Name
Abstract | ||
|---|---|---|
This paper aims to model relational data on edges of networks. We describe appro- priate Gaussian Processes (GPs) for directed, undirected, and bipartite networks. The inter-dependencies of edges can be effectively modeled by adapting the GP hyper-parameters. The framework suggests an intimate connection between link prediction and transfer learning, which were traditionally two separate research topics. We develop an efficient learning algorithm that can handle a large number of observations. The experimental results on several real-world data sets verify superior learning capacity. The goal of this paper is to design a Gaussian process (GP) (13) framework to model the depen- dence structure of networks, and to contribute an efficient algorithm to learn and predict large-scale relational data. We explicitly construct a series of parametric models indexed by their dimension- ality, and show that in the limit we obtain nonparametric GP priors consistent with the dependence of edge-wise measurements. Since the kernel matrix is on a quadratic number of edges and the computation cost is even cubic of the kernel size, we develop an efficient algorithm to reduce the computational complexity. We also demonstrate that transfer learning has an intimate connection to link prediction. Our method generalizes several recent transfer learning algorithms by additionally learning a task-specific kernel that directly expresses the dependence between tasks. |
Year | Venue | Keywords |
|---|---|---|
2007 | NIPS | transfer learning,gaussian process,parametric model,relational data,indexation,computational complexity,link analysis |
Field | DocType | Citations |
Stability (learning theory),Semi-supervised learning,Relational database,Inductive transfer,Active learning (machine learning),Computer science,Link analysis,Transfer of learning,Theoretical computer science,Gaussian process,Artificial intelligence,Machine learning | Conference | 26 |
PageRank | References | Authors |
1.57 | 14 | 2 |