Ziyi Cai

Predicting task-relevant latent futures, not raw pixels, unlocks efficient and effective long-horizon VLA control, outperforming reactive policies by over 11 percentage points.

SeMoCo achieves unprecedented reconstruction accuracy by decoupling semantic meaning from kinematic details in motion representations.

Achieving a 1-1/e guarantee in language generation through randomization reveals new strategies for balancing novelty and validity in output.

ChartArena reveals that even top proprietary models struggle with diagrammatic structures, exposing critical gaps in current chart parsing capabilities.

Ventricular dysfunction can be surprisingly well-predicted in a zero-shot manner from ECG diagnostic probabilities, suggesting a structured encoding of cardiac function within these representations.