Extremely fast graph integration for semi-supervised learning via Gaussian fields with Neumann approximation

Abstract

With the rapid growth in data availability, it has become more important to utilize multiple data sources containing different but complementary information for a given task. Using multiple graphs can technically be interpreted as finding the optimal combination of each graph. There have been various approaches for graph integration or graph fusion, but most of them have suffered from scalability issues as data size increases due to long computation time. This makes them difficult to utilize in the current trend of data size becoming huge. To circumvent this difficulty, our approach introduces a fast graph integration method based on semi-supervised learning (SSL), which incorporates the Neumann approximation during the maximum likelihood estimation process. Empirical studies show that the proposed method significantly reduces computation time by at least a factor of two compared to state-of-the-art methods, while still performing competitively with other methods. This advantage becomes more apparent as the size of the data increases, since the complexity of the proposed method depends mostly on the number of graphs to be integrated and not on the number of nodes, unlike other methods. Experimental results demonstrate the scalability and efficiency of the proposed method for graph integration.