Weighted data contraction based on information content

推广了经典的k-最近邻熵估计方法，提出了一种计算效率高的加权数据信息含量估计方法。为了将我们的基于距离的方法应用于各种数据集，在多核学习和实时建模的框架中考虑了距离度量学习方法。这些方法的有效性证实了聚类问题和集成学习的真实世界的数据。

Generalizing the classical k-nearest neighbor method for entropy estimation, an computationally efficient method for estimating information contents of weighted data is proposed. For utilizing our distance-based method to various kinds of data sets, distance metric learning methods are considered in the framework of multiple kernel learning and just-in-time modeling. Validity of those proposed methods are confirmed by clustering problems and ensemble learning of real-world data.