Self-Distillation with Meta Learning for Knowledge Graph Completion

Abstract

In this paper, we propose a self-distillation framework with meta learning (MetaSD) for knowledge graph completion with dynamic pruning, which aims to learn compressed graph embeddings and tackle the long-tail samples. Specifically, we first propose a dynamic pruning technique to obtain a small pruned model from a large source model, where the pruning mask of the pruned model could be updated adaptively per epoch after the model weights are updated. The pruned model is supposed to be more sensitive to difficult-to-memorize samples (e.g., long-tail samples) than the source model. Then, we propose a one-step meta self-distillation method for distilling comprehensive knowledge from the source model to the pruned model, where the two models co-evolve in a dynamic manner during training. In particular, we exploit the performance of the pruned model, which is trained alongside the source model in one iteration, to improve the source model's knowledge transfer ability for the next iteration via meta learning. Extensive experiments show that MetaSD achieves competitive performance compared to strong baselines, while being 10x smaller than baselines.