Nobuo Sannomiya

Cooperation of multi-agent system and its composition

Abstract Cooperation is considered to be very important for an autonomous multi-agent system (called MAS in short) composed of many autonomous agents. However, keeping a suitable combination of cooperation and diversity is considered to be more important for a MAS behaving in a dynamic environment. In this paper, as the examples of autonomous MAS, two species of fish schools with different cooperation and diversity are proposed for studying the relationship between the cooperation of MAS and its composition. A fish can be considered as an agent is because it can perceive the environment and measure the speeds and directions of other fish by using its sensors, such as eyes and lateral lines. It makes a decision by cooperating with the behavior of other fish and adapting to the environment change. A large number of simulations are carried out by using the two species of fish school models behaving in a water tank. Then the deadlock situations of different fish school models in the trap are compared and analyzed by using the simulations.

A Neural Network Model for Predicting the Error Rates of Students for a Learning Problem

In general it is impossible to know the learning effect of students before teaching them. Therefore, teachers have to predict it in order to perform actual teaching effectively and efficiently. In this paper, we propose a method to predict the error rates of the students for a learning problem and analyze how to teach students effectively through the prediction results. For this purpose, a multi-layer neural network (MNN) model is used. In this model, the input variables are five aptitude abilities of a student and the output variables are three error rates. It is confirmed that the prediction values obtained by using this MNN are reasonable as compared with the experimental results. Moreover, from the sensitivity analysis, the aptitude abilities to reduce the error rates are identified. This result makes it possible to predict the error rates of a learning problem in advance. By using these results, teachers can instruct students more effectively and efficiently.