Weight Poisoning Attacks on Pre-trained Models

Abstract

Recently, NLP has seen a surge in the usage of large pre-trained models. Users download weights of models pre-trained on large datasets, then fine-tune the weights on a task of their choice. This raises the question of whether downloading untrusted pre-trained weights can pose a security threat. In this paper, we show that it is possible to construct ``weight poisoning'' attacks where pre-trained weights are injected with vulnerabilities that expose ``backdoors'' after fine-tuning, enabling the attacker to manipulate the model prediction simply by injecting an arbitrary keyword. We show that by applying a regularization method, which we call RIPPLe, and an initialization procedure, which we call Embedding Surgery, such attacks are possible even with limited knowledge of the dataset and fine-tuning procedure. Our experiments on sentiment classification, toxicity detection, and spam detection show that this attack is widely applicable and poses a serious threat. Finally, we outline practical defenses against such attacks. Code to reproduce our experiments is available at https://github.com/neulab/RIPPLe.

Published as a long paper at ACL 2020

References in corpus (3)

• Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoning
• Security Evaluation of Pattern Classifiers under Attack
• Deceiving Google's Perspective API Built for Detecting Toxic Comments

Cited by in corpus (4)

• Backdoor Attacks and Countermeasures on Deep Learning: A Comprehensive Review
• Can Adversarial Weight Perturbations Inject Neural Backdoors?
• Poison Attacks against Text Datasets with Conditional Adversarially Regularized Autoencoder
• Systematic Attack Surface Reduction For Deployed Sentiment Analysis Models