Locate-then-edit knowledge editing fails at scale due to over-optimized activations and growing edit-matrix norms, and adding MPES early stopping plus a Frobenius norm constraint mitigates the damage.
These are the facts that the model learnt through its pre-training
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Lifelong Knowledge Editing requires Better Regularization
Locate-then-edit knowledge editing fails at scale due to over-optimized activations and growing edit-matrix norms, and adding MPES early stopping plus a Frobenius norm constraint mitigates the damage.