LUMIA: Linear probing for Unimodal and MultiModal Membership Inference Attacks leveraging internal LLM states

TL;DR

LUMIA introduces a layer-by-layer linear probing method to detect membership inference attacks in large language models by analyzing internal activations, significantly improving detection accuracy across unimodal and multimodal tasks.

Contribution

The paper presents LUMIA, a novel linear probing approach that leverages internal LLM states for more effective membership inference attack detection, outperforming previous methods.

Findings

LUMIA achieves an average 15.71% AUC gain over prior techniques.

In 65.33% of cases, LUMIA reaches AUC > 60%.

In multimodal models, visual inputs enhance MIA detection, with 85.90% of experiments exceeding AUC > 60%.

Abstract

Large Language Models (LLMs) are increasingly used in a variety of applications, but concerns around membership inference have grown in parallel. Previous efforts focus on black-to-grey-box models, thus neglecting the potential benefit from internal LLM information. To address this, we propose the use of Linear Probes (LPs) as a method to detect Membership Inference Attacks (MIAs) by examining internal activations of LLMs. Our approach, dubbed LUMIA, applies LPs layer-by-layer to get fine-grained data on the model inner workings. We test this method across several model architectures, sizes and datasets, including unimodal and multimodal tasks. In unimodal MIA, LUMIA achieves an average gain of 15.71 % in Area Under the Curve (AUC) over previous techniques. Remarkably, LUMIA reaches AUC>60% in 65.33% of cases -- an increment of 46.80% against the state of the art. Furthermore, our approach reveals key insights, such as the model layers where MIAs are most detectable. In multimodal models, LPs indicate that visual inputs can significantly contribute to detect MIAs -- AUC>60% is reached in 85.90% of experiments.