This paper proposes a novel approach to few-shot learning using graph neural networks (GNNs). The authors cast the few-shot learning problem as a supervised message-passing task on a partially observed graphical model, where nodes represent images and edges represent trainable similarity kernels. This framework generalizes several existing few-shot learning models and is shown to achieve state-of-the-art performance on datasets like Omniglot and Mini-Imagenet with fewer parameters. The GNN model is also extended to semi-supervised and active learning scenarios, demonstrating its versatility and effectiveness in handling 'relational' tasks. The paper includes detailed experimental results and a discussion of future directions, highlighting the potential of GNNs in meta-learning and related fields.This paper proposes a novel approach to few-shot learning using graph neural networks (GNNs). The authors cast the few-shot learning problem as a supervised message-passing task on a partially observed graphical model, where nodes represent images and edges represent trainable similarity kernels. This framework generalizes several existing few-shot learning models and is shown to achieve state-of-the-art performance on datasets like Omniglot and Mini-Imagenet with fewer parameters. The GNN model is also extended to semi-supervised and active learning scenarios, demonstrating its versatility and effectiveness in handling 'relational' tasks. The paper includes detailed experimental results and a discussion of future directions, highlighting the potential of GNNs in meta-learning and related fields.