Search papers, labs, and topics across Lattice.
1
0
3
A black-box defense that reduces harmful outputs in text-to-image models by 37.7% without needing any model retraining or internal access.
