Recently, Evolutionary Robotics have been applied to evolve the behavior of the autonomous mobile robot. Using Evolutionary Robotics method, the robot controllers are designed in the simulated environment. Evolutionary Robotics are expected as new design methods for the robot controller. However, there exist some errors in the simulated model and the acquired controllers do not always work in the real world. In this research, to overcome these implementation problems, we are focused on the evolution in the real environment using Genetic Algorithm. This approach arises another problem that the load for human operators in fitness evaluation are quite high. To reduce this load, we applied the fitness estimation in the GA procedure. The experimental results show the effectiveness of the proposed method.
The purposes of this paper are to propose and evaluate an immune optimization algorithm using a biological immune co-evolutionary phenomenon and cell-cooperation. The co-evolutionary models searches the solution through the interactions between two kinds of agents, one of the agents is called immune agent which optimizes the cost of its own work. The other is called antigen agent which realizes the equal work assignment. This algorithm solves the division-of-labor problems in multi-agent system (MAS) through the three kinds of interactions:division-and-integration processing is used for optimization of the work-cost of immune agents and, escape processing is used to perform equal work assignment as a result of evolving the antigen agents. The immune agent optimizes own cost using division as well as integration processing based on the immune cell-cooperation which is considered a kind of parallel-distributed system with role differentiation. The 'splicing' is one of the re-combination operator of genes, whose function is used for forming the role. The division as well as integration processing in our method is based on the splicing. And the antigen agent computes even division of work domain using escape processing based on a phenomenon that the antigen evolves to escape from the elimination of immune system. In order to investigate the validity of the proposed method, this algorithm is applied to the "N-th agent's Travelling Salesmen Problem (called the n-TSP)" as a typical problem of MAS. The property that is believed to function as solution driver for MAS shall be clarified using several simulations.
Spaceplane, which is expected as an Advanced Space Transportation Vehicle, has a flight region so extensive that the flow field around the airframe is changed greatly. Since some complex interference between the spaceplane airframe shape and the flight trajectory through this flow field, both must be optimized during conceptual designing of the spaceplane. However, with existing research results and knowledge about spaceplane being so little, it is impossible to specify a suitable initial value during conceptual designing. Therefore, it is necessary to search for its solution automatically in the large solution space, and a designing tool for this problem has been in demand. We thus propose a designing tool using Genetic Algorithm (GA), which is an optimization method for global searching. In addition, the results of applying this tool to an integrated optimization problem of the airframe shape and flight trajectory of a spaceplane is also reported.
We have proposed a method that acquired concept relation rules from some texts with text classes. The method deals with only a training example that all concept class has at most a key concept. The paper proposes a method dealing with a training example that some concept classes have multiple key concepts. The method incorporates the membership function, which processes single key concept as a special case of multiple key concepts in each attribute, into a fuzzy inductive learning method. Also, the paper compares the revised fuzzy inductive learning method with an old fuzzy one and C4.5.
Feature subset selection is of prime important in pattern classification, machine learning and data mining applications. A real world database may contain many noisy, unnecessary and irrelevant features. If it is used for data minihg directly, the quality of the discovered knowledge may be very poor. To cope with this problem, many methods have been proposed. In this paper, we propose a hybrid algorithm by using class mutual information for feature selection, starting from the Rough Sets CORE. If the CORE is empty we use binary mutual information for the first feature selection. Experiments have been conducted on some artificial and real world domains in terms of tree size, test errors rate and subset sizes. The results show the effectiveness of proposed hybrid algorithm.
Ordinary time series analysis operates on time series data measured on interval scale, and/or ratio scale. In the study of psychology, sociology, ethnology, and so on, we often encounter fuzzy time-series data related to human judgement, or recognition in the process of data colletction. In such cases, investigating the trend of responses is one of the most important research activity. One of the ways to clarify the trend is visualize the variation. In order to achieve this purpose, "fuzzy time-series analysis of qualitative data" is proposed as a extention of "time series analysis of qualitative data (Inagaki, 1997). The basic idea behind the method is to assign numerical scores to categories so as to maximize correlation between the category scores and function of times. In the method, the three models, i.e., irreversible model, reversible model, and circulative model, are provided for expressing the trend. The method can be widely used as a tool for analyzing not only time-series data but also fuzzy category data in order to explore the relationship between numerical variables and fuzzy categories. In this paper, mathematical formulation and algebraic solution are showed and some technical problems and relation to other methods are also discussed.