Search papers, labs, and topics across Lattice.
4
0
8
0
Ditch the hand-tuning: this new diffusion sampling method uses a learned, geometry-aware correction that's faster and more accurate.
Mapping LLM attack strategies onto a multiplex network reveals interpretable vulnerability clusters and dramatically improves red teaming efficiency.
VLA models, despite strong performance, crumble under simple paraphrasing of instructions, revealing a surprising lack of semantic grounding in robotic manipulation tasks.
Achieve up to 19.5x speedup in diffusion-prior based inverse problems by replacing slow inner loops with closed-form projections and analytic step sizes.
