyzing the asymmetric neural networks. VI and MT model networks, which are decomposed into nonlinear sub-asymmetrical networks. By the optimization of the asymmetric networks, movement detection equations are derived. Then, it was clarified that the even-odd nonlinearity combined asymmetric networks, are fundamental in the movement detection. These facts are applied to two layered VI and MT networks, in which it was clarified that the second layer MT has an efficient ability to detect the movement.

In general, mathematical subjects are used for computer software programming education in universities. In this case, many students often lose interest in computer software programming because the students have the preconception that it is difficult for them to make program using mathematical knowledge. In particular beginners at the programming tend to lose their interest. Therefore it has been pointed out that subjects who do not require mathematical knowledge should be used as much as possible. In this paper, the authors have tried to create animation generated by the programming instead of mathematical subjects by using C language that is used in many fields. The authors discuss about improvements of the interest of students for programming that is to make animation by programs in a programming class and refer to its effects.

Programming is one of the most important factors in the education of computer literacy. However, in the initial learning of programming, many students often have faced a simple problem, like syntax-errors by carelessness or mis-typing. Therefore, they feel that programming is too hard. The most important problem is the student can't take advantage of the error messages displayed by the computer. These messages are difficult to understand for beginners because the message only indicates a syntactic error and where it occurred. The authors propose a support system for C language programming on UNIX by using agent technology. This system monitors the error messages from the compiler. It analyzes, rewrites the acquired message, and reports to the student and the teacher on the network. The teacher can improve the teaching method by finding the category of error. The student customizes these messages, and they can use the suitable one at each level. Moreover, these messages are sent to other students by the communication line of each agent. Consequently, many students are able to use these useful messages; they can select the most preferable message from the collection of the agent. The authors executed a preliminary experiment and the improvement of efficiency to the complication work was confirmed.

In this paper we present a hybrid text categorization method based on Rough Sets theory. A central problem in good text classification for information filtering and retrieval (IF/IR) is the high dimensionality of the data. It may contain many unnecessary and irrelevant features. To cope with this problem, we propose a hybrid technique using Latent Semantic Indexing (LSI) and Rough Sets theory (RS) to alleviate this situation. Given corpora of documents and a training set of examples of classified documents, the technique locates a minimal set of co-ordinate keywords to distinguish between classes of documents, reducing the dimensionality of the keyword vectors. This simplifies the creation of knowledge-based IF/IR systems, speeds up their operation, and allows easy editing of the rule bases employed. Besides, we generate several knowledge base instead of one knowledge base for the classification of new object, hoping that the combination of answers of the multiple knowledge bases result in better performance. Multiple knowledge bases can be formulated precisely and in a unified way within the framework of RS. This paper describes the proposed technique, discusses the integration of a keyword acquisition algorithm, Latent Semantic indexing (LSI) with Rough Set-based rule generate algorithm, and provides experimental results. The test results show the hybrid method is better than the previous rough set-based approach.

Recommender Systems (RS), allow users to share information about items they like or dislike and obtain, in a timely fashion, recommendations based on predictions about unseen items (physical or information goods and/or services). In this process, users' preferences are considered to be the learning target functions. We study Agent-based Recommender Systems (ARS) under the scope of online learning in Multi-Agent systems (MAS). This approach models the problem as a pool of independent cooperative predictor agents, one per each user (the masters) in the system, in situations in which each agent (the learners) faces a sequence of trials, with a prediction to make in every step, eventually getting the correct value from its master. Each learner is willing to discover the degree of similarity among the target function of its master and those of other agents' masters (i.e. preference similarity). The agent uses this information for the calculation of its own prediction task, the goal being to make as few mistakes as possible. A simple, yet effective method is introduced in order to construct a compound algorithm for each agent by combining memory-based individual prediction and online weighted-majority voting. We give a theoretical mistake bound for this algorithm that is closely related to the total loss of the best predictor agent in the pool. Finally, we conduct some experiments obtaining results that empirically support these ideas and theories.

Five color spaces, RGB, XYZ, YIQ, HSV and LUV color spaces, were examined to know which was the most suitable to extract face regions. From the observations of flesh-colored pixel distributions in those color spaces, an ellipse from the UV plane in the LUV color space successfully extracted a face region. A proposed method using LUV color space and ellipse extraction was an extension of the method proposed by (Hongo et al., 1998) that used a circle instead of an ellipse, and it improved extracted results by 6% in our agreement index defined in this paper. Extracted rotated face images by 20 degrees interval were used to make an automatic face direction detection system. The face direction was estimated through two processes. The first process was a projection into an eigenspace, and the second process was an estimation of direction by a neural network. At first, a seven-dimensional eigenspace was obtained from the principal component analysis of ten face images. Then, a neural network learned to detect the face direction by using the projected coordinates of these images in the eigenspace.

An accounting education CAI system was developed which could be accessed through the Web. The system automatically adjusted the difficulty of questions by means of fuzzy inferences which were obtained from a number of wrong answers, answer time and a number of times of hint-referred.

The chaotic neural networks that learn patterns by associative memory can recall aperiodic pattern sequences including the learned ones. We report that applying presynaptic inhibition to a chaotic neural network to control the chaotic behavior realized the search of cyclic associative memory by matching features with chaos. Nara et al. (1992) reported the search of cyclic associative memory by changing the connection number of neurons. Using the patterns the same as Nara's report, the same searches ware carried out in chaotic neural networks. The results shows that search by the chaotic neural network can attain a better success rate, but the speed is lower. To improve the time for search the presynaptic inhibition function is reformed.

We propose a new concept of a support system for marine engine operation onboard ship. In the concept an agent system is introduced, which has a man-machine interface of speech communication and acts as the skilled engineer by using the chief engineer's knowledge. In the near future, the marine engine system will be operated by a deck officer at the bridge who does not have enough experience. To cope with the emergency and normal start-stop situation, several intelligent control systems are developed, but it can only change the engine load and indicate the failure parts. We have proposed and have been developing the Chief Engineer System (CE-SYS) which instructs the crew about the preferable operation during each step of the engine conditions. CE-SYS is a computer agent system, which has the experienced chief engineer's behavior. CE-SYS creates a comment from the condition of the engine automatically and tells it to the crew as an instruction. CE-SYS is equipped with a learning function of the crew's ability and personality through speech communication to generate an appropriate instruction. This makes the efficiency of the data collection for control. We confirmed the effectiveness of the experiment on the training ship Yuge Maru.

A new off-line writer recognition method is proposed which uses a 2D membership function obtained by piling up characters as a dictionary. This is an extension of our previous work that also used the 2D membership function with /spl alpha/-cut. This time the principal component analysis was used to characterize one's handwriting features, and the dictionary was constructed of characters not far from the mean character. This method was applied to writer recognition of 21 students who wrote 8 types of "hiragana" characters and an 86.5% recognition rate was obtained.

A new off-line writer recognition method is proposed which uses a 2D membership function obtained by piling up characters with /spl alpha/-cuts. Furthermore, principal component analysis is used to characterize the handwriting features and a dictionary of selected characters (those that are not far from the mean character) is constructed. The writer recognition is done by the production of similarities between three types of characters. This method was applied to writer recognition of 108 students who wrote 46 types of "hiragana" characters and a 97.8% recognition ratio was obtained.

There are many cases that a neural-network-based system must memorize some new patterns incrementally. However, if the network learns the new patterns only by referring to them, it probably forgets old memorized patterns, since parameters in the network usually correlate not only to the old memories but also to the new patterns. A certain way to avoid the loss of memories is to learn the new patterns with all memorized patterns. It needs, however, a large computational power. To solve this problem, we propose incremental learning methods with retrieving interfered patterns (ILRI). In these methods, the system employs a modified version of a resource allocating network (RAN) which is one variation of a generalized radial basis function (GRBF). In ILRI, the RAN learns new patterns with a relearning of a few number of retrieved past patterns that are interfered with the incremental learning. In this paper, we construct ILRI in two steps. In the first step, we construct a system which searches the interfered patterns from past input patterns stored in a database. In the second step, we improve the first system in such a way that the system does not need the database. In this case, the system regenerates the input patterns approximately in a random manner. The simulation results show that these two systems have almost the same ability, and the generalization ability is higher than other similar systems using neural networks and k-nearest neighbors.

Article Exploiting heterogeneous parallelism in the presence of communication delays Share on Authors: Dingchao Li Educational Center for Information Processing, Nagoya Institute of Technology, Gokisocho Showaku, Nagoya, 466, Japan Educational Center for Information Processing, Nagoya Institute of Technology, Gokisocho Showaku, Nagoya, 466, JapanView Profile , Yuji Iwahori Educational Center for Information Processing, Nagoya Institute of Technology, Gokisocho Showaku, Nagoya, 466, Japan Educational Center for Information Processing, Nagoya Institute of Technology, Gokisocho Showaku, Nagoya, 466, JapanView Profile , Naohiro Ishii Department of Intelligence and Computer Science, Nagoya Institute of Technology, Gokisocho Showaku, Nagoya, 466, Japan Department of Intelligence and Computer Science, Nagoya Institute of Technology, Gokisocho Showaku, Nagoya, 466, JapanView Profile Authors Info & Claims ICS '98: Proceedings of the 12th international conference on SupercomputingJuly 1998 Pages 157–164https://doi.org/10.1145/277830.277862Online:13 July 1998Publication History 0citation846DownloadsMetricsTotal Citations0Total Downloads846Last 12 Months0Last 6 weeks0 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

Intelligent information services as newly published books information have been carried out on the World Wide Web (WWW). In this study, an intelligent service system of books information are developed by applying the changing multiple convergence algorithm, which is based on the inductive learning method. The method developed can deal with the user's changing interests. To overcome the slow learning of this method, a pre-learning processing is introduced.

One of the difficult problems in database integration is to find and resolve the semantic heterogeneity of the component databases. In this paper, we present a methodology for database integration by using an ontology. The ontology is mainly used to help the designers to understand the semantics of database objects. Anyone who wants to share his database with others should provide an export schema of the database and a mapping between the schema and one or many standard ontologies. According to this mapping, the designer can easily find and resolve the semantic heterogeneity of the component databases and build an integrated schema.

Sunday's OM algorithm can reduce the number of character comparisons by making use of information of character distribution in an alphabet. Smith's adaptive algorithm uses dynamic statistics to reduce comparisons and its performance is close to that of the OM algorithm in the number of character comparisons. Smith's algorithm has the advantage of language independence. Its drawback is that it runs slowly because of maintaining an ordering list. This paper presents an improved adaptive method which dispenses with the ordering list. This method treats the pattern as a circle, and first compares the mismatched character in the last checking operation. This methods is slightly worse than Smith's method in the number of character comparisons, but it much better in the running time. © 1998 John Wiley & Sons, Ltd.

This paper addresses the following scheduling problem: given a precedence graph with communication costs and a machine architecture with different types of processors, construct a schedule that runs on the given architecture at the minimum possible execution time. The main contributions are: Firstly, we present a static scheduling algorithm that keeps processors idle for future important tasks and fills idle time slots incurred due to interprocessor communication. Secondly, to evaluate the effectiveness of the algorithm, we develop a lower bound on the length of a optimal schedule as a yardstick. Experiments show that this new approach produces better schedules and takes much less compile time.

A photometric method called point source illuminating stereo is proposed for determining the 3D shape of an object from multiple shading images under point light source illumination. When the surface is a perfect diffuser with uniform reflectance, an algorithm for the determination of 3D shape with positions is developed by using the method of least squares and relying on the principle of the monocular vision and the inverse square law for illuminance. In the proposed method, the number of the necessary images is four for the general surface, and can be reduced to three for the continuous surface.>