Abstract: Multiconlitron is a general framework for constructing piecewise linear classifiers. This paper introduces a typical nearest point method, i.e. Schlesinger-Kozinec (SK) algorithm to compute the separating hyperplane between two convex polytopes for linearly separable data sets. By using SK algorithm, this paper constructs conlitron and multiconlitron to solve convexly separable and commonly separable problems, respectively. Experiments on both synthetic and real data sets show that the presented method has a good classification performance. Comparing with some other piecewise linear classifiers verifies the effectiveness of the presented method.

Key words: multiconlitron, SK algorithm, general framework, piecewise linear classifier

摘要： 组合凸线性感知器（multiconlitron）是用来构造分片线性分类器的一个通用理论框架。基于此框架，引入一种典型的凸包间最近点求解方法——Schlesinger-Kozinec（SK）算法, 来计算线性可分样本间的分类超平面；然后构造可处理凸可分数据的凸线性感知器和处理叠可分数据的组合凸线性感知器。在人工合成数据集和标准数据集上的实验说明，所构造的感知器具有良好的分类性能，与其他典型分片线性分类器的对比也说明了该方法的有效性。

关键词: 组合凸线性感知器, SK算法, 通用框架, 分片线性分类器