Most of the former design aid systems have acquired some form of representative behavior of many users and designers by means of techniques such as multivariate analysis or interviews. But, such a conventional system is based on an averaged representation of various preference and difficult to adapt to changes in time or differences by age group or culture. In this paper, a design aid system that is based on a particular individual preference is discussed. We propose an easy-to-use design aid system which makes use of an interactive evolutionary system paradigm. Genetic programming, an evolutionary system, is used to formulate a user's preference as a mathematical function based on history of usage. We propose that the function can be used in support of a design aid system that enable designs based on individual values by implementing particular designer preferences into the system. We evaluate synthesizing a function with linear discrimination and examine that synthesizing a function based on history of usage of the system is effective. By simulating a design process with a synthesized function as a virtual user, we examine our proposal that a synthesized function supports the system and show its effect.

This paper proposes to display individuals of interactive evolutionary computation(EC) in an evaluation order to reduce the fatigue of human operators. To display in an evaluation order, two prediction methods which do not depend on application tasks-the first using a neural networks(NN), and the second using Euclidean distance measure-are proposed. The prediction method based on an NN learns the relationship between past individuals and their evaluation values that an interactive EC operator gave in the past EC generations. The prediction method based on Euclidean distance measure predicts the human evaluation to a given individual using a weighting average method of the evaluation values to individuals in the past EC generations. These weights are calculated by the Euclidean distances between the given individual and individuals in the past. We evaluate the performances of these two prediction methods through simulation and subjective tests. The simulation test evaluates how the predicted order is closer to the actual one than similarity between the order that GA produces-conventional order-and the actual one. The comparison of cross-correlations between these two similarities has shown that the proposed method is significantly better than the conventional method. Following the simulation, the subjective tests are conducted to test how the proposed method let human operators feel better than the conventional method. The result has shown that there is no significant difference between the proposed method and the conventional method unlikely to the simulation test. We discuss the results of simulation and subjective tests. Furthermore, we discuss four hints for the future improvement obtained through subjective tests and the reports from subjects.

In the conventional image retrieval, the user has performed the retrieval by using linguistic keywords. However, it is difficult to use this method in the image retrieval treating the user's subjectivity like the impression received from images. Recently, as one of the method treating the user's subjectivity, the interactive genetic algorithms is researched. In this paper, we propose the image retrieval system using the interactive genetic algorithms in the retrieval treating the personal subjectivity. Using the combination of the artificial selection used in the conventional interactive genetic algorithms and natural selection, the number of the branch from a certain generation could be increased. Moreover, the use of the retrieval history can give the user the effective retrieval without performing the complex evaluation of the presented images. We evaluate the system through the experiment using the subjects. In addition, we compare the proposed system with the system using the conventional interactive genetic algorithms which decides the parents by using artificial selection.

We propose an interactive image browsing system which combines similarity-based image retrieval technology and simulated breeding methodology. Here "simulated breeding methodology" refers to the generic framework of human interface for the interactive optimization of subjective problems. Based on this framework, the scenario of interactive image browsing is summarized as follows: (1)A user is first requested to determine the first example image before starting image browsing. (2)The system next retrieves similar images to the given example image by applying current image retrieval parameters. (3)Image retrieved in high orders are displayed on the screen for the user to give some feedback to the system. (4)A significance point to each image given by the user, which is the form of feedback employed in this paper, is transformed into the fitness of individuals; then the underlying optimization algorithm, namely queue-based genetic algorithm(QGA) proposed in this paper as an appropriate algorithm for simulated breeding methodology, plays the role of interactive optimization of image retrieval parameters. To realize such a system, graphical user interface is constructed for the satellite cloud image database that archives 1027 images, where the image representation model used for extracting and indexing image contents is called "hierarchical model of image content elements". Results are analyzed based on the history of image browsing, and they demonstrate steady improvement in terms of the similarity retrieval order of the target image, because of the sequential change of example images and the optimization of image retrieval parameters by QGA.

Interactive genetic algorithm(IGA) has been recognized as a useful tool for searching problems were fitness is determined by human intervention. It has an important aspect as a user-interface when searching for a target according to user's subjective factors such as similarity, attractiveness, preference, and so forth. Two important problems of the IGA are how to assign fitness to populatiions utilizing the human sense, and how to reduce the user's load. We compared three fitness assignment methods, "rating all", "bias", and "fuzzy" in searching for a cartoon which has been imagined by user. The latter two methods have been proposed by us. The user's load was lighter and time to process was shorter in the fuzzy method compared with the rating all method. However, the final populations in the fuzzy method did not become as similar to a target as in the rating all method. This is because the fittest cartoon face is picked up based on a psychometrical measure, while fitness of the other cartoon faces is calculated with a physical measure in the fuzzy method, that is, the fuzzy reasoning is performed in the physical face space. In this paper we improved the fuzzy method by mapping the cartoon faces from the face space to the psychometrical space and performing the fuzzy reasoning in the psychometrical space. The distance between any two faces in the psychometrical space expresses the correct similarity between them. Therefore the fitness of the other faces as well as the picked-up one can be obtained based on the human sense. Finally the fuzzy method can get the advantage of the rating all method, that is coming closest to a target, without sacrificing its own advantages, those are user-friendliness and quickness.

GAs(Genetic Algorithms) evolve populatiions to search solutions for optimization problems by using the crossover, that is the unique operator in similar stochastic direct optimization techniques like SAs(Simulated Annealings). Generation alternation models are important to give controls on search processes. Analytic works should show how to design an encoding/crossover and a generation alternation model for given problems. Existing works have often used SGA(simple or standard GA) for a fixed generation alternation model, although it was often pointed out to be problematic for real applications because of its high selection pressure. This paper shows an analysis focused upon how generation alternation models influence crossover's effect in solving the minimal deceptive problems. We prepared two generation alternation models; SGA as one extreme and the rMGG(routellete minimal generation gap model) for the other extreme from Satoh's experimental analysis. We have made a Markov analysis for the MDPs(minimal deceptive problems) with relatively small populations. We show SGA-based GAs are easily deceived than rMGG-based ones, and stepping stones help crossover's effect in deceptive problems. We have planed three experiments on three bits and four bits FDPs(fully deceptive problems), and demonstrate these conjectures are not violated in these cases.

Eshghi and Kowalski's abductive procedure is not in general sound with respect to the 2-valued stable model semantics, but sound with respect to preferred extension(Dung, 1991). We establish an abduction framework in which a newly proposed integrity constraint allows atoms to be underivable from the theory and the abducibles. In correlation with the newly proposed integrity constraint, we present a phase of derivations augmented to Eshghi and Kowalski's abductive procedure, by which abductive unsucceeding derivations are adjusted as detections of underivability. By dealing with alternating fixpoint semantics as in[Van Gelder 93], in relation with our constraint, we see, with respect to our constraint, the soundness of Eshghi and Kowalski's abductive procedure. The motive of considering such a adjusty derivation comes up from a relaxation that there may be some atom which is not a logical consequence of P^* ∪ Δ and whose negation is neither a logical consequence. Finally the purpose of the paper is(1)to relax the constraint that all the(ground)atom or its negation should be the logical consequence of the union of a given theory and abducibles, and to get a constraint in which some atom may not be cared, (2)to extend Eshghi and Kowalski's abductive procedure to the one with an adjusty derivation phase, sound with respect to our constraint.

Problem solving is a basic human activity which fills gaps between requirements and reality. In general, requirements can be represented by a spatial structure and a temporal one. Similarly, facts of the real world contain spatial and temporal structures. In spite of close relationship between spatial and temporal structures, most of conventional problem solving frameworks determine either of them at a time. This paper proposes a new problem solving framework in which both a spatial structure and a temporal structure are determined concurrently. Moreover, the framework deals with ill-defined problems in which the problem itself is not yet clarified in advance. The framework, first, clarifies the problem in a computer, while the requirements and the facts are represented by primitives and their connections. In the next step, a constraint satisfaction module and a state transition module are performed to maintain consistency in the structures. The former module determines the problem structure spatially, and the latter determines temporally, so that these two modules are complementary. Finally in the third step, some solution candidates are visualized in order to evaluate with the external human knowledge. This step leads a human to represent uncertain part of the problem and enforce to articulate new knowledge. This paper illustrates the proposed framework using a simple example of problem solving. After discussing the results of the example, this paper concludes that the framework has advantages of usability of informal and fragmentary knowledge in problem solving. Furthermore, particularly in cases where creativity is demanded, the features of the framework are expected to make some creative solutions.

In this paper, we propose a new genetic algorithm(GA) for job-shop scheduling problems(JSPs), considering dependencies among machines. We regard the crossover as a main search operator. Crossovers should preserve characteristics between parents and their children in order for GAs to perform well. Characteristics are elements that constitute a solution and determine the fitness of the solution. Chracteristics also should be highly independent of each other. A characteristic has to be found for each problem domain since it depends on a particular problem domain. We basically regard the processing order of jobs as a characteristic for JSPs. We consider job-based order inheritance and position-based order inheritance for ways of inheritance of the processing order by crossovers, and propose two new crossovers; the Inter-machine Job-based Order Crossover(Inter-machine JOX) and the Inter-machine Position-based Order Crossover(Inter-machine POX). By applying them to the benchmark problems of FT10×10 and FT20×5, we demonstrate that the Inter-machine JOX shows better performance than the Inter-machine POX and an existing crossover, the SXX[Kobayashi 95]. The Inter-machine JOX preserves both the processing order of jobs and the technological ordering which causes dependencies among machines. We also propose a new mutation named the Inter-machine Job-based Shift Change for introducing a diversity of population. We confirm its effectiveness by applying it with the Inter-machine JOX to FT10×10 and FT20×5.

The paper proposes a search method for pipe route planning using genetic algorithm incorporated with several heuristics. First, the basic principle of our method is presented using key ideas which include representation of pipe route for GA operations, spatial potential energy to cover design scenarios, fitness function, basic GA operations, coordinates conversion procedure, and route modification procedure using subgoal setting. In order to apply the method to actual problems and to solve them in a practical manner, the study employs various heuristics, which are concept of direction, generation of initial individuals using intermediate point, extended two-points crossover, and dynamic selection. Those heuristics are described separately and their effectiveness in our method is discussed. The paper also shows a prototype system which were developed based on our approach and discusses the validity of the proposed method.

Mobile robots are expected to run in environments where unexpected scenes or obstacles may occur which disturb the exact measurement of distances in the space. In such an environment, it is difficult for a robot to move tactfully, avoiding obstacles e.g.walls, columns, etc. In order to achieve such tactful movements, it is desired that a mobile robot can focus attention to meaningful points around itself, and plan its own movement toward the goal. A key point for such an intelligent recognition-based control is to represent the meaning of the environment for the robot, i.e., whether visible objects help or disturb the robot's movement. In this paper, we present a qualitative reasoning based driving, for a mobile robot to park in an unknown, small square parking lot. The environment, observed by the omni-directional vision of the robot, is segmented according to the motion capability and shape of the robot, to generate a robot-centered logical description of the environment. In experiments, we obtained accurate, safe and efficient parking behaviors due to this description, in various simulated parking lots.

The aim of this work is proposing a method to construct the knowledge model of a trainee for a training support system of plant operation. Following our previous study, trainee's knowledge is represented as two-layered model of task hierarchy and qualitative causality. A link between two nodes in this representation can be thought of as a unit of trainee's knowledge. Possessive states of these units are identified by inferencing trainee's intention and consistency maintenance. In intention inferencing, candidates of operation goal are enumerated from current state of the plant, operation procedures for each goal are generated using the trainee's knowledge model, and then the procedures are matched with operations that actually taken by the trainee. A prototype system of training support system was implemented on a workstation assuming a model plant consisting of a few tanks, valves and pipes. An experiment was performed using this prototype, where subject's intention inferred by the system was verified against subject's utterance. It was consequently shown that intention inferencing was appropriate in most cases, and that heuristics used for prioritizing candidates of intention did not heavily depend on subjects or experimental scenarios.

We present a method of proving the termination of term rewriting systems(computational systems specified by a set of rewrite rules)by using binary decision diagrams(BDD's) for efficient representation of provability. First, we give a recursive definition of the Boolean function that computes the provability based on a partial ordering &sc; on the set of function symbols. Then the construction of the BDD's for this function, in which a primitive expression f &sc; g consisting of two function symbols f and g is associated with the logical variable x_<fg>, is incorporated into an interactive, incremental termination verification procedure. We conduct some experiments to see how the performance of this procedure is affected by some heuristic selection of variable orderings and constraint orderings, and show that our method and heuristics are useful.