Di Wang

This paper theoretically analyzes Continuous Adversarial Training (CAT) for LLMs using In-context Learning (ICL) theory, proving that embedding space perturbations effectively enhance robustness against token-space jailbreaks and proposing a singular value regularization method for improvement.

This paper introduces CoLA, a framework demonstrating that subset training, while efficient, introduces new and potentially greater privacy risks by leaking information about both data membership and the selection process itself.

The paper introduces NaRA, a noise-aware LoRA technique that dynamically adapts fine-tuning parameters based on the noise level during diffusion, significantly improving the performance of Diffusion LLMs.

The paper proposes a novel zeroth-order optimization framework to enhance the robustness of LLM safety alignment, showing that few refinement steps can significantly improve safety while maintaining utility.

The paper introduces Science Earth, a planet-scale scientific runtime that enables diverse, siloed AI capabilities to connect and collaborate dynamically, demonstrating that scientific discovery can become a distributed, self-correcting process.