Tao Cheng

Chasing large token-level teacher-student gaps in agent distillation is largely misguided: local divergence fails to predict task completion, whereas unpenalized minor deviations often trigger catastrophic multi-turn failure.

Random splits in urban representation learning benchmarks inflate performance and alter model rankings, making cross-city generalization claims unreliable.

Unleashing the full potential of multimodal LLMs requires reasoning directly in the visual latent space, and this paper shows how to do it with stable policy optimization.

Reconstructing and simulating wind-driven dynamics from video is now possible with a new differentiable framework that enforces fluid dynamics laws.