Zhao Yang

Uniformly sampling frames in video LLMs is leaving crucial temporal information on the cutting room floor: GroundVTS selectively attends to the most informative segments, substantially boosting grounding performance.

LinkVLA tackles the language-action misalignment problem in autonomous driving by unifying language and action tokens in a shared space, leading to faster and more accurate instruction following.

Diffusion models can generate realistic DNA sequences, outperforming autoregressive models in regulatory element generation by a large margin.

Forget scaling sequence length: carefully integrating proximal epigenomic signals is the key to accurate gene expression prediction, outperforming long-sequence models.