In this paper, we propose a new method for image analysis by using the natural language information extracted from the explanation text of image data. First, we describe the method of extraction of color information from the explanation text. Then, we describe how this color information is used for the extraction of objects from the image data. Our experiments with pictures and text in pictorial book of flora showed the effectiveness of our method.

In this paper, we propose a method of acquiring models of 3D objects in a class. In this method, quantitative and qualitative models of an object are acquired from range images within the same class. The quantitative model is obtained as a typical shape and a shape distribution of the class, whereas the qualitative model is linguistic description of shapes. A primitive shape is expressed by superquadrics which can represent geometric property of the 3D shape by some parameters. A compound shape is considered as combination of the primitive shapes, and is represented by geometrical positions of connected parts. This representation makes it possible to place the 3D shape in the parameter space. Since similar shapes are located close in the parameter space, a quantitative model of the class is acquired as parameter distribution in the parameter space. Thus, the typical shape is obtained by averaging parameters of the representation, and the shape distribution is acquired as variances of the parameters. Then, linguistic description of geometrical property of an object is obtained by interpreting the parameters of the shape representation. This interpretation enables acquisition of a qualitative model flom the quantitative model. As a result, the experiment demonstrates that the proposed method provides valid models for both simple and compound shapes.

Distributed/concurrent systems consist of many concurrent processes, and behave on strict timing conditions. It is important to verify the timing conditions of distributed/concurrent systems. In this paper, we propose formal verification method based on Binary Decision Diagrams(BDDs) for distributed/concurrent systems as follows: (1) We encode timed Kripke structure into Binary Decision Diagrams corresponding to patterns of timing constraints. (2) We verify systems by symbolic model-checking, which consists of both inverse image computation and Difference Bounds Matrices(DBMs). We show our proposed method effective by some example.

Teachers' routine is to give careful consideration to possible effbcts which will be produced by teaching materials, learning environment and their behavior in their classes while they make plans for their classroom instruction (i.e. Instructional-Design). Alignment of teaching items and learning tasks, namely, automatic generation of sequencing teaching items is one of important problems in Instructional-Design. Some researchers have been engaged in proposing 'problem dependent procedures' for this sequencing of every new teaching strategy and criterion for sequencing. Other have been engaged in developping Instructional Design Supported System with a mechanism which generates sequences of teaching items using Expert-System. Our method does not need this kind of procedures for sequencing, but only the partial knowledge which evaluates the sequence and Genetic-Algorithm is enough. It enables us to generate variable sequences similar to the sequences done by conventional procedures. And it also enables us to get sequences which reflect combinations of some criterion and procedures, and to get ones which no conventional procedure can generate. In this paper we describe the effect of using Genetic-Algorithm and the method of applying Genetic-Algorithm which includes evaluation functions, fitness functions, Crossover, Mutation and their control mechanisms to the problem of automatic generation of sequencing teaching items. And at last we describe the result of our try at simulation making a comparison between one of the conventional methods and our method and the simulation case where some conventional processes are included in our method.

This paper proposes a new framework to integrate tutoring systems with microworld systems. While varieties of microworld systems and ITSs (Intelligent Tutoring Systems) have been developed as CAIs, most of them restrict the subject to learn. The present authors have already reported the system architecture that seamlessly integrate multiple microworlds that were independently developed for different subjects to teach. This architecture has standardized the message types, event types, and their processing to provide each component as a reusable object transportable across different microworlds. This architecture has enabled us to develop integrated learning systems whose functions are hard to be provided by a single independent system. This paper further tries to integrate multiple tutoring systems together with multiple microworlds by making their component functions as reusable objects. We have also introduced autonomous agents. Therefore, the tutoring functions can be activated not only by students' operations, but also by these agents. The reusable components for tutoring functions and agents have brought the following improvements: (1) Tools and components brought from other microworlds can be used in cooperation with those in the local microworld. (2) Tutoring systems can reuse already existing components and tools developed for other purposes. (3) Multiple tutoring systems can reside in a single microworld. (4) Even those microworlds with tutoring functions can freely import and export components and tools to and from each other. Furthermore, the provision of reusable components to construct ITS would also make it possible to integrate ITSs and microworlds.

In this paper we propose a framework of program comprehension based on a domain world model. Many researchers have developed plan recognition methods or/and some structural graphs and have tried to construct frameworks of program comprehension mainly based on pattern matching to abstract templates. Certainly, those methods are effective to identify which role block each instruction belongs to or to clarify its structural position in the flow or hierarchy of the program. But we think that we can capture rather surface meaning of the program by those approarches. Program is a description of operations to solve problems in some domain worlds. Therefbre, it is important to analyze what effect each instruction have on the domain world operated by the program. From this standpoint, we think that program comprehension system should have some domain world model and the ability to simulate the behavior of the world on the model and to interpret the meaning of instruction based on the result of the simulation. In this paper, we propose such a new, simulation-based framework of program comprehension.

This paper proposes a new regularization method based on the MDL (Minimum Description Length) principle. An adequate precision weight vector is trained by approximately truncating the maximum likelihood weight vector. The main advantage of the proposed regularizer over existing ones is that it automatically determines a regularization factor without assuming any specific prior distribution with respect to the weight values. Our experiments using regression problems showed that the MDL regularizer significantly improves the generalization error of a second-order learning algorithm and shows a comparable generalization performance to the best tuned weight-decay regularizer.

It is well known that the efficiency of computation depends on knowledge representation systems on which problems and algorithms are based. No general theory has yet been established, however, for changing knowledge representation systems correctly. The difficulty is due to the absence of a theory for discussing various knowledge representation systems in a unified manner. We overcome this difficulty by the adoption of "declarative programs" and "specialization systems" which provides a theoretical basis for the formalization of many different knowledge representation systems and programming languages, including context-free grammars, typed unification grammars, many extensions of Prolog, and CLP languages. In the Prolog case, logic programs are declarative programs on a specialization system which formalizes the interrelation between atoms, ground atoms and substitutions. This paper proposes a new theory for changing knowledge representation systems in the framework of problem-solving based on equivalent transformation of declarative programs on specialization systems, where computation is based, not on logical inference, but on the preservation of declarative semantics of declarative programs. A new notion of "safe extension" of specialization systems is established. We prove that safe extension of a specialization system to another one does not change the declarative semantics of a declarative program on the specialization systems. This theorem has wide applicability for changing konwledge representation systems and provides an important theoretical foundation for representation change.