InfoMax Classification-Enhanced Learnable Network for Few-Shot Node Classification

Graph neural networks have a wide range of applications, such as citation networks, social and knowledge graphs. Among various graph analyses, node classification has garnered much attention. While many the recent network embedding models achieve promising performance, they usually require sufficient labeled nodes for training, which does not meet reality that only few are available in novel classes. few-shot learning is commonly employed vision language domains to address problem insufficient training samples, there still two characteristics non-Euclidean domain investigation: (1) how extract most informative class use it on testing data (2) thoroughly explore limited number support sets maximize amount information transferred query set. We propose an InfoMax Classification-Enhanced Learnable Network (ICELN) these issues, motivated by Deep (DGI), adapts principle summary representation patch node. By increasing shared between representation, ICELN can transfer maximum unlabeled enhance potential. The whole model trained using episodic method, simulates actual environment ensure meta-knowledge learned from previous experience may be used entirely new classes been studied before. Extensive experiments conducted five real-world datasets demonstrate advantages over existing methods.