Exploring Memorization in Fine-tuned Language Models

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zeng, Shenglai, Li, Yaxin, Ren, Jie, Liu, Yiding, Xu, Han, He, Pengfei, Xing, Yue, Wang, Shuaiqiang, Tang, Jiliang, Yin, Dawei
Format: Preprint
Veröffentlicht: 2023
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911782455476224
author Zeng, Shenglai
Li, Yaxin
Ren, Jie
Liu, Yiding
Xu, Han
He, Pengfei
Xing, Yue
Wang, Shuaiqiang
Tang, Jiliang
Yin, Dawei
author_facet Zeng, Shenglai
Li, Yaxin
Ren, Jie
Liu, Yiding
Xu, Han
He, Pengfei
Xing, Yue
Wang, Shuaiqiang
Tang, Jiliang
Yin, Dawei
contents Large language models (LLMs) have shown great capabilities in various tasks but also exhibited memorization of training data, raising tremendous privacy and copyright concerns. While prior works have studied memorization during pre-training, the exploration of memorization during fine-tuning is rather limited. Compared to pre-training, fine-tuning typically involves more sensitive data and diverse objectives, thus may bring distinct privacy risks and unique memorization behaviors. In this work, we conduct the first comprehensive analysis to explore language models' (LMs) memorization during fine-tuning across tasks. Our studies with open-sourced and our own fine-tuned LMs across various tasks indicate that memorization presents a strong disparity among different fine-tuning tasks. We provide an intuitive explanation of this task disparity via sparse coding theory and unveil a strong correlation between memorization and attention score distribution.
format Preprint
id arxiv_https___arxiv_org_abs_2310_06714
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Exploring Memorization in Fine-tuned Language Models
Zeng, Shenglai
Li, Yaxin
Ren, Jie
Liu, Yiding
Xu, Han
He, Pengfei
Xing, Yue
Wang, Shuaiqiang
Tang, Jiliang
Yin, Dawei
Artificial Intelligence
Computation and Language
Machine Learning
Large language models (LLMs) have shown great capabilities in various tasks but also exhibited memorization of training data, raising tremendous privacy and copyright concerns. While prior works have studied memorization during pre-training, the exploration of memorization during fine-tuning is rather limited. Compared to pre-training, fine-tuning typically involves more sensitive data and diverse objectives, thus may bring distinct privacy risks and unique memorization behaviors. In this work, we conduct the first comprehensive analysis to explore language models' (LMs) memorization during fine-tuning across tasks. Our studies with open-sourced and our own fine-tuned LMs across various tasks indicate that memorization presents a strong disparity among different fine-tuning tasks. We provide an intuitive explanation of this task disparity via sparse coding theory and unveil a strong correlation between memorization and attention score distribution.
title Exploring Memorization in Fine-tuned Language Models
topic Artificial Intelligence
Computation and Language
Machine Learning
url https://arxiv.org/abs/2310.06714