Determining motif explanations for learning on graphs

This study focuses on identifying relevant and concise explanations in the machine learning context. A subgraph is considered a relevant pattern in a graph if it is important for making the prediction on the graph. Measurement of the subgraph's relevance in the graph only through the relative frequency of occurrence of the subgraphs isomorphic to the motif (traditional motif definition) is not sufficient to identify motifs that are relevant due to some functional property. Machine learning, however, exactly aims to identify functional relevance by learning properties depending on other properties varying in relevance depending on the task of prediction. Hence this extended abstract demonstrates the application of machine learning for the identification of important subgraph patterns on graphs for property prediction.