Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhou, Tong, Duan, Shijin, Liu, Gaowen, Fleming, Charles, Kompella, Ramana Rao, Ren, Shaolei, Xu, Xiaolin
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2503.13224
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913740891357184
author Zhou, Tong
Duan, Shijin
Liu, Gaowen
Fleming, Charles
Kompella, Ramana Rao
Ren, Shaolei
Xu, Xiaolin
author_facet Zhou, Tong
Duan, Shijin
Liu, Gaowen
Fleming, Charles
Kompella, Ramana Rao
Ren, Shaolei
Xu, Xiaolin
contents Pre-trained models are valuable intellectual property, capturing both domain-specific and domain-invariant features within their weight spaces. However, model extraction attacks threaten these assets by enabling unauthorized source-domain inference and facilitating cross-domain transfer via the exploitation of domain-invariant features. In this work, we introduce **ProDiF**, a novel framework that leverages targeted weight space manipulation to secure pre-trained models against extraction attacks. **ProDiF** quantifies the transferability of filters and perturbs the weights of critical filters in unsecured memory, while preserving actual critical weights in a Trusted Execution Environment (TEE) for authorized users. A bi-level optimization further ensures resilience against adaptive fine-tuning attacks. Experimental results show that **ProDiF** reduces source-domain accuracy to near-random levels and decreases cross-domain transferability by 74.65\%, providing robust protection for pre-trained models. This work offers comprehensive protection for pre-trained DNN models and highlights the potential of weight space manipulation as a novel approach to model security.
format Preprint
id arxiv_https___arxiv_org_abs_2503_13224
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ProDiF: Protecting Domain-Invariant Features to Secure Pre-Trained Models Against Extraction
Zhou, Tong
Duan, Shijin
Liu, Gaowen
Fleming, Charles
Kompella, Ramana Rao
Ren, Shaolei
Xu, Xiaolin
Cryptography and Security
Machine Learning
Pre-trained models are valuable intellectual property, capturing both domain-specific and domain-invariant features within their weight spaces. However, model extraction attacks threaten these assets by enabling unauthorized source-domain inference and facilitating cross-domain transfer via the exploitation of domain-invariant features. In this work, we introduce **ProDiF**, a novel framework that leverages targeted weight space manipulation to secure pre-trained models against extraction attacks. **ProDiF** quantifies the transferability of filters and perturbs the weights of critical filters in unsecured memory, while preserving actual critical weights in a Trusted Execution Environment (TEE) for authorized users. A bi-level optimization further ensures resilience against adaptive fine-tuning attacks. Experimental results show that **ProDiF** reduces source-domain accuracy to near-random levels and decreases cross-domain transferability by 74.65\%, providing robust protection for pre-trained models. This work offers comprehensive protection for pre-trained DNN models and highlights the potential of weight space manipulation as a novel approach to model security.
title ProDiF: Protecting Domain-Invariant Features to Secure Pre-Trained Models Against Extraction
topic Cryptography and Security
Machine Learning
url https://arxiv.org/abs/2503.13224