Skewed Memorization in Large Language Models: Quantification and Decomposition
This work addresses privacy and security risks for users of LLMs by analyzing memorization behaviors, though it is incremental as it builds on existing analyses.
The paper tackled the problem of skewed memorization in large language models during supervised fine-tuning, linking it to training duration, dataset size, and inter-sample similarity, and provided strategies for detection and mitigation to enhance privacy.
Memorization in Large Language Models (LLMs) poses privacy and security risks, as models may unintentionally reproduce sensitive or copyrighted data. Existing analyses focus on average-case scenarios, often neglecting the highly skewed distribution of memorization. This paper examines memorization in LLM supervised fine-tuning (SFT), exploring its relationships with training duration, dataset size, and inter-sample similarity. By analyzing memorization probabilities over sequence lengths, we link this skewness to the token generation process, offering insights for estimating memorization and comparing it to established metrics. Through theoretical analysis and empirical evaluation, we provide a comprehensive understanding of memorization behaviors and propose strategies to detect and mitigate risks, contributing to more privacy-preserving LLMs.