Hiroyasu Matsushima

Hybrid directional-biased evolutionary algorithm for multi-objective optimization

This paper proposes the hybrid Indicator-based Directionalbiased Evolutionary Algorithm (hIDEA) and verifies its effectiveness through the simulations of the multi-objective 0/1 knapsack problem. Although the conventional Multi-objective Optimization Evolutionary Algorithms (MOEAs) regard the weights of all objective functions as equally, hIDEA biases the weights of the objective functions in order to search not only the center of true Pareto optimal solutions but also near the edges of them. Intensive simulations have revealed that hIDEA is able to search the Pareto optimal solutions widely and accurately including the edge of true ones in comparison with the conventional methods.

Dynamic matching range in Exemplar-based Learning Classifier System

This paper proposes the extended version of Exemplar-based Learning Classifier System (ECS) called DMR-ECS which introduces the basis function for the dynamic matching selection in ECS. In comparison with our previous match selection in ECS, the proposed dynamic match selection in DMR-ECS can control an appropriate range of the match selection automatically to extract the exemplars that cover given problem space. Intensive simulation on the cargo layout problem has revealed that DMR-ECS contributes to not only improving the performance but also reducing the number of the exemplars with an appropriate range of the match selection.

Exemplar-Based Learning Classifier System with Dynamic Matching Range for Imbalanced Data

In this paper, we propose a method to improve ECSDMR which enables appropriate output for imbalanced data sets. In order to control generalization of LCS in imbalanced data set, we propose a method of applying imbalance ratio of data set to a sigmoid function, and then, appropriately update the matching range. In comparison with our previous work (ECSDMR), the proposed method can control the generalization of the appropriate matching range automatically to extract the exemplars that cover the given problem space, wchich consists of imbalanced data set. From the experimental results, it is suggested that the proposed method provides stable performance to imbalanced data set. The effect of the proposed method using the sigmoid function considering the data balance is shown.

Evolutionary algorithm for uncertain evaluation function

This paper proposes the evolutionary algorithm (EA) for the uncertain evaluation function in which fitness values change even with the same input. In detail, the proposed method employs the probability model to acquire the appropriate attributes that can drive the good solutions. To investigate the effectiveness of the proposed method, we apply it to sleep stage estimation problem where an accuracy of sleep stage estimation changes even in the same estimation filter (correspondingly the solutions). The experimental results have revealed the following implications: (i) The proposed method succeeded to acquire the robust estimation filters which stably derive a high accuracy of the sleep stage estimation; (ii) in detail, the proposed method with the roulette selection shows higher performance than the one with the random selection; and (iii) the proposed method shows high performance and robustness to the different days in comparison with the conventional sleep stage estimation method.

What kinds of human negotiation skill can be acquired by changing negotiation order of bargaining agents

This paper focuses on developing human negotiation skills through interactions between a human player and a computer agent, and explores its strategic method towards a human skill improvement in enterprise. For this purpose, we investigate the negotiation skill development through bargaining game played by the player and an agent. Since the acquired negotiation strategy of the players is affected by the negotiation order of the different types of agents, this paper aims at investigating what kind of the negotiation strategies can be learned by negotiating with different kinds of agents in order. Through an intensive human subject experiment, the following implications have been revealed: (1) human players, negotiating with the human-like behavior agent firstly and the strong/weak attitude agent secondly, can neither obtain the large payoff nor win many games, while (2) human players, negotiating with the strong/weak attitude agent firstly and the human-like behavior agent secondly, can obtain the large payoff and win many games.

Exemplar-based Learning Classier System: Towards cargo layout optimization

This paper focuses on generalization of learning classifier system (LCS) and explores the method for reducing the time of generalizing conscious rules that have the real number. For this purpose, we pay attention on exemplars (i.e., good examples) and, propose exemplar-based LCS (ECS) that extracts useful exemplars as generalized rules by deleting unnecessary exemplars (some overlapping exemplars) as much as possible. To validate the effectiveness of ECS, this paper applies it to the cargo layout optimization problems. Intensive simulations have revealed the following implications; that (1) the gap between a center of gravity of HTV and its actual center is minimized by ECS in comparison with the other cases that employ 2000 exemplars and the randomly selected exemplars; (2) ECS can minimize the gap with the small numbers of exemplars (i.e., less than 2000 exemplars); and (3) such effectiveness of ECS is maintained even when the predetermined range of the match set is varied, which show the robustness of ECS against the parameter setting.