2020/06/11 by Jinheon Baek, Baek, Jinheon, Dong Bok Lee +3 · 1 citation
Biochemistry, Genetics and Molecular Biology · Computer Science · #Advanced Graph Neural Networks #Bayesian Modeling and Causal Inference #Bioinformatics and Genomic Networks #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2006.06648
openalex publication_date 2020/06/11 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/01
Many practical graph problems, such as knowledge graph construction and\ndrug-drug interaction prediction, require to handle multi-relational graphs.\nHowever, handling real-world multi-relational graphs with Graph Neural Networks\n(GNNs) is often challenging due to their evolving nature, as new entities\n(nodes) can emerge over time. Moreover, newly emerged entities often have few\nlinks, which makes the learning even more difficult. Motivated by this\nchallenge, we introduce a realistic problem of few-shot out-of-graph link\nprediction, where we not only predict the links between the seen and unseen\nnodes as in a conventional out-of-knowledge link prediction task but also\nbetween the unseen nodes, with only few edges per node. We tackle this problem\nwith a novel transductive meta-learning framework which we refer to as Graph\nExtrapolation Networks (GEN). GEN meta-learns both the node embedding network\nfor inductive inference (seen-to-unseen) and the link prediction network for\ntransductive inference (unseen-to-unseen). For transductive link prediction, we\nfurther propose a stochastic embedding layer to model uncertainty in the link\nprediction between unseen entities. We validate our model on multiple benchmark\ndatasets for knowledge graph completion and drug-drug interaction prediction.\nThe results show that our model significantly outperforms relevant baselines\nfor out-of-graph link prediction tasks.\n