Search papers, labs, and topics across Lattice.
University of California, Santa Barbara
4
0
9
Spatial attention in VLMs is nearly irrelevant to accuracy, with self-consistency emerging as the true indicator of reliability.
Current VideoQA models falter in understanding complex narratives, but StoryVideoQA and PlotTree redefine how we tackle deep video comprehension.
LLMs can dangerously misinterpret cues from users with eating disorders, leading to harmful advice that could exacerbate their conditions.
Spot poisoned LoRA adapters without running them: a weight-space analysis achieves 97% accuracy in detecting backdoors, even when the trigger is unknown.
