Target unbiased meta-learning for graph classification

Abstract Even though numerous works focus on the few-shot learning issue by combining meta-learning, there are still limits to traditional graph classification problems. The antecedent algorithms directly extract features from samples, and do not take into account preference of trained model previously “seen” targets. In order overcome aforementioned issues, an effective strategy with training unbiased meta-learning algorithm was developed in this paper, which sorted out problems target under paradigm. First, interactive attention extraction module as a supplement feature employed, improved separability vectors, reduced for certain target, remarkably generalization ability new task. Second, neural network used fully mine relationship between samples constitute structures complete image tasks at node level, greatly enhanced accuracy classification. A series experimental studies were conducted validate proposed methodology, where semisupervised problem has been effectively solved. It also proved that our better than methods real-world datasets.