Ziqi Yang

LLMs can expose sensitive entity-related information through subtle interrogations, achieving up to 97% accuracy in membership inference.

Fragmented privacy patches are insufficient for Embodied AI: a unified, lifecycle-level approach is needed to prevent systemic privacy leakage in real-world deployments.

Robots can now learn complex skills from a single human demonstration while maintaining safety and intuitive interaction, thanks to a novel layered control framework.