Structure-enhanced meta-learning for few-shot graph classification

Graph classification is a highly impactful task that plays crucial role in myriad of real-world applications such as molecular property prediction and protein function prediction. Aiming to handle the new classes with limited labeled graphs, few-shot graph has become bridge existing solutions practical usage. This work explores potential metric-based meta-learning for solving classification. We highlight importance considering structural characteristics solution propose novel framework which explicitly considers global structure local input graph. An implementation upon GIN, named SMF-GIN, tested on two datasets, Chembl TRIANGLES, where extensive experiments validate effectiveness proposed method. The constructed fill gap lacking large-scale benchmark evaluation, released together SMF-GIN at: https://github.com/jiangshunyu/SMF-GIN.