检索结果-内蒙古大学图书馆

3rd European Network Intelligence Conference (ENIC)

作者： Przewozniczek, Michal Wroclaw Univ Sci & Technol Dept Computat Intelligence Wroclaw Poland

ISBN: (纸本)9781509034550

Coevolution is a well-known way of population diversity preservation in the field of Evolutionary Computation. Usually, the methods that use coevolution are effective in solving hard optimization problems, because they are less likely to stuck in local optima. Therefore, they are more capable of reaching a breakthrough, find new, promising regions in the solution search space and propose solutions of better quality. One of the main drawbacks of coevolution is the high computation cost of many coevolving subpopulations maintenance. If the coevolution is used, the determination of subpopulations number is crucial and may be very challenging. Therefore, the strategies of dynamic subpopulation number control (SDSNC). When SDSNC is used, the number of coevolving subpopulations is automatically adjusted to the current solution search state and changes during the method run. Therefore, this paper analyzes, the application of the recent SDSNC propositions to the multi-population method designed for solving the Routing and Spectrum Allocation with Joint Anycast and Unicast (RSA-JAU).

关键词： Coevolution dynamic subpopulation Management dynamic subpopulation number control Evolutionary Algorithms Linkage Learning Elastic Optical Networks Routing and Spectrum Allocation problem Anycasting

来源：评论

学校读者我要写书评

暂无评论

Splitting the fitness and penalty factor for temporal diversity increase in practical problem solving

引用

EXPERT SYSTEMS WITH APPLICATIONS 2020年 145卷 113126-113126页

作者： Przewozniczek, Michal Witold Datta, Rituparna Walkowiak, Krzysztof Komarnicki, Marcin Michal Wroclaw Univ Sci & Technol Dept Computat Intelligence Wybrzeze Wyspianskiego 27 PL-50370 Wroclaw Poland Univ S Alabama Dept Comp Sci 150 Jaguar Dr Mobile AL 36688 USA Wroclaw Univ Sci & Technol Dept Syst & Comp Networks Wybrzeze Wyspianskiego 27 PL-50370 Wroclaw Poland

In this paper, we propose an optimization system based on an evolutionary algorithm developed to solve a real-life optimization problem related to the optimization of the computer networks. In particular, we focus on an NP-hard flow allocation problem in computer networks related to network survivability and addressing various constraints specific to computer networks. The constrained problems are usually handled by evolutionary methods by the introduction of the so-called penalty factor. However, such techniques raise difficulties regardless of their simple working principle. The more oppressive the penalty factor is, the more likely the evolutionary method is to stuck in the local optima with feasible solutions and propose final results of low-quality. On the other hand, a low penalty factor may lead the method to regions of high-rated solutions that violate the constraints and are useless due to their infeasibility. Therefore, we address the issue of penalty handling by proposing Ranking-based Fitness and Penalty Weighting (RFPW). RFPW is inspired by a theory-driven penalty parameter estimation using a bi-objective and weighted sum approach. It separates the objective and penalty functions and removes the necessity of hand-made weighting. RFPW was introduced to the recent proposition of an evolutionary method dedicated to solving hard NP-complete practical problem. The performed experiments related to the flow allocation problem confirm that employing RFPW may lead to a significant improvement in results quality. In our opinion, the proposed RFPW framework of the penalty function handling has the potential to be adapted to a wide range of optimization systems that utilize the idea of penalty function. RFPW allows the optimization method to reduce or increase the strength of a convergence automatically. Therefore, RFPW has the potential of results quality improvement and may replace expensive commonly-used diversity preservation techniques (e.g., island models). (

关键词： Penalty factor separation Diversity preservation Evolutionary computation dynamic subpopulation number control Messy-coding Genetic algorithm

来源：评论

学校读者我要写书评

暂无评论

subpopulation initialization driven by linkage learning for dealing with the Long-Way-To-Stuck effect

引用

INFORMATION SCIENCES 2020年第0期521卷 62-80页

作者： Przewozniczek, Michal Witold Wroclaw Univ Sci & Technol Dept Computat Intelligence Fac Comp Sci & Management Wybrzeze Wyspianskiego 27 PL-50370 Wroclaw Poland

The maintenance of many subpopulations is an important technique employed in evolutionary methods. However, the use of a multi-population approach has its drawbacks. Among all, it requires spending high amounts of available resources. Therefore, the methods that dynamically manage the number of subpopulations gain an increasing interest due to their capability of adjusting the subpopulation number to their current state. They are shown to be capable of reaching excellent results. In this paper, we identify the Long-Way-To-Stuck effect and show it on the base of the practical, NP-complete problem. Such a phenomenon occurs when a randomly initialized population of an evolutionary method must be processed through many iterations before it is incapable of improving the best-found solution. If so, then a large amount of computational resources must be spent on subpopulation initialization, which may turn multi-population methods ineffective. Therefore, in this paper, we propose the Linkage Learning-Driven subpopulation Initialization (LLDSI) that limits the costs of subpopulation initialization and significantly improves the effectiveness methods dynamically managing the subpopulation number. (C) 2020 Elsevier Inc. All rights reserved.

关键词： Genetic algorithms Island models dynamic subpopulation number control Linkage learning Long-Way-To-Stuck effect

来源：评论

学校读者我要写书评

暂无评论

The transformation of the k-Shortest Steiner trees search problem into binary dynamic problem for effective evolutionary methods application

引用

INFORMATION SCIENCES 2019年 479卷 1-19页

作者： Przewozniczek, Michal Witold Walkowiak, Krzysztof Sen, Arunabha Komarnicki, Marcin Lechowicz, Piotr Wroclaw Univ Sci & Technol Fac Comp Sci & Management Dept Computat Intelligence Wybrzeze Wyspianskiego 27 PL-50370 Wroclaw Poland Wroclaw Univ Sci & Technol Fac Elect Dept Syst & Comp Networks Wybrzeze Wyspianskiego 27 PL-50370 Wroclaw Poland Wroclaw Univ Sci & Technol ENGINE Ctr Wroclaw Poland Arizona State Univ Sch Comp Informat & Decis Syst Engn Tempe AZ USA

Evolutionary methods are well-known tools used for solving hard computational problems. In this paper, we consider k-Shortest Steiner Trees (kSST) problem appearing in a diverse set of domains, e.g., multicast tree construction in communication networks in general, and optical networks in particular. The kSST is relatively new and has not been widely investigated in the literature. Thus, only a few algorithms have been proposed, each requiring significant resources amount and long execution times, partially following from the NP-hard nature of the problem. The kSST problem solution is a set of different trees, where each single tree can be easily represented using a genotype-encoding. However, encoding the tree set may be challenging, as each tree must be unique. Especially, in most applications the number of trees is large, resulting with the genotype containing high number of necessary genes. Thus, in this paper, we propose a transformation of the kSST problem into a dynamic problem, and when applied in the evolutionary method, a single individual encodes a single tree instead of a whole tree set. The results of extensive numerical experiments executed on two representative network topologies show that the proposed transformation improves the effectiveness of all considered state-of-the-art evolutionary methods. (C) 2018 Elsevier Inc. All rights reserved.

关键词： K shortest Steiner trees Parameter-less population pyramid Island models dynamic problems dynamic subpopulation number control Linkage Learning

来源：评论

学校读者我要写书评

暂无评论

Universal strategy of dynamic subpopulation number management in practical network optimization problems

引用

APPLIED SOFT COMPUTING 2019年 82卷 105592-000页

作者： Przewozniczek, Michal Witold Goscien, Roza Wroclaw Univ Sci & Technol Fac Comp Sci & Management Dept Computat Intelligence Wybrzeze Wyspianskiego 27 PL-50370 Wroclaw Poland Wroclaw Univ Sci & Technol Fac Elect Dept Syst & Comp Networks Wybrzeze Wyspianskiego 27 PL-50370 Wroclaw Poland

Multi-population approach, also known as Island Model (IM), is a well-known technique improving diversity preservation. It is employed by evolutionary and Swarm Intelligence methods. The diversity increased through the maintenance of many subpopulations, leads to finding higher quality solutions. Since IM is computationally expensive, it is important to keep the balance between the diversity maintenance and the convergence. Therefore, this paper concentrates on proposing the new strategy of dynamic subpopulation number control (SDSNC). When SDSNC is used, the number of subpopulations may change during the method run, adjusting the subpopulation number to the current method state, making such method highly-effective. In this paper, we show that the SDSNC strategies that were proposed for two up-to-date, NP-hard, practical problems are not universal. Therefore, we propose Store and Use Later (SUL) strategy that applied to problem-dedicated methods outperforms or at least equals two other strategies for both considered problems. (C) 2019 Elsevier B.V. All rights reserved.

关键词： Evolutionary algorithms Island Models dynamic subpopulation number control Linkage learning Elastic Optical Networks Routing and spectrum allocation Flow optimization

来源：评论

学校读者我要写书评

暂无评论

Problem Encoding Allowing Cheap Fitness Computation of Mutated Individuals

Problem Encoding Allowing Cheap Fitness Computation of Mutat...

引用

IEEE Congress on Evolutionary Computation (CEC)

作者： Przewozniczek, Michal Witold Wroclaw Univ Sci & Technol Dept Computat Intelligence Wybrzeze Wyspianskiego 27 PL-50370 Wroclaw Poland

ISBN: (纸本)9781509046010

In the Evolutionary Computation field, it is frequent to assume that a computation load necessary for fitness value computation is, at least, similar for all possible cases. The main objective of this paper is to show that the above assumption is frequently false. Therefore, the examples of evolutionary methods that use problem encoding which allows for significant optimization of the fitness computation process are pointed out and analyzed. The definition of Problem Encoding Allowing Cheap Fitness Computation of Mutated Individuals (PEACh) is proposed. Another objective of the paper is to start a discussion concerning the computation load measurement in the evolutionary computation field. As shown, the Fitness Function Evaluation number is not always a fair measure and may be significantly affected by the quality of method implementation.

关键词： Fitness Computation Cost Computation Load Measurement Fitness Computation Optimization Evolutionary Algorithms Problem Encoding Linkage Learning Messy Coding dynamic subpopulation number control

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：