Experimental Evaluation of a GPU-based Frequent Subgraph Miner Using Synthetic Databases

Frequent subgraph mining, i.e. enumeration of subgraphs appearing frequently in graph databases, is one of the most fundamental problems in graph mining. Several optimization techniques as well as parallel implementations are developed to alleviate the problem that subgraph miners require a long computation time if target databases are huge or given frequency threshold is low. In this paper, we investigate on a GPU-based implementation of subgraph miner developed recently. In order to obtain significant insights for further performance improvements, we re-implement the GPU-based subgraph miner with a few modifications, and conduct a series of intensive experiments by using synthetic graph databases.