日本語
Hiroshima City University 
Graduate School of Information Sciences 
Dept. of Intelligent Systems 

Professor 
Takahama Tetsuyuki 

Tel.082-830-1698  
 
My website is here.  

Academic background
Kyoto University  Faculty of Engineering  電気工学第二学科  1982  Graduated 
Kyoto University  Graduate School, Division of Engineering  電気工学第二専攻  Doctor course  1987  Withdrawn after completion of required course credits 

Academic degrees
Master of Engineering  Kyoto University 
Doctor of Engineering  Kyoto University 

Research Areas
Computational Intelligence 
Evolutionary Computation 
Optimization 

Research keywords
Natural Computing 
Evolutionary Algorithms 
Optimization Algorithms 

Subject of research
Learning Fuzzy Control Rules  1990-Present 
Multiobjective optimization method  1996-Present 
Solving Constrained Optimization Problems by Alpha Constrained Methods  1997-Present 
Structural Learning by Degeneration  1999-Present 

Bibliography
"A Study on an Equivalent Penalty Coefficient Value for Adaptive Control of the Penalty Coefficient in Constrained Optimization by Differential Evolution", in Advanced Studies in Economic Sciences: Information Systems, Economics and OR  Setsuko Sakai and Tetsuyuki Takahama  65-86  Kyushu University Press  2019/02  The penalty function method has been widely used for solving constrained optimization problems. In the method, an extended objective function, which is the sum of the objective value and the constraint violation weighted by the penalty coefficient, is optimized. However, it is difficult to control the coefficient properly because proper control varies in each problem. In this study, the equivalent penalty coefficient value (EPC) is proposed for population-based optimization algorithms (POAs). EPC can be defined in POAs where a new solution is compared with the old solution. EPC is the penalty coefficient value that makes the two extended objective values of the solutions the same. Search that gives priority to objective values is realized by selecting a small EPC in a population. Search that gives priority to constraint violations is realized by selecting a large EPC. It is expected that the adaptive control of the penalty coefficient can be realized by selecting an appropriate EPC. The proposed method is introduced to differential evolution. The nature of the proposed method is shown by solving several constrained optimization problems. 
"A Study on Selecting an Oblique Coordinate System for Rotation-Invariant Blend Crossover in a Real-Coded Genetic Algorithm", in Recent Studies in Economic Sciences: Information Systems, Project Managements, Economics, OR and Mathematics  Setsuko Sakai and Tetsuyuki Takahama  65-87  Kyushu University Press  2018/02  Differential Evolution (DE) has been successfully applied to various optimization problems. The performance of DE is affected by algorithm parameters such as a scaling factor F and a crossover rate CR. Many studies have been done to control the parameters adaptively. One of the most successful studies on controlling the parameters is JADE. In JADE, the values of each parameter are generated according to one probability density function (PDF) which is learned by the values in success cases where the child is better than the parent. However, search performance might be improved by learning multiple PDFs for each parameter based on some characteristics of search points. In this study, search points are divided into plural groups according to some criteria and PDFs are learned by parameter values in success cases for each group. Objective values and distances from a reference point, which is the best search point or the centroid of search points, are adopted as the criteria. The effect of JADE with group-based learning is shown by solving thirteen benchmark problems. 
"A Comparative Study on Grouping Methods for an Adaptive Differential Evolution", in Challenging Researches in Economic Sciences: Legal Informatics, Environmental Economics, Economics, OR and Mathematics  Setsuko Sakai and Tetsuyuki Takahama  51-91  Kyushu University Press  2017/03  Differential Evolution (DE) has been successfully applied to various optimization problems. The performance of DE is affected by algorithm parameters such as a scaling factor F and a crossover rate CR. Many studies have been done to control the parameters adaptively. One of the most successful studies on controlling the parameters is JADE. In JADE, the values of each parameter are generated according to one probability density function (PDF) which is learned by the values in success cases where the child is better than the parent. However, search performance might be improved by learning multiple PDFs for each parameter based on some characteristics of search points. In this study, search points are divided into plural groups according to some criteria and PDFs are learned by parameter values in success cases for each group. Objective values and distances from a reference point, which is the best search point or the centroid of search points, are adopted as the criteria. The effect of JADE with group-based learning is shown by solving thirteen benchmark problems. 
"A Comparative Study on Detecting Ridge Structure for Population-Based Optimization Algorithms", in Contemporary Works in Economic Sciences: Legal Informatics, Economics, OR and Mathematics  Setsuko Sakai, Tetsuyuki Takahama  61-82  Kyushu University Press  2016/02 
"A Study on Adaptive Parameter Control for Interactive Differential Evolution Using Pairwise Comparison", in New Solutions in Legal Informatics, Economic Sciences and Mathematics  Setsuko Sakai, Tetsuyuki Takahama  101-121  Kyushu University Press  2015/03 
View details...

Papers
Research paper (international conference proceedings)  Joint  A New Binomial Crossover Considering Correlation Among Decision Variables for Adaptive Differential Evolution  Tetsuyuki Takahama and Setsuko Sakai  Proc. of Joint 10th International Conference on Soft Computing and Intelligent Systems and 19th International Symposium on Advanced Intelligent Systems (SCIS&ISIS2018)  467-473  2018/12  In population-based optimization methods such as evolutionary algorithms, various information can be obtained from the distribution of good search points. When problems with strong dependency among decision variables are optimized, a characteristic distribution, which is a thin elliptical distribution, may appear. In order to generate good children, it is necessary to change the variables simultaneously along the long axis of the elliptical distribution. A similar distribution also may appear when the search points are far from the optimal solution even in problems with independent variables. In this study, we propose a new crossover CBX which uses correlation coefficients of search points in order to detect such distribution and realizes efficient movement toward the optimal solution. The crossover points are decided so that highly correlated variables are inherited at the same time. However, if only CBX is used, the diversity of the search points tends to be lost rapidly. The adaptive control of the probability for applying CBX is also proposed. The advantage of the proposed method is shown by solving several benchmark problems. 
Research paper (scientific journal)  Joint  The Velocity Updating Rule According to an Oblique Coordinate System with Mutation and Dynamic Scaling for Particle Swarm Optimization  T.Takahama, S.Sakai  Artificial Life and Robotics  23/ 4, 618-627  2018/12  Particle swarm optimization (PSO) has been showing powerful search performance especially in separable and unimodal problems. However, the performance is deteriorated in non-separable problems such as rotated problems. In this study, a new velocity updating rule according to an oblique coordinate system, instead of an orthogonal coordinate system, is proposed to solve non-separable problems. Two mutation operations for the best particle and the worst particle are proposed to improve the diversity of particles and to decrease the degradation of moving speed of particles. Also, the vectors generated according to the oblique coordinate system is dynamically scaled in order to improve the robustness and efficiency of the search. The advantage of the proposed method is shown by solving various problems including separable, non-separable, unimodal, and multimodal problems, and their rotated problems and by comparing the results of the proposed method with those of standard PSO. 
Research paper (international conference proceedings)  Joint  Grouping of Genes According to Correlation Coefficients and Grouping-Based Crossover for Adaptive Differential Evolution  Tetsuyuki Takahama and Setsuko Sakai  The 50th ISCIE International Symposium on Stochastic Systems Theory and Its Applications (SSS'18), extended abstract  2018/11  When problems with strong dependency among deci- sion variables are optimized, a characteristic distribu- tion, which is a thin elliptical distribution, may appear. In order to generate good children, it is necessary to change the variables (genes) simultaneously along the long axis of the elliptical distribution. Since binomial crossover in differential evolution determines whether each gene is crossed or not with the same probability, it is difficult to change some genes simultaneously. In this study, we propose a crossover operation GBX which uses correlation coefficients of search points in order to detect such distribution. The highly correlated genes are grouped and the genes in each group are crossed (or not crossed) simultaneously. However, if only GBX is used, the diversity of the search points tends to be lost rapidly. The adaptive control of the probability for applying GBX is also proposed. The advantage of the proposed method is shown by solving several bench- mark problems. 
Research paper (bulletin of university, research institution)  Joint  変数間依存性を解消する変換を導入したブレンド交叉の提案  阪井節子,高濱徹行  数理解析研究所講究録2078  65-72  2018/07  本研究では,変数依存性を弱めるために空間を変換し,変換された空間上で変数分離型問題に強いBLX-αを適用するTBLX-αを提案した.TBLX-αでは,まず個体集団の分散共分散行列にCholesky分解を適用し変換行列を求め,個体集団に変換行列を適用して変数間依存性のない空間に変換する.次に,変換空間においてを適用し子個体を求める.最後に子個体を元の空間に戻す.本手法を単峰性,多峰性,回転問題などを含む13個のテスト関数に適用した.拡張率αの調整を行うことにより変数間依存性が強い問題ではある程度の性能が得られたが,優れた性能を実現するには、さらに検討が必要であるという課題が残った. 
Research paper (scientific journal)  Joint  Particle swarm optimisation with dynamic search strategies based on rank correlation  T.Nishio, J.Kushida, A.Hara, T.Takahama  International Journal of Computational Intelligence Studies  6/ 4, 311-332  2018/01 
View details...

Research presentations
Oral presentation(general)  A New Binomial Crossover Considering Correlation Among Decision Variables for Adaptive Differential Evolution  Joint 10th International Conference on Soft Computing and Intelligent Systems and 19th International Symposium on Advanced Intelligent Systems (SCIS&ISIS2018)  2018/12/06  In population-based optimization methods such as evolutionary algorithms, various information can be obtained from the distribution of good search points. When problems with strong dependency among decision variables are optimized, a characteristic distribution, which is a thin elliptical distribution, may appear. In order to generate good children, it is necessary to change the variables simultaneously along the long axis of the elliptical distribution. A similar distribution also may appear when the search points are far from the optimal solution even in problems with independent variables. In this study, we propose a new crossover CBX which uses correlation coefficients of search points in order to detect such distribution and realizes efficient movement toward the optimal solution. The crossover points are decided so that highly correlated variables are inherited at the same time. However, if only CBX is used, the diversity of the search points tends to be lost rapidly. The adaptive control of the probability for applying CBX is also proposed. The advantage of the proposed method is shown by solving several benchmark problems. 
Oral presentation(general)  集団的降下法に対するペナルティ係数の適応的調整法の提案  京都大学数理解析研究所共同研究(公開型)「不確実性の下での意思決定の数理とその周辺」  2018/11/19 
Oral presentation(general)  Grouping of Genes According to Correlation Coefficients and Grouping-Based Crossover for Adaptive Differential Evolution  The 50th ISCIE International Symposium on Stochastic Systems Theory and Its Applications (SSS'18), extended abstract  2018/11/02  When problems with strong dependency among deci- sion variables are optimized, a characteristic distribu- tion, which is a thin elliptical distribution, may appear. In order to generate good children, it is necessary to change the variables (genes) simultaneously along the long axis of the elliptical distribution. Since binomial crossover in differential evolution determines whether each gene is crossed or not with the same probability, it is difficult to change some genes simultaneously. In this study, we propose a crossover operation GBX which uses correlation coefficients of search points in order to detect such distribution. The highly correlated genes are grouped and the genes in each group are crossed (or not crossed) simultaneously. However, if only GBX is used, the diversity of the search points tends to be lost rapidly. The adaptive control of the probability for applying GBX is also proposed. The advantage of the proposed method is shown by solving several bench- mark problems. 
Oral presentation(general)  差分進化における相関係数に基づく遺伝子のグループ化とグループ単位の交叉の提案  情報処理学会 第120回数理モデル化と問題解決研究会  2018/09/26  最適化が困難な変数間依存性の強い問題に出現する特徴的な分布として,細い楕円形状の分布がある.このような場合に優れた子個体を生成するには,楕円形状の長軸に沿って変数を同時に変更する必要がある.また,同様の分布は,変数分離型の問題において探索点集合が最適解から離れている場合にも出現する.差分進化における2項交叉は,各変数(遺伝子) について同じ確率で交叉を行うかどうかを決定しているため,特定の遺伝子を同時に交叉することは困難である.本研究では,このような形状を検出するために探索点の相関係数を利用する方法を提案する.探索点の分布から相関行列を求め,相関の強い遺伝子をグループ化し,グループ単位で遺伝子を同時に交叉する(あるいは,交叉しない).本手法を差分進化の代表的手法であるJADE に導入し,幾つかのベンチマーク問題を最適化し,性能を比較することにより,本手法の効果を調べる. 
Oral presentation(general)  変数間依存性を解消する変換を導入したブレンド交叉の提案  京都大学数理解析研究所RIMS共同研究(公開型)「不確実性の下での意思決定理論とその応用 :計画数学の展開」  2017/11/15 
View details...

Prizes
IEEE Computational Intelligence Society, CEC2010 Constrained Real-Parameter Optimisation Competition Award  2010/07 
2006 IEEE Congress on Evolutionary Computation Competition Program Award -- Nonlinear Programming  2006/07 
View details...

Allotted class
機械学習特論 
データ構造とアルゴリズムⅠ 
数理計画法 
機械学習 
View details...

Memberships of academic societies
Institute of Electronics, Information and Communication Engineers 
Association for Natural Language Processing 
Japanese Society for Artificial Intelligence 
情報処理学会 
IEEE  1999/05/01-Present 
View details...

Open lecture
模擬授業「自然から学ぶアルゴリズム-ナチュラル・コンピューティング-」  Others  2015/07-2015/07  自然界には,多くの有益なアイデアが存在しています.例えば,生物は長い年月をかけて進化し,環境に適応しています.また,単純な生物でも群としては驚くほど知的な振る舞いをします.このような現象をコンピュータ上に実現し,情報処理に応用する研究であるナチュラルコンピューティングを紹介します. 
模擬授業「自然から学ぶアルゴリズム-ナチュラル・コンピューティング-」  Others  2009/11-2009/11  自然界には,多くの有益なアイデアが存在しています.例えば,生物は長い年月をかけて進化し,環境に適応しています.また,単純な生物でも群としては驚くほど知的な振る舞いをします.このような現象をコンピュータ上に実現し,情報処理に応用する研究であるナチュラルコンピューティングを紹介します. 
模擬授業「人工知能」  Open lecture  2008/10-2008/10  人工知能は,知的活動が可能な知的機械(コンピュータ)の実現を目指す研究分野です.推論,探索などの基礎分野から音声・画像認識,自然言語理解などの応用分野まで,人間の知的活動の一部を実現するために様々な研究が行われています.本講義では,ゲーム木探索を取り上げ,オセロゲームや将棋のようなゲームにおいて,指し手をコンピュータで計算する方法について,基本的な考え方を紹介します. 
平成17年度広島市立大学情報科学部公開講座  Open lecture  2005/11-2005/11  自然から学ぶアルゴリズム -ナチュラル・コンピューティング- 
View details...