Search papers, labs, and topics across Lattice.
The Ohio State University
1
0
3
Machine unlearning methods may only *appear* successful because they simply misalign the last-layer features with the classifier, leaving the underlying representations largely unchanged and vulnerable to recovery.
