摘要：The subpath kernel is a useful positive definite kernel, which takes arbitrary rooted trees as input, no matter whether they are ordered or unordered, We first show that the subpath kernel can exhibit excellent classification performance in combination with SVM through an intensive experiment. Secondly, we develop a theory of irreducible trees, and then, using it as a rigid mathematical basis, reconstruct a bottom-up linear-time algorithm for the subtree kernel, which is a correction of an algorithm well-known in the literature. Thirdly, we show a novel top-down algorithm, with which we can realize a linear-time parallel-computing algorithm to compute the subpath kernel.
Loading...
