Reproduce “Adversarial Robustness of Representation Learning for Knowledge Graphs” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Peru Bhardwaj

Research at Trinity College Dublin developed novel data poisoning attacks that reveal significant adversarial vulnerabilities in Knowledge Graph Embeddings, demonstrating how strategic modifications to training data can efficiently degrade KGE model predictive performance for missing link prediction.

Key insights

  • Knowledge Graph Embedding models exhibit significant adversarial vulnerabilities, with simpler influence-based deletion attacks proving highly effective and computationally efficient.
  • Targeting KGEs' inherent inductive biases, particularly the symmetry inference pattern, yields potent adversarial additions, underscoring this as a major point of exploit.
  • KGE vulnerability is influenced by dataset characteristics such as sparsity and the presence of specific relation inference patterns, indicating varied robustness across different knowledge graphs.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.