Image de couverture de livre

Transferable Embedding Inversion Attack: Uncovering Privacy Risks in Text Embeddings without Model Queries

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Huang, Yu-Hsiang, Tsai, Yuche, Hsiao, Hsiang, Lin, Hong-Yi, Lin, Shou-De
Format:	Preprint
Publié:	2024
Sujets:	Cryptography and Security Computation and Language Machine Learning
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

_version_	1866929674473439232
author	Huang, Yu-Hsiang Tsai, Yuche Hsiao, Hsiang Lin, Hong-Yi Lin, Shou-De
author_facet	Huang, Yu-Hsiang Tsai, Yuche Hsiao, Hsiang Lin, Hong-Yi Lin, Shou-De
contents	This study investigates the privacy risks associated with text embeddings, focusing on the scenario where attackers cannot access the original embedding model. Contrary to previous research requiring direct model access, we explore a more realistic threat model by developing a transfer attack method. This approach uses a surrogate model to mimic the victim model's behavior, allowing the attacker to infer sensitive information from text embeddings without direct access. Our experiments across various embedding models and a clinical dataset demonstrate that our transfer attack significantly outperforms traditional methods, revealing the potential privacy vulnerabilities in embedding technologies and emphasizing the need for enhanced security measures.
format	Preprint
id	arxiv_https___arxiv_org_abs_2406_10280
institution	arXiv
publishDate	2024
record_format	arxiv
spellingShingle	Transferable Embedding Inversion Attack: Uncovering Privacy Risks in Text Embeddings without Model Queries Huang, Yu-Hsiang Tsai, Yuche Hsiao, Hsiang Lin, Hong-Yi Lin, Shou-De Cryptography and Security Computation and Language Machine Learning This study investigates the privacy risks associated with text embeddings, focusing on the scenario where attackers cannot access the original embedding model. Contrary to previous research requiring direct model access, we explore a more realistic threat model by developing a transfer attack method. This approach uses a surrogate model to mimic the victim model's behavior, allowing the attacker to infer sensitive information from text embeddings without direct access. Our experiments across various embedding models and a clinical dataset demonstrate that our transfer attack significantly outperforms traditional methods, revealing the potential privacy vulnerabilities in embedding technologies and emphasizing the need for enhanced security measures.
title	Transferable Embedding Inversion Attack: Uncovering Privacy Risks in Text Embeddings without Model Queries
topic	Cryptography and Security Computation and Language Machine Learning
url	https://arxiv.org/abs/2406.10280