On extending extreme learning machine to non-redundant synergy pattern based graph classification

Abstract

Graph patterns are widely used to define the feature space for building an efficient graph classification model. Synergy graph patterns refer to those graphs, where the relationships among the nodes are highly inseparable. Compared with the general graph patterns, synergy graph patterns which have much higher discriminative powers are more suitable as the classification features. Extreme Learning Machine (ELM) is a simple and efficient Single-hidden Layer Feedforward neural Networks (SLFNs) algorithm with extremely fast learning capacity. In this paper we propose the problem of extending ELM to non-redundant synergy pattern based graph classification. The graph classification framework being widely used consists of two steps, namely feature generation and classification. The first issue is how to quickly obtain significant graph pattern features from a graph database. The next step is how to effectively build a graph classification model with these graph pattern features. An efficient depth-first algorithm, called GINS, was presented to find all non-redundant synergy graph patterns. Also, based on the proposed Support Graph Vector Model (SGVM) and ELM algorithm, the graph classification model was constructed. Extensive experiments are conducted on a series of real-life datasets. The results show that GINS is more efficient than two representative competitors. Besides, when the generated graph patterns are considered as the classification features, the GINS+ELM classification accuracy can be improved much.