Data-Free Privacy-Preserving for LLMs via Model Inversion and Selective Unlearning
This addresses privacy concerns for LLM users by enabling data-free unlearning, though it is an incremental improvement over existing unlearning methods.
The paper tackles the problem of removing sensitive personally identifiable information (PII) from large language models without access to training data, achieving effective PII removal while maintaining model utility as demonstrated on the AI4Privacy PII-Masking dataset with Pythia models.
Large language models (LLMs) exhibit powerful capabilities but risk memorizing sensitive personally identifiable information (PII) from their training data, posing significant privacy concerns. While machine unlearning techniques aim to remove such data, they predominantly depend on access to the training data. This requirement is often impractical, as training data in real-world deployments is commonly proprietary or inaccessible. To address this limitation, we propose Data-Free Selective Unlearning (DFSU), a novel privacy-preserving framework that removes sensitive PII from an LLM without requiring its training data. Our approach first synthesizes pseudo-PII through language model inversion, then constructs token-level privacy masks for these synthetic samples, and finally performs token-level selective unlearning via a contrastive mask loss within a low-rank adaptation (LoRA) subspace. Extensive experiments on the AI4Privacy PII-Masking dataset using Pythia models demonstrate that our method effectively removes target PII while maintaining model utility.