On Graph Kernels: Hardness Results and Efficient Alternatives

Abstract

Abstract. As most 'real-world' data is structured, research in kernel methods has begun investigating kernels for various kinds of structured data. One of the most widely used tools for modeling structured data are graphs. An interesting and important challenge is thus to investigate kernels on instances that are represented by graphs. So far, only very specific graphs such as trees and strings have been considered. This paper investigates kernels on labeled directed graphs with general structure. It is shown that computing a strictly positive definite graph kernel is at least as hard as solving the graph isomorphism problem. It is also shown that computing an inner product in a feature space indexed by all possible graphs, where each feature counts the number of subgraphs isomorphic to that graph, is NP-hard. On the other hand, inner products in an alternative feature space, based on walks in the graph, can be computed in polynomial time. Such kernels are defined in this paper.

The Pharmacophore Kernel for Virtual Screening with Support Vector Machines

“…In this section we show that the pharmacophore kernel can be seen as an extension of the walk-count graph kernels (28) to the 3D representation of molecules. The walk-count graph kernel is based on the representation of a molecule m as a labeled graph m = (V, E), defined by a set of vertices V, a set of edges E ⊂ V × V connecting pairs of vertices, and a labeling function l : V ∪ E → A, assigning a label l(x) in an alphabet A to any vertex or edge x.…”

Predicting measurements at unobserved locations in an electrical transmission system

“…With the flexible definition in equation (7) two special cases of interest are discussed by Gärtner;Flach and Wrobel (2003). Firstly, the exponential kernel with λ(t) := β t t!…”

“…Secondly, the geometric kernel with λ(t) := β t and |β| < 1 employs the geometric series ∞ t=0 β t in equation (7). Gärtner;Flach and Wrobel (2003) show that the geometric series of a matrix only converge if β < 1 a with a := min {|N(s)|, s ∈ {1, . .…”

Using Graph-Kernels to Represent Semantic Information in Text Classification

“…Gartner et al [8] prove that the computation of a kernel function able to completely recognise graph structure is NP-hard and introduce a walk based kernel function that computes in polynomial time including both previous kernels as special cases. This kernel, known as product graph kernel is based on the concept of the direct product graph counting walks on that graph.…”