Knowledge about algorithm similarity is an important aspect of meta-learning, where the information gathered from previous learning tasks can be used to guide the selection of algorithms for new datasets. Usually this...
详细信息
ISBN:
(纸本)9781479956180
Knowledge about algorithm similarity is an important aspect of meta-learning, where the information gathered from previous learning tasks can be used to guide the selection of algorithms for new datasets. Usually this task is done by comparing global performance measures across different datasets or alternatively, comparing the performance of algorithms at the instance-level. In both cases, the previous similarity measures do not consider misclassification costs, and hence they neglect an important information that can be exploited in different learning contexts. In this paper we present algorithm similarity measures that deals with cost proportions and different threshold choice methods for building crisp classifiers from learned models. Experiments were performed in a meta-learning study with 50 different learning tasks. The similarity measures were adopted to cluster algorithms according to their aggregated performance on the learning tasks. The clustering process revealed similarity between algorithms under different perspectives.
Although the diversity of metaheuristic algorithms has been frequently highlighted, the similarity of these algorithms is not studied comprehensively. This work studies the similarity of metahruristic algorithms from ...
详细信息
ISBN:
(纸本)9781538694183
Although the diversity of metaheuristic algorithms has been frequently highlighted, the similarity of these algorithms is not studied comprehensively. This work studies the similarity of metahruristic algorithms from their performance perspective captured in a newly proposed fractional ranking method, which can map comprehensive performance measures into a scalar framework. The fractional ranking data is clustered using a k-medoids clustering to find similarities between algorithms. Results show that the proposed similarity analysis scheme reveals a new perspective of metaheuristic algorithms.
In the context of metaheuristic search algorithms, two recent approaches have been proposed to measure algorithm similarity. The first one is based on shared search strategies, such as mutation or selection. The secon...
详细信息
ISBN:
(纸本)9798400701207
In the context of metaheuristic search algorithms, two recent approaches have been proposed to measure algorithm similarity. The first one is based on shared search strategies, such as mutation or selection. The second one involves an empirical analysis that uses a set of performance metrics on benchmark problems. This paper explores whether high Component similarity corresponds to Performance similarity by analyzing these two similarity metrics on 9 CMA-ES variants on the COCO benchmark.
Metaheuristics are found to be efficient in different applications where the use of exact algorithms becomes short-handed. In the last decade, many of these algorithms have been introduced and used in a wide range of ...
详细信息
Metaheuristics are found to be efficient in different applications where the use of exact algorithms becomes short-handed. In the last decade, many of these algorithms have been introduced and used in a wide range of applications. Nevertheless, most of those approaches share similar components leading to a concern related to their novelty or contribution. Thus, in this paper, a pool template is proposed and used to categorize algorithm components permitting to analyze them in a structured way. We exemplify its use by means of continuous optimization metaheuristics, and provide some measures and methodology to identify their similarities and novelties. Finally, a discussion at a component level is provided in order to point out possible design differences and commonalities.
Metaheuristic Search is a successful strategy for solving optimization problems, leading to over two hundred published metaheuristic algorithms. Consequently, there is an interest in understanding the similarities bet...
详细信息
ISBN:
(数字)9783031210945
ISBN:
(纸本)9783031210938;9783031210945
Metaheuristic Search is a successful strategy for solving optimization problems, leading to over two hundred published metaheuristic algorithms. Consequently, there is an interest in understanding the similarities between metaheuristics. Previous studies have done theoretical analyses based on components and search strategies, providing insights into the relationship between different algorithms. In this paper, we argue that it is also important to consider the classes of optimization problems that the algorithms are capable of solving. To this end, we propose a method to measure the similarity between metaheuristics based on their performance on a set of optimization functions. We then use the proposed method to analyze the similarity between different algorithms as well as the similarity between the same algorithm but with different parameter settings. Our method can show if parameter settings of the same algorithm are more similar between themselves than to other algorithms and suggest a clustering based on the performance profile.
暂无评论