Xu Lin

YOLO-PEFT transforms the fine-tuning landscape for real-time detectors by replacing trial-and-error with structured, auditable planning, achieving notable performance gains.

StepOPSD shows that focusing on individual agent steps, rather than entire trajectories, unlocks significant performance gains in multi-turn agent reinforcement learning.

Coordinating multiple agents with a shared visual contract can dramatically improve the structural consistency and visual alignment of automatically generated scientific papers.