Effective Population and Training Data Partitioning in Parallel Distributed Evolutionary Knowledge Acquisition

为了从数值数据中获取基于规则的知识，已经提出了进化知识获取。该方法的主要问题是当我们将其应用于大数据时需要巨大的计算成本。本研究提出了进化知识获取的并行分布式实现，其中群体和训练数据分别分为子群体和训练数据子集。计算实验表明，在不降低泛化能力的情况下，可以大幅降低计算成本。还检查了并行分布式实现的各种规范的效果。

Evolutionary knowledge acquisition has beenproposed in order to obtain rule-based knowledge from numerical data. The main problem of this method is that huge computation cost is necessarywhen we apply it to large data. This study proposes parallel distributed implementation of evolutionary knowledge acquisition where both a population and training data are divided into subpopulations and training data subsets, respectively. The computational experiments show that the computational cost can be drastically reduced without the deterioration of the generalization ability. The effects of various specifications for parallel distributed implementation are also examined.