Search papers, labs, and topics across Lattice.
Sharif University of Technology
2
0
4
Unlearning nudity in Stable Diffusion can severely degrade its ability to compose scenes correctly, showing that current unlearning methods struggle to balance concept removal with broader generative capabilities.
Despite advances in vision-language models, they still fail at rebus puzzles, highlighting a critical gap in cognitive visual reasoning that neither scaling nor in-context learning can fix.
