Development of highly-scalable ILP systems

(1)该方法基于遗传算法（GA）的一种变体共生进化，以提高未知样本分类的预测精度。我们假设遗传算法中结果的多样性增加了对未知数据的适应度。我们开发了一个ILP系统，称为ILP/SE，它在假设搜索任务中使用共生进化，在其他任务中使用Progol的学习算法。ILP/SE利用在重复执行中获得的多个假设，以多数判断未知数据的类别。利用诱变数据集进行了实验，验证了ILP/SE的性能。(2)我们提出了一种方法来选择需要检查的事实，以得出更准确的假设。该方法采用溯因法选择事实，然后将考试结果加入背景知识。我们称这种方法为主动背景知识选择，因为它类似于数据挖掘中的主动数据选择。最后，给出了一个实证实验的结果，并讨论了该方法的有效性。(3)我们的研究目的是利用现有的已知具有成就和解释能力的分类规则，提高分类精度，发现有助于我们修改旧的基于知识的分类规则的知识。我们首先预测给定分类器的错误分类。如果预测分类规则的结果是正确的，我们接受它。如果被预测为错误分类，则使用ILP获得的新分类规则选择新的类别标签。我们将该方法应用于英语句子词性标注，这是ILP应用最成功的领域之一。

(1)The method is based on symbiotic evolution, a variant of genetic algorithm(GA), for improving the predictive accuracy in classifying unknown example. We postulate that the diversity of the results in GA increases the fitness to unknown data. We have developed an ILP system called ILP/SE, which uses symbiotic evolution for the hypothesis search task and uses the learning algorithm of Progol for the other task. ILP/SE judges the class of unknown data by majority using multiple hypothesises obtained in repeated execution. Experiments were conducted to show the performance of ILP/SE using the mutagenesis dataset. (2)We propose a method to choose facts which should be examined for inducing a more accurate hypothesis. The proposed method uses abduction to choose the facts and then adds the results of examinations to background knowledge. We call this method active background-knowledge selection, since it is analogous to active data selection in data mining. Finally, we show the result of an empirical experiment and discuss the effectiveness of our method. (3)The purposes of our research are using existing classification rule which is known to have achievement and explanation power, improving the classification accuracy, and discovering knowledge which helps us for modifying the old knowledge-based classification rule. We firstly predict misclassifications of a given classifier. If a result of the classification rule is predicted to be correct, we accept it. If it is predicted to be misclassification, we choose a new class label using a new classification rule acquired by ILP. We apply this method to Part-of-Speech(POS) tagging in English sentences, which is one of the most successful field for ILP applications.