Timezone: »

Graph Policy Network for Transferable Active Learning on Graphs
Shengding Hu · Zheng Xiong · Meng Qu · Xingdi Yuan · Marc-Alexandre Côté · Zhiyuan Liu · Jian Tang

Thu Dec 10 09:00 PM -- 11:00 PM (PST) @ Poster Session 6 #1845

Graph neural networks (GNNs) have been attracting increasing popularity due to their simplicity and effectiveness in a variety of fields. However, a large number of labeled data is generally required to train these networks, which could be very expensive to obtain in some domains. In this paper, we study active learning for GNNs, i.e., how to efficiently label the nodes on a graph to reduce the annotation cost of training GNNs. We formulate the problem as a sequential decision process on graphs and train a GNN-based policy network with reinforcement learning to learn the optimal query strategy. By jointly training on several source graphs with full labels, we learn a transferable active learning policy which can directly generalize to unlabeled target graphs. Experimental results on multiple datasets from different domains prove the effectiveness of the learned policy in promoting active learning performance in both settings of transferring between graphs in the same domain and across different domains.

Author Information

Shengding Hu (Tsinghua University)
Zheng Xiong (Tsinghua University / University of Oxford)
Meng Qu (Mila)
Xingdi Yuan (Microsoft Research)
Marc-Alexandre Côté (Microsoft Research)
Zhiyuan Liu (Tsinghua University)
Jian Tang (Mila)

More from the Same Authors