Search papers, labs, and topics across Lattice.
The Hong Kong University of Science and Technology (Guangzhou)
8
0
13
8
MLLMs can now achieve over 8% improvement in spatial reasoning for egocentric scenes by leveraging a novel Ego-element Graph for enhanced perception.
OmniCoT recalibrates the challenges of panoramic reasoning, enabling MLLMs to leverage global evidence for complex multi-step inference.
SCPO transforms reinforcement learning for LLM agents by ensuring that intermediate steps receive consistent credit, leading to remarkable performance gains on complex tasks.
Forget brittle heuristics – NestedKV's multi-timescale anomaly detection and surprise-gated routing lets you compress KV caches far more aggressively without sacrificing long-context performance.
Text watermarks can now survive even aggressive paragraph-level paraphrasing, thanks to a new self-anchoring technique that breaks the robustness-quality tradeoff.
Q-Gate dynamically optimizes keyframe selection by intelligently routing attention based on query intent, outperforming traditional methods that struggle with modality noise.
Ground-truth labels don't improve LLM reasoning, but building a consensus-based knowledge graph of reasoning steps does, boosting accuracy by 10%+.
Existing affordance prediction models fall flat when confronted with the wide-angle, distorted reality of panoramic vision, but a new training-free pipeline called PAP rises to the challenge.
