Prevailing methods for graphs require abundant label and edge information for learning. When data for a new task are scarce, meta-learning can learn from prior experiences and form much-needed inductive biases for fast adaption to new tasks. Here, we introduce G-Meta, a novel meta-learning algorithm for graphs. G-Meta uses local subgraphs to transfer subgraph-specific information and learn transferable knowledge faster via meta gradients. G-Meta learns how to quickly adapt to a new task using only a handful of nodes or edges in the new task and does so by learning from data points in other graphs or related, albeit disjoint label sets. G-Meta is theoretically justified as we show that the evidence for a prediction can be found in the local subgraph surrounding the target node or edge. Experiments on seven datasets and nine baseline methods show that G-Meta outperforms existing methods by up to 16.3%. Unlike previous methods, G-Meta successfully learns in challenging, few-shot learning settings that require generalization to completely new graphs and never-before-seen labels. Finally, G-Meta scales to large graphs, which we demonstrate on a new Tree-of-Life dataset comprising of 1,840 graphs, a two-orders of magnitude increase in the number of graphs used in prior work.

Weaknesses: - The main weakness of the work relates to the computational complexity of 1) computing the local subgraphs (are shortest paths computed ahead of the training process?), 2) evaluating each node's label individually. Can authors comment on the impact on training/evaluation time? - Another important missing element from the paper is the value of neighborhood size h, as well as an analysis of its influence over the model's performance. This is the key parameter of the proposed strategy and providing readers with intuitive knowledge of the value of h to use, and the robustness of the method with respect to larger or smaller neighborhoods is essential. Similarly, different hyperparameter sets are used per dataset, which is not ideal. Can authors provide insights into how performance varies with a constant set of parameters? - Certain aspects of the training set-up needs clarifying.

This paper proposes a meta-learning method for graph data. The use of the local subgraphs provides more flexibility and allows us to adopt the same framework to different scenarios. The effectiveness of the proposed method is supported by theory and experiments. However, Meta-GNN also computes the nodes representations using a sub-graph based on the number of aggregation layers. Also, the performance improvement of the proposed method compared with Meta-GNN might come from the usage of metric learning.

Prevailing methods for graphs require abundant label and edge information for learning. When data for a new task are scarce, meta-learning can learn from prior experiences and form much-needed inductive biases for fast adaption to new tasks. Here, we introduce G-Meta, a novel meta-learning algorithm for graphs. G-Meta uses local subgraphs to transfer subgraph-specific information and learn transferable knowledge faster via meta gradients. G-Meta learns how to quickly adapt to a new task using only a handful of nodes or edges in the new task and does so by learning from data points in other graphs or related, albeit disjoint label sets.