Split-and-Denoise: Protect large language model inference with local differential privacy. (arXiv:2310.09130v1 [cs.AI])

Large Language Models (LLMs) shows powerful capability in natural language
understanding by capturing hidden semantics in vector space. This process
enriches the value of the text embeddings for various downstream tasks, thereby
fostering the Embedding-as-a-Service (EaaS) business model. However, the direct
transmission of text to servers poses a largely unaddressed risk of privacy
leakage. To mitigate this issue, we introduce Split-N-Denoise (SnD), an
innovative framework that split the model to execute the token embedding layer
on the client side at …

as-a-service business differential privacy eaas hidden language language models large large language model llms local natural natural language privacy process protect servers service space text transmission understanding value