Designing similarity measures for discrete data structures, and their applications to machine learning

对于离散数据结构（如字符串、树和图形），称为核函数的相似性度量允许对非数值变量进行统计分析和机器学习。在这项研究中，一个新的和一般的框架设计核函数的离散数据结构已经开发。该框架是Haussler卷积核的推广，通过计算两个结构之间所有可能的结构对应。此外，树的相似性度量的多样性已被开发的基础上计算它们的共同子结构。

The similarity measures called kernel functions for discrete data structures such as strings, trees, and graphs allow for statistical analysis and machine learning for non-numerical variables. In this study, a novel and general framework for designing kernel functions for discrete data structures has been developed. The framework is a generalization of Haussler's convolution kernel by counting all possible structural correspondences between two structures. Moreover, a diversity of similarity measures for trees have been developed based on counting their common substructures.