The authors approximate influence functions with Fisher matrices and zeroth-order optimization to unlearn knowledge graph triples fast, claiming large gains over graph unlearning baselines.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Efficient Knowledge Graph Unlearning with Zeroth-order Information
The authors approximate influence functions with Fisher matrices and zeroth-order optimization to unlearn knowledge graph triples fast, claiming large gains over graph unlearning baselines.