Liang Li

UniDDT achieves a groundbreaking balance between multimodal understanding and generation, outperforming existing models in both tasks with enhanced semantic coherence.

VAIC enables humanoid robots to perform complex tasks in real-world settings without the need for perfect state observability, significantly advancing their practical deployment.

DocFormFlow boosts document formatting accuracy while slashing token consumption, revealing that target localization is the key to success.

PINNs can now reconstruct 3D magnetic fields with an order of magnitude greater accuracy than previously thought possible, even in complex experimental settings.

MARL-optimized collaboration between large and small models in LEO satellites slashes service delays by nearly a third.

Robot manipulation models trained on mostly VR data can perform as well as those trained on real-world data, but at 1/20th the cost.

Achieve significantly more realistic and lip-synced movie dubbing by modeling the cognitive processes of professional actors with a novel diffusion transformer architecture.