1 paper
Hojun Jeong, Gyunyeop Kim, Sangwoo Kang
Fine-tuning-based knowledge editing is simple and architecture-agnostic, but standard cross-entropy increases the edited target probability without explicitly constraining changes…