Preserving Utility in Continual LLM Unlearning via Orthogonal Projections
Orthogonal projections keep LLM utility stable during repeated unlearning.
Explainer · on the wire
The explainer for this story is still on the wire.
The claim, the mechanism, and the figures land here once generated.
Source · arXiv