Search papers, labs, and topics across Lattice.
Nanyang Technological University
2
0
5
Text-to-image models can be tricked into generating images containing malicious text with over 90% success, even when standard jailbreak methods fail.
Open-source VLN agents can nearly double their navigation success by remembering where they've been, thanks to a new hierarchical memory system.
