Context-Aware Membership Inference Attacks against Pre-trained Large Language Models

TL;DR

This paper introduces a novel membership inference attack tailored for pre-trained large language models, leveraging perplexity dynamics to effectively identify training data membership, surpassing previous methods.

Contribution

The paper proposes a new MIA method for LLMs that accounts for their generative nature and context-dependent memorization, improving attack accuracy.

Findings

Outperforms prior MIAs on LLMs

Reveals context-dependent memorization patterns

Effective in identifying training data membership

Abstract

Membership Inference Attacks (MIAs) on pre-trained Large Language Models (LLMs) aim at determining if a data point was part of the model's training set. Prior MIAs that are built for classification models fail at LLMs, due to ignoring the generative nature of LLMs across token sequences. In this paper, we present a novel attack on pre-trained LLMs that adapts MIA statistical tests to the perplexity dynamics of subsequences within a data point. Our method significantly outperforms prior approaches, revealing context-dependent memorization patterns in pre-trained LLMs.