Search papers, labs, and topics across Lattice.
Imperial College London
1
0
0
Head-to-head benchmarking of 12 diffusion unlearning algorithms reveals that reported safety gains frequently mask severe, previously hidden degradations in generative quality and concept retention.