Yan Xu

U-TTT adapts dynamically during inference, achieving state-of-the-art PET image denoising even under challenging distribution shifts.

UniPET adapts to varying dose reduction factors in PET imaging without sacrificing quality, outperforming traditional models that rely on fixed assumptions.

Forget end-to-end training: MedVol-R1 leverages reinforcement learning to decouple evidence grounding from volumetric delineation, achieving state-of-the-art performance in 3D medical image segmentation by grounding clinical reasoning to verifiable 2D evidence.

Stop wasting idle compute: ProAct agents anticipate user needs and proactively gather information, slashing task completion time and hallucinations.

LLM agents struggle to generalize from experience to reusable skills, often performing worse than simply replaying past trajectories, revealing a critical gap in current abstraction methods.

Forget handcrafted prompts: a hierarchical multi-agent framework turns diffusion models into coherent storytelling engines by globally optimizing for semantic coherence.