Search

As a replacement for the lecture-driven teaching of the past, we are aiming to establish an individually initiated, efficient independent study environment using a computer network. This paper focuses in particular on the collection of Web information needed by researchers, based on efficient paper retrieval and collection Web sites using XML. XML can handle data on the Web more flexibly than the present HTML, and research into its practical applications is progressing rapidly. In the artificial intelligence field, ontology engineering for authoring tools to be used in creating teaching materials is under way. In this context, because XML has been defined by the W3C, it is an accessible standard, and application examples have appeared. However, few of them focus on retrieval. We describe a “paper retrieval system” for efficient retrieval, constructed only in HTML on the basis of XML, RDF, and XHTML. © 2002 Wiley Periodicals, Inc. Syst Comp Jpn, 34(1): 21–28, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10014 [ABSTRACT FROM AUTHOR]

This paper discusses a method in which periodic motion, such as biped walking, is acquired by the framework of reinforced learning. For an object with multiple degrees of freedom, it is in general difficult to acquire the policy by reinforced learning. There must be a framework of learning which is matched to the motion task under consideration. From this viewpoint, this paper proposes “dynamic policy,” that is, a policy with internal state and dynamics. Its learning procedure is then considered. When periodic motion such as biped walking is to be acquired by learning, dynamical entrainment to the control object can be utilized if the policy has internal dynamics. Furthermore, by providing a policy with internal state, the controller is made robust to time delay and noise in the sensor inputs. The use of the policy gradient is discussed as a method of learning the policy without explicitly considering its internal state. As an example, dynamic policy is applied to a three-link biped walking robot model. The dynamic policy in this example is realized by a feedback controller based on a neural oscillator and sensor inputs. By using the policy gradient, appropriate rules for sensor feedback to the neural oscillator are learned by trial and error. It is shown by simulation that a dynamic policy realizing biped walking is acquired. The method is further applied to a five-link biped walking robot in order to investigate the possibility of extension to a system with multiple degrees of freedom. © 2007 Wiley Periodicals, Inc. Syst Comp Jpn, 38(4): 25–38, 2007; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20441 [ABSTRACT FROM AUTHOR]

This paper devises a novel kernel function for natural language processing tasks. The new kernels, called Hierarchical Directed Acyclic Graph (HDAG) kernels, directly accept graphs whose nodes could contain graphs. HDAG data structures are needed to fully reflect the syntactic and semantic structures inherently possessed by natural language data. In this paper, we define the kernel function and describe how to achieve efficient calculation. Experimental results demonstrate that the proposed kernels are superior to other kernel functions, sequence kernels, dependency structure kernels, and bag-of-words kernels. © 2006 Wiley Periodicals, Inc. Syst Comp Jpn, 37(10): 58–68, 2006; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20485 [ABSTRACT FROM AUTHOR]

Turbo codes are known as a class of error-correcting codes which have high error-correcting performance with efficient decoding algorithm. Characteristics of the iterative decoding algorithm have been studied in detail through a variety of numerical experiments, but theoretical results are still insufficient. In this paper, this issue is addressed from the information geometrical viewpoint. As a result, a mathematical framework for analyzing turbo codes is obtained, and some of the fundamental properties of turbo decoding are elucidated based on this framework. Recently, it has been pointed out that the turbo decoding algorithm is related to the decoding algorithm of low-density parity check codes, the computation method of Bethe approximation in statistical physics, and the belief propagation algorithm of Bayesian networks. The mathematical framework given in the present paper can also be used to analyze these wide classes of iterative computation methods, and hence represent a new analysis tool. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 36(1): 79–87, 2005; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10359 [ABSTRACT FROM AUTHOR]

This paper proposes a knowledge model for large-scale knowledge structurization of documents. When an elementary concept such as input for information retrieval is handled as an object, generally a structure composed of the relations in the set of objects is observed. The proposed model generates objects by statistical machine learning of the occurrence patterns of words in the word set and quantifies the relations among the objects by matching to the occurrence patterns. The idea is to approach knowledge structurization in order to achieve systematic understanding by relating pieces of existing knowledge by quantifying the relations among objects as elementary concepts. In this paper, the proposed model is applied to problems in the life science field, where there is a strong requirement for new knowledge with a comprehensive overview, and its application to gene function analysis is discussed. Experimental results on the extraction of disease-related genes as relations between genes and a disease, and on the estimation of gene functions combined with knowledge about organs, are presented and the effectiveness of the approach is demonstrated. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 35(8): 21–30, 2004; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10670 [ABSTRACT FROM AUTHOR]

Learning from examples in the presence of overwhelmingly many irrelevant variables has been studied in the field of machine learning theory, where the learner is required to derive as succinct hypotheses as possible. In this paper, we consider the following computational learning problem: Given a set of negative examples in n Boolean variables, the learner is required to find O(log n) relevant variables whose conjunction gives a consistent hypothesis. This problem belongs to a complexity class called β2. We prove that this problem is equivalent, by probabilistic log-space Turing reduction, to the problem of finding satisfiable assignments for polynomial-size ${\bf \Pi}^{0}_{3}$-circuits with O(log2n) inputs. As a consequence, we can show that learning of O(log n)-length Boolean conjunctions from negative examples is neither β2-hard unless P ⊆ POLYLOG-SPACE, nor polynomial-time computable unless NP ⊆ DTIME(2O(√n)). In this paper, we also discuss the polynomial-time learnability of an O(log n)-length monotone Boolean conjunction from negative examples, when the examples are drawn from the uniform distribution. © 2001 Scripta Technica, Syst Comp Jpn, 33(1): 1–7, 2002 [ABSTRACT FROM AUTHOR]

This paper proposes an architecture of intelligent learning environments that aims at assisting students in acquiring the ability to apply physical laws to complex systems. This paper also presents its application to a spring-system learning environment. The learning process in the architecture consists of three steps: to interpret data collected by experiment, to predict results of experiment, and to verify the prediction by experiment. If a student encounters difficulties, the intelligent learning environment provides assistance such as prompting to collect additional data that are useful for interpretation, or providing a new physical system that is suitable for the student's state of understanding. To realize such tailored assistance for each individual student, we need a method to model a student's ability to apply rules. This paper proposes a new student model represented by a multidimensional space. Axes of the space correspond to measures of generality of a subject domain. We implemented an intelligent learning environment called “SpringMaster” to validate the proposed architecture. The result of its evaluation shows that students who learned by SpringMaster had better understanding than students who learned by real instruments. © 2001 Scripta Technica, Syst Comp Jpn, 32(8): 1–9, 2001 [ABSTRACT FROM AUTHOR]

In a radio communications network, all of the nodes cannot communicate with each other directly. Then packets are transferred from a source node to a destination node through several nodes. Therefore, we need to schedule transfer timing at each node, and communications routes must be assigned to minimize the total transfer time when many packet transfers are requested. This problem is divided into two problems: the communications route assignment problem and the scheduling problem. The former problem is subdivided into the communication route candidate extraction problem and the communication route selection problem. This paper first proposes an evaluation function (Cost) which gives the lowest limit of the total transfer time for the communication route assignment problem. Next this paper proposes a k-shortest route extraction procedure for the communication route extraction problem. This procedure is based on the k-shortest route algorithm. It prevents extraction of a route whose number of hops is more than the upper limit, sets an appropriate number of extraction routes, prevents loops, and prevents redundant route extraction. We also propose a greedy neural network procedure for the communication route selection problem. This procedure introduces the ω function into the operation equation of the neuron initial value setting based on the number of hops and a cost minimization term for the evaluation function. The procedure uses an appropriate termination condition for iterative computation. It also uses the first-order maximum neuron. Through simulations of 500-vertex communication network examples, the proposed procedure has the merits of high precision, short computation, and smaller number of computations in a region. © 1999 Scripta Technica, Syst Comp Jpn, 30(13): 52–60, 1999 [ABSTRACT FROM AUTHOR]

This paper proposes a method of retrieving landscape images in which an index is first generated automatically and then, to enable retrieval even when the index includes errors, a pixel-based object labeling technique for the landscape images is employed and similar images are retrieved by using object sketches. The automatic index generation requires recognition processing. Outdoor images are conventionally recognized by using a segment-based approach and heuristic knowledge and, as a consequence, problems of segment generation and knowledge acquisition arise. The proposed recognition method employs pixel-based, independent processing which does not use heuristic knowledge and, therefore, compared with conventional techniques for recognizing outdoor scenes, the process of index generation can be automated more easily. Generally, it is not possible to avoid errors completely in recognition results. With respect to this problem, this paper proposes a retrieval method using object sketches in which the user specifies object types and their position inside the image. Based on experiments, it is shown that, when this method is combined with a pixel-based object labeling technique, it is possible to retrieve with approximately the same degree of accuracy as when using a similar index built by humans, even when the recognition results are partially wrong. © 1999 Scripta Technica, Syst Comp Jpn, 30(12): 85–94, 1999 [ABSTRACT FROM AUTHOR]

This paper presents a method for designing neural networks using a genetic algorithm (GA) with deterministic mutation (DM) based on learning. The GA presented in this paper has a large framework including DM, which is performed on the basis of the results from neural network learning. It can achieve better convergence properties than traditional GAs. This framework is an evolutional system based on mutual interaction between DM and traditional genetic operators including stochastic mutation. It is also a model of transcription and reverse transcription in DNA. We show that the present method is better than conventional GAs with respect to convergence in learning. © 1998 Scripta Technica. Syst. Comp Jpn, 29(3): 10–17, 1998 [ABSTRACT FROM AUTHOR]

This paper considers cooperative and distributed problem-solving in an environment where similar problems are repeatedly posed, and discusses methods for plan reuse. In distributed planning including multiple agents, the agents generate the plan cooperatively from various viewpoints which results in a plan with high cost. The reuse of the plan realizes efficient planning by utilizing the plan for similar previous problems. The first feature to note in applying plan reuse to a multi-agent system is the problem similarity. Even if an agent decides that two problems are the same, another agent may consider a totally different goal or viewpoint. In this paper, it is shown that under certain conditions, the reuse of a plan is effective independently of the goals of other agents. A planning method including reuse is presented. Finally, it is shown experimentally that efficient planning can be realized. © 1997 Scripta Technica, Inc. Syst Comp Jpn, 28(6): 60–67, 1997 [ABSTRACT FROM AUTHOR]

This paper describes the framework of a system that can understand a textbook for machine operation, based on the model of an object world that can retain informations not only by symbolic representation but also by image. In any understanding of the description for machine operation, spatial information is important, and there must be a model that can retain the informations with natural reflection of the continuity and relative location, independently of the particular point of observation. Such a representation is difficult, dealing only with the symbolic knowledge representation used in the conventional language understanding system. From such a viewpoint, this paper notes the usefulness of the representation by the image and proposes the imagery object world model, which can utilize partially the properties of the image, using mathematical expressions in a two-dimensional real coordinate system. Mechanisms both to execute the simulation on the proposed model and to read the text using the simulation mechanism, are investigated. A system is constructed experimentally based on the proposed idea, which can construct the model for the input sentence, while executing a simulation and generating symbolic recognition expressions. With those functions, the identity or the similarity of the language expressions derived from the same object or event from different viewpoints, can adequately be decided. [ABSTRACT FROM AUTHOR]

Today, technological expertise gained from studies on natural languages is applied to various practical systems, including machine translation systems. English-Japanese machine translation, however, still remains difficult in the translation of lengthy sentences that contain multiple complicated expressions. Due to Limited information volume installed in current machine translation systems, such systems cannot handle the complicated sentences actually used in English. This paper discusses comparative structures, which frequently appear in long English sentences, clarifies characteristics of such structures through grammatical studies and the statistical analysis of the sentences contained in Corpus, and explains the development of comparative structure models and a new analysis method using such models. Having more flexibility than the conventional machine translation method, the new method proposed in this paper has been proved suitable for long sentences for which the grammatical rules and vocabulary installed in conventional machine translation systems are insufficient. English newspaper articles were analyzed with a commercial machine translation system and the new method. The result showed that the new method provided translations that were 41.6 percent more accurate than the conventional system translations. [ABSTRACT FROM AUTHOR]

This paper presents a neural network algorithm called SOLAR (Supervised One-shot Learning Algorithm for Real-valued inputs) that can complete learning by a single presentation of the real-valued inputs. The similarity matrix is introduced in SOLAR, which can measure the Euclidean distance between training instances represented by a real number. Based on the geometrical structure of the similarity matrix, the structure of the network, the connection weights, the learning parameters and the linear disciminant function are determined. Then, the learning is executed by a single presentation of the training set. The nonlinearly separable case by decomposing the set of supervisor signals into linearly separable sets. The main contribution of this paper is that SOLAR realizes a high-speed learning by a single presentation of the input data composed of real values and a large-scale complex set of data can be learned. This has been difficult in previous models. In addition, since the number of hidden units and the parameters are determined by the algorithm, there is no problem such as trial and error. [ABSTRACT FROM AUTHOR]

This paper describes the utterance generation system Linta-II, which was developed Aiming at the realization of the flexible human-robot interface. Linta-II is implemented on an autonomous mobile robot. Based on the information from the sensor installed on the robot, the utterance can be generated according to the situation. In Linta-II, the sensor information is acquired effectively using the attention mechanism. The attention mechanism is constructed as a set of entities executing simple symbol processings. It realizes two kinds of attentions, i.e., the involuntary and voluntary attentions. The involuntary attention has the function of paying immediate attention to the external event. The voluntary attention is activated by other modules such as the action control unit, and executes the top-down processing. Linta-II utilizes those two kinds of attention mechanisms, and can generate the utterance according to the situation, even if the rule to determine the sensor information to be referred to cannot completely be described. This paper demonstrates the effectiveness of the attention mechanism by presenting several examples of utterances by Linta-II. [ABSTRACT FROM AUTHOR]

The number of interconnections between cells in a neural network, which is proportional to square of the number, causes a problem in the structure of the network when the number is large. The fast Fourier transform is a method to speed up matrix operation by using a butterfly operation. This paper proposes a method of reducing the number of connections between cells in a neural network by using a multilayer structure which is based on a butterfly operation. Since the neural network inevitably a multilayer structure, it is necessary to employ a new learning method which replaces the conventional backpropagation method. To satisfy this condition, this paper proposes an inverse function method which generates an intermediate target of learning in an inverse direction from a teacher signal at the last state, and a bidirectional learning method which is an improvement of the inverse function method. The bidirectional learning method improves the learning effect, since the learning process is more stable than the inverse function method, and intermediate targets are generated taking into account the state of the output of each step. This is demonstrated by using a two-dimensional pattern transform problem. [ABSTRACT FROM AUTHOR]

In ordinary written Japanese, words are not separated by spaces. Therefore morphological analysis involves segmenting and tagging sentences. Since each sentence has a huge number of possible tagged segmentations, various criteria have been proposed for making plausible decisions. However, there are still no unified frameworks that incorporate various heuristics, and there has been no comparative evaluation of commonly used heuristics. This paper presents a clear framework to describe various heuristics, and an N-best algorithm for extracting optimal solutions. The time complexity of this algorithm is O(nNlog2(1 + N)), where n is the sentence length. The advantage of the N-best algorithm over the standard beam search algorithm is also discussed. This paper also presents a comparative evaluation of three major heuristics, and proposes a precise and portable rule-based heuristic. Estimation was done using the aforementioned algorithm and six criteria. The newly proposed heuristic is based upon the Extended Least Bunsetsu (Phrase) Number method. [ABSTRACT FROM AUTHOR]

This paper proposes a multimodule neural network for complex pattern association. In the associative memory discussed in this paper, it is assumed that the memorized pattern can be divided into a certain number of attributes. An associative module is allocated to each of the divided attributes. The network is constructed by connecting those modules. Then, the relations among the attributes are memorized in the connections among the modules, as the mutual relations among the modules. By this structure, an association function is realized, in which given a partial information concerning some of the attributes, all information related to the input can be retrieved by tracing the relations among the attributes. Furthermore, by providing a module for each attribute, similar patterns that differ only in a very limited number of attributes, can be memorized and associated and separate patterns. By combining the modules of the associative memory so that multiple patterns can be produced sequentially as the output for an input pattern, a many-to-many association can be realized by the network composed of such modules, where multiple candidate patterns are recalled from various combinations of partial information. [ABSTRACT FROM AUTHOR]

This paper presents an expert system for automatically recognizing handwritten characters of dollar amounts on personal checks. The dollar amounts composed of numerals and marks "-," "x x," etc., are freely handwritten in various formats that individuals favor, inside the amount field. Then frequently there is contact between not only horizontally adjacent characters but also vertically adjacent ones. Recognition of such unconstrained character strings handwritten by the public whose formats are undecided beforehand is a problem in optical character recognition. The system developed in this paper has the ability to segment the dollar amount image into individual characters automatically and to recognize them accurately. The recognition process of the dollar amount consists of the following steps: divide a digitized image into groups, called a "block," of line elements connected with one another, give attributes to each block by using the knowledge on arrangement patterns of characters; and segment the block into individual characters according to the attributes and recognize them. It is suggested that this system will enable the saving of considerable operator labor in check processing. [ABSTRACT FROM AUTHOR]

This paper introduces an advanced explanation capability for the intelligent tutoring system (ITS) to assist students in understanding objects such as electric circuits designed by humans. In general, In an object can be explained in various. To advance the explanation capability, it is necessary to clarify how these explanations are used properly according to some tutoring goals in the domains. Moreover, the design of ITS based on such a capability requires: (1) classification of explanations based on the tutoring goals, and (2) development of a facility to generate the classified explanations. First, this paper defines object understanding by using two concepts, i,e, viewpoint and object model, and classifies explanations by regarding the defined object understanding as a goal. Next, the explanation facility, Explanations Structure model (EXSEL) is proposed. EXSEL has a mechanism to generate an explanation structure that is regarded as a resource of explanation and is designed in due consideration of being applied to ITS. The definition of object understanding the classification and EXSEL are applicable to the objects explainable at three abstraction levels: structure, behavior, and function, Providing the mechanisms to operate EXSEL according to tutoring requisites, such as the student's understanding, it is possible to realize various ITSs to support greatly object understanding. [ABSTRACT FROM AUTHOR]

Based on the context-free tree grammar (CFTC) proposed by W. C. Rounds [7J, the concept of pushdown tree automata (PDTA) was newly introduced. Such a PDTA was a top-down PDTA. By contrast, it is desired to introduce its dual concept, i.e., a bottom-up PDTA. With such a background, this paper proposes a new concept of bottom-up PDTA and it is shown that one of the fundamental properties of the pushdown automaton (PDA) such as the final state acceptance, equivalent to the empty stack acceptance, is also applicable to the bottom-up PDTA. Next, we show that the tree language accepted by the bottom-up PDTA is always accepted by a single-state bottom-up PDTA. It is shown also that the tree language accepted by the single-state bottom-up PDTA is accepted by the single-state linear stack bottom-up PDTA, where the stack structure is restricted to be linear. This paper discusses only the fundamental properties of the bottom-up PDTA. We hope to give a basis for the theoretical development of tree language and related fields. [ABSTRACT FROM AUTHOR]

This paper describes the outline and implementation of the declarative debugging system DDS. The purpose of DOS is to provide the following debugging system for the functional program without a side effect: (1) The programmer specifies the result of execution and the intermediate result expected to the program (defines in a declarative way, and the system detects the bug automatically based on that specification; (2) an efficient guide is provided to the programmer so that the required input is less and the input can easily be given; and (3) the system has an efficient bug detection mechanism in terms of the memory capacity and the computational complexity. It is an interactive system which can be applied to a wide range of programs. This paper proposes a mechanism called filter, which gives an efficient realization of DDS. The filter is a mechanism which realizes the tabulation technique by a small-capacity memory, by extending the technique from the viewpoint of the bug detection. The filter is a mechanism which selectively retains the instances needed in the bug detection and controls the function call of the program using the retained data. An experimental system of DDS is constructed on VAX/VMS using Common Lisp. Debugging examples are shown, indicating the usefulness of DDS. [ABSTRACT FROM AUTHOR]

This paper proposes a method of constructing a multilayered neural network, using the multiple logistic model (MLM). The model is a nonlinear multivariate analysis considering the output logistic function of each unit, which is used in the back- propagation method (BP). The idea can be applied directly to the determination of the multilayered neural network structure. The model can also be utilized as a systematic method to introduce such information as pattern distribution into the neural network structure. Considering the speaker-independent vowel recognition as the problem, this paper compares the results by the proposed method (MLM), the construction by the linear multiple regression analysis (MRA), the learning by BP with the weight being defined at random as the initial value, and the learning by BP with the initial weight determined by MLM or MRA. It is seen as a result that the recognition rate is the best when BP is applied after introducing the speaker distribution information by the proposed method. It is seen also that the computation time is reduced compared with the BP, with the initial weight being defined at random. [ABSTRACT FROM AUTHOR]

This paper proposes a neural network model, which performs deterministic, parallel pattern matching. The pattern is recognized by matching an input pattern containing noise and a template constructed through learning. In this paper, it is assumed that no feature detection has already been accomplished. We consider a circuit structure which merely compares a template with the noisy feature distribution pattern that appears at the output of the feature detector. This pattern matching method is based on our previous work using neural networks to determine the correspondence between stereo images. The input from the left eye, for example, corresponds to the feature distribution pattern, and the input from the right eye corresponds to the template. The difference in the positions of features in the template and the input pattern corresponds to the binocular parallax. The proposed model is composed of a pattern matching layer, a minimum distance detection layer (which selects the template close to the input feature distribution pattern, and feeds it back to the pattern matching layer for pattern matching) and the recognition layer which classifies the matched template. The results of our computer simulation indicated that perfect matching was achieved for inputs having feature distribution pat- terns corresponding to those used during learning. Since sequences of labels or character strings can be considered to be feature distribution patterns, the proposed network can be applied to a wide range of problems related to pattern matchings [ABSTRACT FROM AUTHOR]

We have been studying the parallel coordinated computing system composed of several AI workstations ELIS, as well as the knowledge processing on that system. From the viewpoint of the modularity and the hierarchy, an object-oriented approach is adopted as the fundamental paradigm. In the knowledge representation on such a system, the object for representing the knowledge must be modified in a dynamical way since the knowledge always changes by itself. In this paper, such a modifiable object is called a dynamic object. In the traditional system, however, the object has been considered as static, and instance variables, for example, cannot be added or deleted dynamically during execution. The dynamic object is realized by modifying the object mechanism of TAO, which is the kernel language of ELIS. This paper proposes a dynamic object-oriented programming based on this dynamic object, which is suited to the representation of the changing knowledge. It is indicated that from the standpoint of the dynamic knowledge representation, the dynamic object-oriented programming is more natural and more efficient than the traditional object-oriented programming. [ABSTRACT FROM AUTHOR]

This paper describes a stereo-matching algorithm which works effectively even for a scene with a large number of discontinuities such as in the case of multiple interlaced surfaces. Prazdny (61 pointed out that the discontinuity can be handled by dissolving the ambiguity based on the facilitation by the similarity, rather than employing the discontinuity of disparity as the penalty. However, his method has a problem in that the ability to dissolve the ambiguity is decreased in correspondence to the reduction of scene restriction. In this paper, to determine the dispartity field maximizing the facilitation of similarity, several matching rules and matching primitives are combined, and the result is integrated into an energy-minimum model of a network similar to Hopfield's neural network [7]. This resulted in a strong disambiguation power. Especially, we propose a new multiresolution processing method robust against discontinuity, in which processing of the different levels is in parallel. Only the information data concerning the image features corresponding to the same physical entity are propagated from the coarse to fine levels. By the reasonings in this paper, a natural extension is derived for the Marr-Poggio first stereo algorithm [2]. An experiment was made for a scene through a fence, and it was verified that the interactions built in the network work effectively. [ABSTRACT FROM AUTHOR]

This paper aims at a constructive description of which neural network model can represent the human problem-solving process and how the process is related to the various human memories by what context. From such a viewpoint, a neural network model is constructed for the representation of the problem-solving process. In this paper, a model for the addition process is constructed as the first step, using the associative memory model HASP (human associative processor) proposed by one of the present authors. The model is composed of the following components: the procedure memory, which stores and retrieves the procedures of the addition; semantics memory, which stores the knowledge about the addition such as 2 + 2 = 4; the working memory, which temporarily stores the external and internal information; and the goal memory, which stores the problem-solving process as a chain of subgoals, and manages the flow of procedures from one higher level in the control hierarchy. In this study, a computer simulation was performed for the addition processes for several numbers with multiple digits, and it is shown that the addition process can be represented. Then the computer simulation is made for the generation of four kinds of bugs in the addition. It is shown that the bug generation can be represented by the model. [ABSTRACT FROM AUTHOR]

In machine learning tasks on large-scale datasets, the labeled data essential to the classification are not always sufficient, which degrades the learning accuracy. Meanwhile, unlabeled data are always abundant. Hence, semisupervised learning which uses both unlabeled and labeled data to improve the learning accuracy is currently of great interest. In this paper, we use a graph to represent the underlying distribution of both labeled and unlabeled data and split it by using multiway cut to classify unlabeled data. Additionally, we propose a graph-based feature selection algorithm to improve the learning accuracy of our graph-based semisupervised learning algorithm. In our algorithm, we first propose an evaluation criterion for the attribute relevance using the graph density. Then, we extract the relevant attribute subset by finding the clique on the graph where each vertex stands for the attribute and each edge stands for the relevance of a feature pair. © 2007 Wiley Periodicals, Inc. Syst Comp Jpn, 38(9): 32–43, 2007; Published online in Wiley InterScience (www.interscience. wiley.com). DOI 10.1002/scj.20757 [ABSTRACT FROM AUTHOR]

This paper presents a new false-name-proof multi-unit auction protocol called the Greedy ALlocation (GAL) protocol. Internet auctions have become an integral part of Electronic Commerce and a promising field for applying agent and Artificial Intelligence technologies. Although the Internet provides an excellent infrastructure for executing auctions, the possibility of a new type of cheating called false-name bids has been pointed out. A false-name bid is a bid submitted under a fictitious name. A protocol called the Iterative Reducing (IR) protocol has been developed for multi-unit auctions and has proven to be false-name-proof, that is, using false-name bids is useless. For Internet auction protocols, being false-name-proof is important since identifying each participant on the Internet is virtually impossible. One shortcoming of the IR protocol is that it requires the auctioneer to carefully predetermine the reservation price for one unit. Our newly developed GAL protocol is easier to use than the IR, since the auctioneer does not need to set the reservation price nor any other parameters. The evaluation results show that the GAL protocol can obtain a social surplus that is very close to Pareto efficient. Furthermore, the obtained social surplus and seller's revenue are much better than those of the IR protocol even if the reservation price is set optimally. © 2006 Wiley Periodicals, Inc. Syst Comp Jpn, 37(13): 89–98, 2006; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10664 [ABSTRACT FROM AUTHOR]

In this paper, we applied various optimal solution search methods from the field of artificial intelligence to multiple sequence alignment, which is a basic problem of sequence analysis in the field of bioinformatics, to conduct performance evaluation experiments. The techniques applied were A*, Enhanced A*, Iterative Deepening A* (IDA*), Recursive Best First Search (RBFS), MREC, and Partial Expansion A* (PEA*). Although MREC had been considered ineffective for the 15-puzzle since it could only reduce the number of states by a mere 1% compared with IDA*, the results of the performance evaluation experiments showed that extension factor (EF-1α) multiple sequence alignment, which could previously be solved for only up to 8 sequences, could be solved successfully by MREC for up to 11 sequences, and that MREC could be executed quickly and with little memory. Experimental results using the Benchmark Alignment dataBASE (BAliBASE) show that MREC obtained the optimal solution the fastest and using the least amount of memory among the currently available techniques. [ABSTRACT FROM AUTHOR]

In the field of multiagent systems, some methods use the policy gradient method for behavior learning. In these methods, the learning problem in the multiagent system is reduced to each agent's independent learning problem by adopting an autonomous distributed behavior determination method. That is, a probabilistic policy that contains parameters is used as the policy of each agent, and the parameters are updated while calculating the maximum gradient so as to maximize the expectation value of the reward. In this paper, first, recognizing the action determination problem at each time step to be a minimization problem for some objective function, the Boltzmann distribution, in which this objective function is the energy function, was adopted as the probabilistic policy. Next, we showed that this objective function can be expressed by such terms as the value of the state, the state action rule, and the potential. Further, as a result of an experiment applying this method to a pursuit problem, good policy was obtained and this method was found to be flexible so that it can be adapted to use of heuristics and to modification of behavioral constraint and objective in the policy. © 2006 Wiley Periodicals, Inc. Syst Comp Jpn, 37(10): 101–109, 2006; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20248 [ABSTRACT FROM AUTHOR]

Chiefly for applications to robots that operate in a real environment, it is desirable to construct a compact vision system with real-time properties and adaptability to the environment. The silicon retina is an image sensor considered from such a viewpoint, in which the preprocessing mechanism in the biological retina is simulated in an analog CMOS integrated circuit. This paper proposes a computation scheme which estimates the optical flow (the moving velocity field of an object that is in relative motion with respect to the imaging system) robustly even in images of a real environment, focusing on the function and structure of the silicon retina. In the proposed method, the edge of the moving object is extracted from the response image on the silicon retina. By applying local computation to the result, the optical flow can be derived efficiently even in a real environment. An experiment is performed using the silicon retina and a computer based on the proposed method, and satisfactory results are obtained. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(11): 12–23, 2005; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20343 [ABSTRACT FROM AUTHOR]

This paper proposes a neural network model which receives visual inputs from a robot moving freely in a room, and extracts its position and direction information separately. The model has three-dimensional structure in which two-dimensional neural fields (unit planes) are arranged in a toroidal form. The learning algorithm takes a form of neural gas in each unit plane and self-organizing map (SOM) in the toroidal direction. It is shown by computer simulation that position and view direction of the robot can be extracted from visual inputs, and the effectiveness of the learning algorithm is verified. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(11): 1–11, 2005; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20349 [ABSTRACT FROM AUTHOR]

Self-Generating Neural Networks (SGNN) have the feature of fast processing by automatically constructing a self-generating neural tree (SGNT) from a given training data set. The prediction accuracy of SGNNs for chaotic time series prediction is improved by ensemble averaging of various SGNNs. However, the computation time increases with the number of SGNNs on a single processor. In this paper, we investigate improvement of the prediction accuracy and parallel efficiency of ensemble SGNNs for chaotic time series prediction problems on an MIMD parallel computer. We allocate each SGNN to one processor. Our results show that as the number of node processors increases, a continuing improvement of prediction accuracy is obtained for all problems, while maintaining the high-speed processing performance of the single SGNN. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(10): 82–92, 2005; Published online in Wiley InterScience (www.interscience. wiley.com). DOI 10.1002/scj.10297 [ABSTRACT FROM AUTHOR]

This paper proposes to apply machine learning techniques to the task of combining outputs of multiple LVCSR models, where, as features of machine learning, information such as the models which output the hypothesized word, its part-of-speech, and its syllable length are useful for improving the word recognition rate. Experimental results show that the combination result outperforms several baselines including model combination by voting such as ROVER in the word recognition rate. Furthermore, unlike model combination by voting, word recognition rate of model combination by machine learning is not damaged even in the case where only the minority of the participating models perform well in the word recognition task. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(10): 9–15, 2005; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20340 [ABSTRACT FROM AUTHOR]

In a layered neural network, the error backpropagation method is generally used as the supervised learning procedure for the hidden layer, which cannot be observed from the outside. In order to apply the method, however, the output function of the neuron model must be differentiable. This paper proposes a supervised learning method for the hidden layer neurons in which the teaching signal is calculated for the hidden layer neurons, by utilizing the learning rule for the connection weight and the duality in the output layer neuron. The method is applicable so long as the neuron model contains duality, and it does not require that the output layer neurons or the hidden layer neurons be differentiable. As an example of a case in which the error backpropagation cannot be applied, a perceptron composed of neurons with a step output function is considered. The proposed method is applied, and the learning rule for the whole network is constructed. The XOR problem was actually learned by the network, and the same learning success rate was obtained as in the error backpropagation method for a perceptron composed of neurons with a sigmoid output function. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(9): 34–42, 2005; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20206 [ABSTRACT FROM AUTHOR]

In this paper, we propose a method to improve the accuracy of classifiers by replacing the connection between the output layer and the immediately preceding hidden layer with an optimal linear transformer. This approach is intended to improve the performance of a breast cancer image diagnosis assistance system. The proposed classifier is composed of a three-layer MLP (multilayer perceptron) and a Mahalanobis classifier. The MLP has only one output unit, and produces output for two categories. If it is assumed that the value from the hidden layer immediately preceding the output layer forms a multivariable normal distribution for each class, that is, a Gaussian distribution, then the optimal linear transformer is a classifier based on the generalized Mahalanobis distance. Thus, the optimal classification is realized in the MLP after learning in which the generalized Mahalanobis distance with the hidden layer immediately preceding the output layer as the input is examined, and classification is performed on the basis of the likelihood. The proposed breast cancer image diagnosis assistance system, the system using only the conventional Mahalanobis classifier, and the system using only the conventional MLP classifier are compared. The best results are given by the proposed method, and it is shown that the performance of the breast image diagnosis assistance system can be improved. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(9): 65–76, 2005; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20173 [ABSTRACT FROM AUTHOR]

In order to achieve fast parallel processing operation by a neural network, it is desirable to implement all neurons in parallel in hardware rather than to realize operation by serial operation in software. The authors have proposed a pulsed neural network based on ΔΣ modulation which is suited to the hardware implementation of digital circuits. The proposed neural network can be realized on a small circuit scale, since the variable can be represented by a ΔΣ-modulated 1-bit pulse signal. Highly accurate operation can be realized even though only 1 bit is used. This paper proposes a hardware implementation of the above neural network. A neural network with built-in GHA learning rule is hardware implemented on an FPGA, and its operation is investigated. The proposed neural network and CPU are implemented on an FPGA and are compared in terms of the number of bits that can be processed in unit time per unit of circuit scale. The proposed scheme achieved an evaluation score 200 times as great as a software realization on a CPU. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(9): 14–24, 2005; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20235 [ABSTRACT FROM AUTHOR]

In this paper the authors propose a method for high-accuracy recognition of large character sets using a new combination of a statistical method and neural networks. In their method, a hierarchical structure that has several neural networks arranged in a line after the statistical method is used. First, recognition using a statistical method is performed, and this represents the final result if the top candidate does not belong to a predefined set of similar characters. If it does, then the input character is discriminated in a neural network which designates the top candidate by determining the similar characters. The results are output as final results. The basic idea of this method is the functional division of a statistical method and neural networks, and the use of a neural network as determined by a statistical method. The results of recognizing 3201 character types including JIS-1 Kanji showed an improvement in the correct recognition rate due to the combined use of a statistical method and neural networks, thereby demonstrating the validity of the authors' approach. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(9): 97–107, 2005; Published online in Wiley InterScience (www.interscience. wiley.com). DOI 10.1002/scj.20330 [ABSTRACT FROM AUTHOR]

A forward-propagation learning scheme has been proposed to acquire inverse models of controlled objects in multilayered neural networks. This scheme is quite different from back-propagation learning rule. The algorithm of the forward-propagation rule consists of two stages. One is the estimation of the instruction signal at each layer by the Newton-like method, and the other is the updating of the connection weights by linear multiple regression. In this scheme, convergence of learning has been faster than other schemes based on backpropagation rule. However, the problems arise that complex parameters must be set for learning and the learning process is too complex and sometimes stops. This paper proposes to use a method of recursive least squares in the forward-propagation rule. The effectiveness of the proposed method is confirmed by computer simulation for the learning of the inverse dynamics model for a two-link arm. © 2005 Wiley Periodicals, Inc. Syst Comp Jpn, 36(8): 71–80, 2005; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20237 [ABSTRACT FROM AUTHOR]

With the development of our highly information-oriented society, there is an increasing demand for large-scale and high-level information processing. Toward this goal, studies have sought to create a new computation principle having an information processing ability exceeding the existing Neumann-type computer, such as the creation of a new computation theory or the integration of the frameworks of existing computation theories. As one such approach, quantum neural computing is considered to be interesting, which integrates neural computing and quantum computation. This paper constructs the feed-forward neural network, which is widely used in practice, based on the qubit neuron model. The 4-bit parity-check problem and the general function identification problem are considered. The performance is compared to the feed-forward network based on the conventional neuron model, and it is shown that the proposed model has a higher performance than the conventional model, using the learning diagram composed of convergence rate and the number of learning iterations. The reason for the better performance is also discussed. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 35(13): 43–51, 2004; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10342 [ABSTRACT FROM AUTHOR]

It is known that the Hopfield-type neural network can solve the combinatorial optimization problem with a high speed. In order to realize that advantage, however, the energy function must be constructed for each problem, and there is no general framework to be applied to various problems. This paper notes that the logic programming language Prolog can be represented as the logical operation of facts and the rules, and proposes an algorithm, which converts the logical operation network to the Hopfield-type neural network. Using this conversion algorithm, the power of the Hopfield-type network as an optimization machine can be utilized in the logic inference described by Prolog. Then, it is expected that the range of its application will be enlarged greatly. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 35(13): 63–70, 2004; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10102 [ABSTRACT FROM AUTHOR]

The cerebellar model arithmetic computer (CMAC) is a kind of neural network that is capable of high-speed learning. In the use of the CMAC, various parameters must be defined, but it is difficult to set the parameters adequately. The pseudo-bacterial genetic algorithm (PBGA) is a kind of evolutionary computation which offers local search. This paper considers adequate determination of the location of the boundary line, which is one of the parameters of the CMAC. For that purpose the PBGA is applied, and PBGA/CMAC, which realizes the efficient search of the boundary line of the CMAC, is proposed. Furthermore, since PBGA/CMAC requires a large amount of computation, a hardware implementation is considered in order to improve the computation speed. The computation involved in the learning rules of the CMAC is divided into four stages, and the structure is implemented as a pipeline. The pipeline implementation produces read-after-write hazards for the granular cell values. However, it is shown that the difference in performance from conventional CMAC learning rule is small, indicating the possibility of achieving high-speed learning. The speed of PBGA/CMAC hardware is improved to 140 times the speed of operation by software. In order to present a guide for PBGA/CMAC hardware, the hardware volume is evaluated, and it is shown that it can be implemented with precision sufficient for practice. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 35(11): 14–23, 2004; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10437 [ABSTRACT FROM AUTHOR]

A complex neural network is obtained from an ordinary network by extending the (real-valued) parameters, such as the weights and the thresholds, to complex values. Applications to problems involving complex numbers, such as communications systems, are expected. This paper presents the following uniqueness theorem. When a complex function is given, the three-layered neural network that approximates the function is uniquely determined by a certain finite group, if it is irreducible. The above finite group specifies the redundancy of the parameters in the complex neural network, but has a structure which is different from that of the real-valued neural network. The order of the finite group is examined, and it is shown that the redundancy of the complex-valued neural network is an exponent multiple of the redundancy of the real-valued neural network. Analysis of the redundancy is important in the theoretical investigation of the basic characteristics of complex-valued neural networks, such as the local minimum property. A sufficient condition is derived for the given three-layered complex-valued neural network to be minimal. The above results are shown, in essence, by extending the approach of Sussmann for real-valued neural networks. © 2003 Wiley Periodicals, Inc. Syst Comp Jpn, 34(14): 54–62, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10363 [ABSTRACT FROM AUTHOR]

In the application of layered neural networks to practical problems, a high generalization power is required. This paper discusses a method of improving the generalization power of neural networks. The knowledge of the object to be learned is assumed to include the fact that the output function of the object to be learned remains invariant for a certain range of input pattern variation. An attempt is made to improve the generalization power by reflecting this invariant property in the weighting of the neural network. It is shown for the case in which the variation of the input pattern can be represented by a linear transformation that a sufficient condition for the neural network to have such invariance is that the linear dependency constraint is introduced into the weight expression. A learning process is proposed in which this kind of constraint for the weight expression is introduced into the evaluation function as an additional term. The proposed method can be considered as a generalization of the method in which deletion learning methods such as the weight decay method and the structural learning method are included as special cases. There has been discussion of the relation between the generalization power and the VC dimension. The improvement of the proposed method can be evaluated, by introducing a linear constraint into the weight expression, based on the reduction of the VC dimension. Lastly, results are presented for an experiment in which the proposed method is applied to the character recognition problem. © 2003 Wiley Periodicals, Inc. Syst Comp Jpn, 34(14): 83–91, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.1224 [ABSTRACT FROM AUTHOR]

The problem of the pulse mode neural network is that the synapse weight is bounded between +1 and -1. In order to enhance the weight range, this paper proposes a new weighting method that is a combination of pulse density modulation and pulse position modulation (PPM). The feasibility of the proposed method is verified by experimental MNN with on-chip learning. A modified back-propagation algorithm is used for the on-chip learning. The experimental results show that the proposed MNN architecture has the ability to handle high-resolution classifying problems and good on-chip learning capability. © 2003 Wiley Periodicals, Inc. Syst Comp Jpn, 34(13): 36–46, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10474 [ABSTRACT FROM AUTHOR]

Studies of immune systems include the work of Jerne on the equilibrium state of a network modeling an immune network. The authors have previously studied the information processing capabilities of immune systems and have proposed models capable of acquiring immune memory based on immune responses in an immune system. These models are based on the interactions between cells constituting an immune system, and these interactions are controlled by the state of a network formed by the cells in advance. However, in these models, cell groups are constructed by repetitions of immune responses in the immune system of a real biological organism and the fact that the immune responses are made by cell groups is not taken into account. Therefore, in this paper an immune system model involving an interactive process of network generation by cell groups generated on the basis of the clonal selection theory is proposed. The immune properties of the proposed network are shown by simulation and its effectiveness is demonstrated by applying it to a pattern recognition system. © 2003 Wiley Periodicals, Inc. Syst Comp Jpn, 34(12): 56–63, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10243 [ABSTRACT FROM AUTHOR]

This paper proposes a new neural network (the concepts-of-sentences associative neural network, CSANN) which performs memory and recall of sentences represented by a semantic network, as an approach toward natural language processing. The semantic network is represented in terms of the triples representation of concepts. The triples representation of concepts consists of concepts linked by relations in sentences represented by natural language. CSANN handles sentence data, which are difficult to handle since the data size is not unique, by decomposing the data into triples representations of concepts and integrating them into the semantic network format. It is verified by computer simulation that the proposed network can store sentences and recall sentences even if the input information is incomplete. Thus, the effectiveness of the proposed network in natural language processing is demonstrated. © 2003 Wiley Periodicals, Inc. Syst Comp Jpn, 34(11): 43–54, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10476 [ABSTRACT FROM AUTHOR]

With the widespread use of the Internet in recent years, documents are increasingly handled by electronic means. However, due to physical and time limitations it is difficult to review all of these documents. To allow selection of desired documents, it is useful to prepare a summary representing the content of each document. Consequently, this paper proposes a panoramic view system which extracts important sentences by a unique method. The system surveys the whole text and detects the keywords that express the features of the text. The feature keywords are extracted as words closely related to the main topic of the text and help to extract important relevant sentences. © 2003 Wiley Periodicals, Inc. Syst Comp Jpn, 34(11): 81–90, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.1215 [ABSTRACT FROM AUTHOR]

This paper presents a fast line extraction method using the line segments found in digital images. A digital line can be decomposed into line segments, which consist of continuous edge pixels, in four directions. The directions of the line segments are varied and limited by the relationship between the line segments and the slopes of analog lines. The proposed method attains high speed and accuracy by tracking each line segment in the same direction which comes from the relationship. Experimental results are included to show that the proposed method can achieve high accuracy with a large reduction in the computation time and has robustness in the presence of noise. © 2003 Wiley Periodicals, Inc. Syst Comp Jpn, 34(10): 76–89, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.10124 [ABSTRACT FROM AUTHOR]