paper-with-me

홈 › Papers

TrojFM: Resource-efficient Backdoor Attacks against Very Large Foundation Models

2024-05-27 · Yuzhou. Nie, Yanting. Wang, Jinyuan. Jia, Michael J. De Lucia, Nathaniel D. Bastian, Wenbo. Guo, Dawn. Song

One key challenge in backdoor attacks against large foundation models is the resource limits. Backdoor attacks usually require retraining the target model, which is impractical for very large foundation models. Existing backdoor attacks are mainly designed for supervised classifiers or small foundation models (e.g., BERT). None of these attacks has successfully compromised a very large foundation model, such as Llama-3-70B, especially with limited computational resources. In this paper, we propose TrojFM, a novel backdoor attack tailored for very large foundation models. Our primary technical contribution is the development of a novel backdoor injection method. This method forces a backdoored model to generate similar hidden representations for poisoned inputs regardless of their actual semantics. Our approach injects such backdoors by fine-tuning only a very small proportion of model parameters. This enables TrojFM to efficiently launch downstream task-agnostic backdoor attacks against very large foundation models under limited computational resources. Moreover, we optimize the fine-tuning process with our customized QLoRA technique, enabling launching our attack via only~\textit{one A100 GPU}. Furthermore, we design a new trigger injection method to ensure our attack stealthiness. Through extensive experiments, we first demonstrate that TrojFM can launch effective backdoor attacks against widely used large GPT-style models without jeopardizing their normal functionalities (and outperforming existing attacks on BERT-style models). Furthermore, we show that TrojFM is resilient to SOTA defenses and is insensitive to changes in key hyper-parameters. Finally, we conduct a resource analysis to quantify that our method can significantly save computational and memory costs compared to existing backdoor attacks.

📄 PDF Abstract BibTeX arXiv:2405.16783

Code (1)

ucsb-mlsec/troj_fm 공식 구현 pytorch

Tasks

Backdoor AttackGPU

Similar Papers 제목 키워드 기반

ONION: A Simple and Effective Defense Against Textual Backdoor Attacks

2020-11-20 · EMNLP 2021 11 · Fanchao Qi, Yangyi Chen, Mukai Li, Yuan YAO 외

Backdoor attacks are a kind of emergent training-time threat to deep neural networks (DNNs). They can manipulate the output of DNNs and possess high insidiousness. In the field of natural language processing, some attack…

Backdoor Attackbackdoor defense

Look, Listen, and Attack: Backdoor Attacks Against Video Action Recognition

2023-01-03 · Hasan Abed Al Kader Hammoud, Shuming Liu, Mohammed Alkhrashi, Fahad Albalawi 외

Deep neural networks (DNNs) are vulnerable to a class of attacks called "backdoor attacks", which create an association between a backdoor trigger and a target label the attacker is interested in exploiting. A backdoored…

Action RecognitionTemporal Action Localization

Fine-Pruning: Defending Against Backdooring Attacks on Deep Neural Networks

2018-05-30 · Kang Liu, Brendan Dolan-Gavitt, Siddharth Garg

Deep neural networks (DNNs) provide excellent performance across a wide range of classification tasks, but their training requires high computational resources and is often outsourced to third parties. Recent work has sh…

On Certifying Robustness against Backdoor Attacks via Randomized Smoothing

2020-02-26 · Binghui Wang, Xiaoyu Cao, Jinyuan Jia, Neil Zhenqiang Gong

Backdoor attack is a severe security threat to deep neural networks (DNNs). We envision that, like adversarial examples, there will be a cat-and-mouse game for backdoor attacks, i.e., new empirical defenses are developed…

Backdoor Attack

Detecting Backdoors in Deep Text Classifiers

2022-10-11 · You Guo, Jun Wang, Trevor Cohn

Deep neural networks are vulnerable to adversarial attacks, such as backdoor attacks in which a malicious adversary compromises a model during training such that specific behaviour can be triggered at test time by attach…

Data Poisoningtext-classificationText Classification