Reversibility-Verified De-identification for Cloud-Local LLM Inference: A Locally Certified Dehydrate-Rehydrate Loop with Layered Assurance (DR-SL)
本文针对「云-端协同 LLM 推理」中的隐私取舍问题提出 DR-SL 框架（Dehydrate-Rehydrate with Self-Learning loop，脱水-再水化自学习闭环）。现实困境在于：企业希望把敏感用户数据留在本地，却又