進化計算学会論文誌
Online ISSN : 2185-7385
ISSN-L : 2185-7385
3 巻, 3 号
選択された号の論文の12件中1~12を表示しています
論文
  • 安藤 大地, 伊庭 斉志
    2012 年 3 巻 3 号 p. 73-84
    発行日: 2012年
    公開日: 2012/12/31
    ジャーナル フリー
    The use of Interactive Evolutionary Computation(IEC) is suitable to the development of art-creation aid system for beginners. This is because of important features of IEC, like the ability of optimizing with ambiguous evaluation measures, and not requiring special knowledge about art-creation. With the popularity of Consumer Generated Media, many beginners in term of art-creation are interested in creating their own original art works. Thus developing of useful IEC system for musical creation is an urgent task. However, user-assist functions for IEC proposed in past works decrease the possibility of getting good unexpected results, which is an important feature of art-creation with IEC. In this paper, we propose a new IEC evaluation process named ``Shopping Basket'' flow IEC. In the process, an user-assit function called Similarity-Based Reasoning allows for natural evaluation by the user. The function reduces user's burden without reducing the possibility of unexpected results.
  • 小野 智司, 武田 和大, 中山 茂
    2012 年 3 巻 3 号 p. 85-97
    発行日: 2012年
    公開日: 2012/12/31
    ジャーナル フリー
    Anagram is a word, a phrase, or a sentence generated by reordering the letters of a given, another word, phrase, or sentence. People have used anagrams as a word game, pseudonym, or cryptography. In previous work which proposed an automatic generation method for Japanese anagram senteces, it is hard to incorporate user prefrence, even when a user has flash of some interesting words or phrases. Although Interactive Evolutionary Computation (IEC) is a promising approach to incorporate user preference, IEC faces the trade-off problem between search performance and cost due to user fatigue during solution evaluation. Therefore, this paper proposes a method for Japanese anagram sentence generation using Cooperative Evolution by User and System (CEUS). CEUS is a method enabling to change the search roles between a user and a system in problems involving both qualitative and quantitative objective functions. The proposed anagram generation method using CEUS allows a user to give a bias to the search, not to make much operations, and to directly incorporate good words or phrases the user thinks of during the search. Experimental results showed that, by dynamic search role changes and qualitative objective function prediction, the proposed method could generate anagram sentences more efficiently than IEC-based method with keeping good support for user's divergent thinking.
  • 裴 岩, 高木 英行
    2012 年 3 巻 3 号 p. 98-108
    発行日: 2012年
    公開日: 2012/12/31
    ジャーナル フリー
    We propose a triple comparison and a quadruple comparison-based mechanism to enhance differential evolution (DE), especially interactive DE (IDE) search without increasing IDE user's fatigue largely. Besides a target vector and a trial vector of normal DE, their opposing vectors generated by opposition-base learning are used to determine offspring, and the best vector among them becomes offspring in the next generation. We evaluate the proposed methods by comparing them with conventional IDE and conventional opposition-based IDE using a Gaussian mixture model with four different dimensions for evaluating simulated IDE. We also compare them using 24 benchmark functions for evaluating DE. The experiments show that our proposed methods can enhance IDE and DE search efficiently from several evaluation indexes including the converged fitness values at the same generation numbers and the same fitness calculation numbers, fitness calculation cost, success rates of convergence, and acceleration rates.
巻頭言:「進化型多目的最適化(EMO)」特集号
招待論文:「進化型多目的最適化(EMO)」特集号
  • Current and Future Research Topics
    Carlos A. Coello Coello
    2012 年 3 巻 3 号 p. 110-121
    発行日: 2012年
    公開日: 2012/12/31
    ジャーナル フリー
    After more than 25 years of existence, evolutionary multi-objective optimization has become a mature discipline within evolutionary computation, producing an important flow of publications each year. This paper presents a brief overview of the main topics on which researchers in this area are currently working, as well as some discussion of the areas which, from the author's perspective, constitute promising research directions for the next few years. The topics discussed include algorithmic design, scalability, efficiency, hybridization, parameter control, theory and incorporation of user's preferences. The contents of this paper intends to provide a quick overview of the current state and challenges within evolutionary multi-objective optimization, and is intended to be useful for those interested in pursuing research in this area.
論文:「進化型多目的最適化(EMO)」特集号
  • 多数目的0/1ナップザック問題における性能検証
    佐藤 寛之, カルロス コエロ, エルナン アギレ, 田中 清
    2012 年 3 巻 3 号 p. 122-132
    発行日: 2012年
    公開日: 2012/12/31
    ジャーナル フリー
    Crossover controlling the number of crossed genes (CCG) significantly improves the search performance of multi-objective evolutionary algorithms (MOEAs) in many-objective optimization problems (MaOPs). CCG controls the number of crossed genes by using a static parameter α. To achieve high search performance by using the static CCG, we have to find out an appropriate parameter α* by conducting many experiments. To avoid time consuming parameter tuning and find out an appropriate α* in a single run of the algorithm, in this work we propose a self-adaptive CCG which dynamically controls the parameter α during the solutions search in a single run of the algorithm. Through experiments using many-objective 0/1 knapsack problems, we show that the values of α controlled by the self-adaptive CCG is converged to an appropriate value even when the self-adaptation is started from any initial values. Also, we show the self-adaptive CCG achieves 80~90% with a single run of the algorithm for the maximum search performance obtained by the static CCG using an optimal α*.
  • 立川 智章, 大山 聖, 藤井 孝藏
    2012 年 3 巻 3 号 p. 133-142
    発行日: 2012年
    公開日: 2012/12/31
    ジャーナル フリー
    In this study, we conducted a Multi-Objective Design Exploration (MODE) using Genetic Programming (GP) for extracting design information from non-dominated solutions. Tree-based Genetic Programming is applied to non-dominated solutions and to discover design information between objective function and design parameters as expressions in symbolic form without prior knowledge of the problem. The unique feature of GP is that it finds not only the linear relationship between parameters, but also the nonlinear relationship automatically. In MODE, GP can be used as a symbolic regression technique to extract the relationship between objective functions and design parameters as symbolic equations.
    We addressed two problems. First one is a test problem in which the relationship between objective function and design parameters of data set is given as a symbolic equation which includes nonlinear terms. Objective functions of tree-based GP are minimization of the number of nodes for simplicity of equation and mean absolute error for accuracy of equation. As the result, various optimal equations which include from simple equations with large residual to complex equations with small residual are obtained. These optimal equations are called ``Non-dominated equations'' here. Nonlinear terms included in data set can be extracted by analyzing the trend of non-dominated equations. In practical problem, the relationship between objective function and design parameters of non-dominated solutions is unknown. Second one is the multi-objective aerodynamic design optimization problem of flapping airfoil motion. Objective functions for optimization are maximization of the time-averaged lift(CL,ave) and the time-averaged thrust(CT,ave), and minimization of the time-averaged required power(CPR,ave). The objective values are evaluated using a two-dimensional incompressible Navier-Stokes solver and a multi-objective evolutionary algorithm code is used to obtain non-dominated solutions. Tree-based GP is applied to each objective function for optimization separately. As the result of analyzing non-dominated equations, pitch offset and frequency have large effect on CL,ave and CT,ave. Furthermore frequency and the square of frequency significantly affect CPR,ave.
  • 渡邉 真也, 横内 直樹
    2012 年 3 巻 3 号 p. 143-154
    発行日: 2012年
    公開日: 2012/12/31
    ジャーナル フリー
    In this paper, a new local search (LS) method using an approximate gradient for multi-objective optimization problems (MOOPs) is proposed. The proposed method has two key features; local Pareto optimality and an interpolation mechanism for capturing the whole of Pareto subsets. First feature aims to guarantee approximate local Pareto optimality for solutions, and second one tries to find the whole of Pareto front.
    In order to guarantee local Pareto optimality, there are two considerable things; a kind of local optimality condition for MOOPs and a judgment mechanism for detecting whether a solution satisfies the local optimality condition or not. The proposed method uses Frits John conditions, which expand Karush-Kuhn-Tucker (KKT) conditions, and applies steepest descent method to candidate solutions until solutions satisfy this condition.
    Also, the proposed method incorporates a new interpolation mechanism for detecting local Pareto subsets exhaustively and capturing the entire shape of each Pareto subset. This mechanism is based on fundamental assumptions that non-dominated front is formed by plural non-dominated subsets and it is not difficult to find an entire non-dominated subset within same non-dominated subset.
    The proposed method is one of posteriori LSs, which are applied to final solutions obtained by EMO (or randomly generated solutions). Since the proposed method is based on an approximate gradient, only continuous typical EMO examples were used for investigating the effectiveness of the proposed method.
  • 内種 岳詞, 畠中 利治
    2012 年 3 巻 3 号 p. 155-162
    発行日: 2012年
    公開日: 2013/01/08
    ジャーナル フリー
    Particle swarm optimization (PSO) is a stochastic multi point search algorithm. A PSO family that is applied to multi-objective optimization problems is called multi-objective PSO. The main differences between single-objective PSO and multi-objective PSO are ``archive'' and ``guide selection''. Non-dominated solutions obtained by multi-objective PSO are stored in archive and archived members are candidates to approximate the Pareto optimal front. Therefore the archived members should be close to the true Pareto front and cover the true Pareto front widely and uniformly. In order to get better solutions, it is important that how to select the guides, what parameter to be used and how to approximate the Pareto optimal front. In this paper, a guide selection method in which either a personal best or a global best is selected depending on single interested objective function among all functions is proposed. Some combinations of the proposed guide selection method with conventional guide selection methods are examined by using several well known benchmark problems. The results show that employing the proposed guide selection method leads to better coverage on and faster convergence to the Pareto optimal front.
  • 平野 博之, 吉川 大弘
    2012 年 3 巻 3 号 p. 163-172
    発行日: 2012年
    公開日: 2012/12/31
    ジャーナル フリー
    Particle Swarm Optimization (PSO) is one of the most effective search methods in optimization problems. Multi-objective Optimization Problems (MOPs) have been focused on and PSO researches applied to MOPs have been reported. On the other hand, the problem that the search performance using conventional methods for MOPs becomes low is reported in Many-objective Optimization Problems (MaOPs) which have four or more objective functions. This paper proposes two-step searching method based on PSO for MaOPs. In the first step, dividing the population into some groups and each group performs the single objective search for each objective function and the center of them. In the second step, the search is performed to acquire the diversity of Pareto solutions by PSO search with the goal, global-best, based on the solutions acquired in the first step. This paper defines the real coded multi-objective knapsack problem and studies the performance of the proposed method applied to this problem.
  • 下山 幸治, 鄭 信圭, 大林 茂
    2012 年 3 巻 3 号 p. 173-184
    発行日: 2012年
    公開日: 2012/12/31
    ジャーナル フリー
    This paper compares the criteria for updating the Kriging response surface models in multi-objective optimization: expected improvement (EI), expected hypervolume improvement (EHVI), estimation (EST), and those combination (EHVI+EST). EI has been conventionally used as the criterion considering the stochastic improvement of each objective function value individually, while EHVI has been recently proposed as the criterion considering the stochastic improvement of the front of non-dominated solutions in multi-objective optimization. EST is the value of each objective function, which is estimated non-stochastically by the Kriging model without considering its uncertainties. Numerical experiments were implemented in the welded beam design problem, and empirically showed that, in a non-constrained case, EHVI keeps a balance between accurate, wide, and uniform search for non-dominated solutions on the Kriging models in multi-objective optimization. In addition, the present experiments suggested future investigation into the techniques for handling constraints with uncertainties to enhance the capability of EHVI in a constrained case.
論文
  • 宮川 みなみ, 佐藤 寛之
    2013 年 3 巻 3 号 p. 185-196
    発行日: 2013年
    公開日: 2013/01/25
    ジャーナル フリー
    When multi-objective optimization problems include several constraints, multi-objective EAs (MOEAs) need to introduce a mechanism to obtain feasible solutions from infeasible ones. In this work we propose a novel constrained MOEA introducing a parents selection based on two-stage non-dominated sorting of solutions and a directed mating in objective space. In the proposed algorithm, first, we classify the entire population into several fronts by non-dominated sorting based on constraint violation values. Then, we re-classify each obtained front by non-dominated sorting based on objective function values, and select the parents population from higher fronts. In this way, superiority of solutions in the same non-dominance level of constraint violation values is determined by non-dominance level of objective function values. It leads to find feasible solutions having better objective function values. In addition, to generate an offspring, after we select a primary parent, we pick solutions dominating the primary parent from entire population including infeasible solutions. Then we select the secondary parent from the picked solutions and apply genetic operators. In this way, we utilize valuable genetic information of infeasible solutions to converge the primary parent towards its search direction in objective space. Through performance verification using SRN, TNK, OSY and m objectives k knapsacks problems, we show that the proposed algorithm achieves higher search performance than the conventional CNSGA-II (Constrained NSGA-II) and RTS algorithms proposed by Ray et al.
feedback
Top