SnapE -- Training Snapshot Ensembles of Link Prediction Models

TL;DR

This paper introduces SnapE, a method for training snapshot ensembles of link prediction models in knowledge graphs, which improves robustness and performance without increasing training time.

Contribution

It adapts snapshot ensemble techniques to link prediction in knowledge graphs and proposes a novel negative sampling method using previous snapshots.

Findings

Outperforms single models across multiple datasets

Maintains constant training time while improving accuracy

Demonstrates robustness of ensemble approach in link prediction

Abstract

Snapshot ensembles have been widely used in various fields of prediction. They allow for training an ensemble of prediction models at the cost of training a single one. They are known to yield more robust predictions by creating a set of diverse base models. In this paper, we introduce an approach to transfer the idea of snapshot ensembles to link prediction models in knowledge graphs. Moreover, since link prediction in knowledge graphs is a setup without explicit negative examples, we propose a novel training loop that iteratively creates negative examples using previous snapshot models. An evaluation with four base models across four datasets shows that this approach constantly outperforms the single model approach, while keeping the training time constant.