Yun Li

Current LLMs struggle to translate visual scenes into executable geometric functions, failing even basic trajectory generation tasks.

Explicitly linking multimodal patient data to biomedical knowledge graphs unlocks significant gains for clinical LLMs, with performance plummeting when these crucial connections are removed.

CARE boosts first-sighting recall in adaptive LiDAR sensing by strategically reserving budget for unexplained camera detections, outperforming conventional methods.

Current video generation benchmarks overlook crucial aspects of physical plausibility and temporal coherence, highlighting the need for holistic evaluation metrics like PhyScore.

Overcome the bandwidth bottleneck in remote sensing with a collaborative edge-cloud approach that transmits structural priors, enabling high-fidelity super-resolution and boosting downstream perception tasks even under extreme compression.

Achieve near-lossless 2x speedups in long video diffusion by learning to attend only to the most relevant parts of the video, guided by motion and semantic content.

Autonomous driving models can get a 30% performance boost simply by restructuring text inputs to highlight causal relationships between intent and environmental constraints.

Forget full intrinsic decomposition: LightCtrl achieves state-of-the-art single-image relighting by learning sparse, physically meaningful latent proxies from limited PBR data.