Privacy-Preserving Large Language Models

The field of natural language processing is moving towards developing privacy-preserving large language models (LLMs) that can protect sensitive user information while maintaining their utility. Researchers are exploring innovative techniques to balance privacy and performance, such as localized LLMs, modular separation of language intent parsing, and split learning. These approaches aim to eliminate the need for LLMs to process encrypted prompts, enabling practical deployment of privacy-preserving LLM-centric services. Noteworthy papers in this area include:

  • Agentic-PPML, which proposes a novel framework to make PPML in LLMs practical by employing a general-purpose LLM for intent understanding and delegating cryptographically secure inference to specialized models.
  • PRvL, which presents a comprehensive analysis of LLMs as privacy-preserving PII Redaction systems and provides an open-source suite of fine-tuned models and evaluation tools for general-purpose PII Redaction.

Sources

From Patient Burdens to User Agency: Designing for Real-Time Protection Support in Online Health Consultations

Agentic Privacy-Preserving Machine Learning

VFLAIR-LLM: A Comprehensive Framework and Benchmark for Split Learning of LLMs

Current State in Privacy-Preserving Text Preprocessing for Domain-Agnostic NLP

PRvL: Quantifying the Capabilities and Risks of Large Language Models for PII Redaction

Built with on top of