Opt-Out: Investigating Entity-Level Unlearning for Large Language Models via Optimal Transport

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Choi, Minseok, Rim, Daniel, Lee, Dohyun, Choo, Jaegul
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915329336147968
author Choi, Minseok
Rim, Daniel
Lee, Dohyun
Choo, Jaegul
author_facet Choi, Minseok
Rim, Daniel
Lee, Dohyun
Choo, Jaegul
contents Instruction-following large language models (LLMs), such as ChatGPT, have become widely popular among everyday users. However, these models inadvertently disclose private, sensitive information to their users, underscoring the need for machine unlearning techniques to remove selective information from the models. While prior work has focused on forgetting small, random subsets of training data at the instance-level, we argue that real-world scenarios often require the removal of an entire user data, which may require a more careful maneuver. In this study, we explore entity-level unlearning, which aims to erase all knowledge related to a target entity while preserving the remaining model capabilities. To address this, we introduce Opt-Out, an optimal transport-based unlearning method that utilizes the Wasserstein distance from the model's initial parameters to achieve more effective and fine-grained unlearning. We also present the first Entity-Level Unlearning Dataset (ELUDe) designed to evaluate entity-level unlearning. Our empirical results demonstrate that Opt-Out surpasses existing methods, establishing a new standard for secure and adaptable LLMs that can accommodate user data removal requests without the need for full retraining.
format Preprint
id arxiv_https___arxiv_org_abs_2406_12329
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Opt-Out: Investigating Entity-Level Unlearning for Large Language Models via Optimal Transport
Choi, Minseok
Rim, Daniel
Lee, Dohyun
Choo, Jaegul
Computation and Language
Instruction-following large language models (LLMs), such as ChatGPT, have become widely popular among everyday users. However, these models inadvertently disclose private, sensitive information to their users, underscoring the need for machine unlearning techniques to remove selective information from the models. While prior work has focused on forgetting small, random subsets of training data at the instance-level, we argue that real-world scenarios often require the removal of an entire user data, which may require a more careful maneuver. In this study, we explore entity-level unlearning, which aims to erase all knowledge related to a target entity while preserving the remaining model capabilities. To address this, we introduce Opt-Out, an optimal transport-based unlearning method that utilizes the Wasserstein distance from the model's initial parameters to achieve more effective and fine-grained unlearning. We also present the first Entity-Level Unlearning Dataset (ELUDe) designed to evaluate entity-level unlearning. Our empirical results demonstrate that Opt-Out surpasses existing methods, establishing a new standard for secure and adaptable LLMs that can accommodate user data removal requests without the need for full retraining.
title Opt-Out: Investigating Entity-Level Unlearning for Large Language Models via Optimal Transport
topic Computation and Language
url https://arxiv.org/abs/2406.12329