Search papers, labs, and topics across Lattice.
University of Electronic Science and Technology
1
0
2
Unlearning sensitive information in LLMs can be achieved without sacrificing performance, with ADU preserving up to 98% of model utility while effectively forgetting.
