Auditing and Mitigating Privacy Leakage in Cloud-Edge Collaborative Decoding

Abstract

Applications such as personalized assistance and proprietary document analysis require large language models (LLMs) to generate outputs from private data. Yet powerful LLMs typically cannot be deployed on the resource-constrained devices where private data resides, and uploading private data to cloud-hosted LLMs exposes sensitive information. Recent work addresses this tension with a cloud-edge collaborative decoding paradigm, where private data are kept on the edge with a small language model (

Categories

Framework mappings

Suggested from the entry's categories.

@misc{zhang2026auditing,
  title = {{Auditing and Mitigating Privacy Leakage in Cloud-Edge Collaborative Decoding}},
  author = {Ke-Jia Zhang and Tianyuan Zou and Zi-Xuan Gu and Yang Liu},
  year = {2026},
  month = aug,
  eprint = {2608.29111},
  archivePrefix = {arXiv},
  url = {https://www.semanticscholar.org/paper/96dedb871e68ffa16687ef79c2cff3a2ca3b1a92}
}