Yue Wang

Fréchet Distance, previously deemed impractical for training, unlocks surprisingly high-fidelity image generation when optimized in representation space with decoupled batch sizes.

LLMs, like humans, exhibit a "frequency bias," performing better when prompted and fine-tuned with more common textual expressions.

Finally, a model that can generate realistic videos of human-object interactions, like pouring liquid, by conditioning on actions, text, and images.