Our paper “Erase or Hide? Suppressing Spurious Unlearning Neurons for Robust Unlearning” and “Bilinear relational structure fixes reversal curse and enables consistent model editing” have been accepted to ICLR 2026.

Updated: