ReCaLL: Membership Inference via Relative Conditional Log-Likelihoods

2024-06-23

Conference on Empirical Methods in Natural Language Processing

AI Summary

To address the challenges of data provenance tracking and insufficient transparency in large language model (LLM) pretraining, this paper proposes a black-box membership inference attack (MIA). The core methodological innovation is the **Relative Conditional Log-Likelihood Ratio (RCLLR)**, a novel metric that assesses whether a target sample belongs to the training set by comparing its conditional probability under random versus synthetic prefixes—without requiring access to the training data or model gradients. The approach integrates conditional language modeling with both sequence-level and token-level behavioral modeling, coupled with ensemble-based inference. Evaluated on the WikiMIA benchmark, it achieves state-of-the-art performance, substantially outperforming existing black-box MIAs. Moreover, it provides the first empirical evidence that LLMs implicitly encode membership information at both the sequence and token levels—a fundamental insight into their internal representational mechanisms.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Humans and AI: Human-in-the-loop Machine Learning

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

The rapid scaling of large language models (LLMs) has raised concerns about the transparency and fair use of the data used in their pretraining. Detecting such content is challenging due to the scale of the data and limited exposure of each instance during training. We propose ReCaLL (Relative Conditional Log-Likelihood), a novel membership inference attack (MIA) to detect LLMs’ pretraining data by leveraging their conditional language modeling capabilities. ReCaLL examines the relative change in conditional log-likelihoods when prefixing target data points with non-member context. Our empirical findings show that conditioning member data on non-member prefixes induces a larger decrease in log-likelihood compared to non-member data. We conduct comprehensive experiments and show that ReCaLL achieves state-of-the-art performance on the WikiMIA dataset, even with random and synthetic prefixes, and can be further improved using an ensemble approach. Moreover, we conduct an in-depth analysis of LLMs’ behavior with different membership contexts, providing insights into how LLMs leverage membership information for effective inference at both the sequence and token level.

Problem

• Detects LLMs' pretraining data using conditional log-likelihoods
• Measures log-likelihood changes with non-member context prefixes
• Analyzes LLMs' behavior with membership contexts for better inference

Innovation

• Uses relative conditional log-likelihoods
• Leverages non-member context prefixes
• Ensemble approach improves performance