Gao Huang

Bypassing final-layer perturbations can significantly enhance reasoning capabilities in aligned LLMs, achieving better performance with zero memory overhead.

MemoryVLA++ achieves up to 28% performance gains in robotic manipulation tasks by integrating memory and imagination, transforming how robots handle temporal dependencies.

Achieving comparable text-to-image quality with a linearized model that accelerates inference by up to 1.47 times, all while leveraging pretrained weights.

Decomposing complex reasoning problems into verifiable subproblems unlocks significant performance gains in LLM reasoning, especially on hard problems previously stuck in gradient dead zones.

Decoupling visual perception from motor control in robot learning yields a 27% performance boost and better generalization.