From Unfamiliar to Familiar: Detecting Pre-training Data via Gradient Deviations in Large Language Models

Ruiqi Zhang, Lingxiang Wang, Hainan Zhang, Zhiming Zheng, Yanyan LanarXiv 2026 · DOI

BibTeX
@misc{zhang2026from,
  author = {Ruiqi Zhang and Lingxiang Wang and Hainan Zhang and Zhiming Zheng and Yanyan Lan},
  title = {{From Unfamiliar to Familiar: Detecting Pre-training Data via Gradient Deviations in Large Language Models}},
  year = {2026},
  eprint = {2603.04828},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2603.04828},
}

Loading related content...

Figure