Zhen Lei

Forget training, just nudge your text embeddings: RGSE closes the open-vocabulary object detection gap under distribution shift by directly and efficiently adapting text embeddings at test time.

Generate realistic talking avatars 120x faster by distilling multi-step diffusion models into a single-step generator, without sacrificing quality.

Forget storing old face forgeries: this method distills the *difference* between real and fake distributions to replay past forgery cues, achieving better continual learning with less memory and improved privacy.

Existing forgery detection methods struggle with AI-generated CT images, but a new spatial-frequency detector achieves superior generalization across diverse generative models.

A principled framework for General World Models reveals the limitations of current systems and the architectural requirements for future progress.