Yujin Huang

Yujin Huang, Terry Yue Zhuo, Qiongkai Xu et al.

Large-scale language models have achieved tremendous success across various natural language processing (NLP) applications. Nevertheless, language models are vulnerable to backdoor attacks, which inject stealthy triggers into models for steering them to undesirable behaviors. Most existing backdoor attacks, such as data poisoning, require further (re)training or fine-tuning language models to learn the intended backdoor patterns. The additional training process however diminishes the stealthiness of the attacks, as training a language model usually requires long optimization time, a massive amount of data, and considerable modifications to the model parameters. In this work, we propose Training-Free Lexical Backdoor Attack (TFLexAttack) as the first training-free backdoor attack on language models. Our attack is achieved by injecting lexical triggers into the tokenizer of a language model via manipulating its embedding dictionary using carefully designed rules. These rules are explainable to human developers which inspires attacks from a wider range of hackers. The sparse manipulation of the dictionary also habilitates the stealthiness of our attack. We conduct extensive experiments on three dominant NLP tasks based on nine language models to demonstrate the effectiveness and universality of our attack. The code of this work is available at https://github.com/Jinxhy/TFLexAttack.

Yujin Huang, Zhi Zhang, Qingchuan Zhao et al.

On-device deep learning (DL) has rapidly gained adoption in mobile apps, offering the benefits of offline model inference and user privacy preservation over cloud-based approaches. However, it inevitably stores models on user devices, introducing new vulnerabilities, particularly model-stealing attacks and intellectual property infringement. While system-level protections like Trusted Execution Environments (TEEs) provide a robust solution, practical challenges remain in achieving scalable on-device DL model protection, including complexities in supporting third-party models and limited adoption in current mobile solutions. Advancements in TEE-enabled hardware, such as NVIDIA's GPU-based TEEs, may address these obstacles in the future. Currently, watermarking serves as a common defense against model theft but also faces challenges here as many mobile app developers lack corresponding machine learning expertise and the inherent read-only and inference-only nature of on-device DL models prevents third parties like app stores from implementing existing watermarking techniques in post-deployment models. To protect the intellectual property of on-device DL models, in this paper, we propose THEMIS, an automatic tool that lifts the read-only restriction of on-device DL models by reconstructing their writable counterparts and leverages the untrainable nature of on-device DL models to solve watermark parameters and protect the model owner's intellectual property. Extensive experimental results across various datasets and model structures show the superiority of THEMIS in terms of different metrics. Further, an empirical investigation of 403 real-world DL mobile apps from Google Play is performed with a success rate of 81.14%, showing the practicality of THEMIS.

Ye Sang, Yujin Huang, Shuo Huang et al.

The increasing popularity of deep learning (DL) models and the advantages of computing, including low latency and bandwidth savings on smartphones, have led to the emergence of intelligent mobile applications, also known as DL apps, in recent years. However, this technological development has also given rise to several security concerns, including adversarial examples, model stealing, and data poisoning issues. Existing works on attacks and countermeasures for on-device DL models have primarily focused on the models themselves. However, scant attention has been paid to the impact of data processing disturbance on the model inference. This knowledge disparity highlights the need for additional research to fully comprehend and address security issues related to data processing for on-device models. In this paper, we introduce a data processing-based attacks against real-world DL apps. In particular, our attack could influence the performance and latency of the model without affecting the operation of a DL app. To demonstrate the effectiveness of our attack, we carry out an empirical study on 517 real-world DL apps collected from Google Play. Among 320 apps utilizing MLkit, we find that 81.56\% of them can be successfully attacked. The results emphasize the importance of DL app developers being aware of and taking actions to secure on-device models from the perspective of data processing.

Zijian Wang, Shuo Huang, Yujin Huang et al.

In recent years, on-device deep learning has gained attention as a means of developing affordable deep learning applications for mobile devices. However, on-device models are constrained by limited energy and computation resources. In the mean time, a poisoning attack known as sponge poisoning has been developed.This attack involves feeding the model with poisoned examples to increase the energy consumption during inference. As previous work is focusing on server hardware accelerators, in this work, we extend the sponge poisoning attack to an on-device scenario to evaluate the vulnerability of mobile device processors. We present an on-device sponge poisoning attack pipeline to simulate the streaming and consistent inference scenario to bridge the knowledge gap in the on-device setting. Our exclusive experimental analysis with processors and on-device networks shows that sponge poisoning attacks can effectively pollute the modern processor with its built-in accelerator. We analyze the impact of different factors in the sponge poisoning algorithm and highlight the need for improved defense mechanisms to prevent such attacks on on-device deep learning applications.

Yujin Huang, Han Hu, Chunyang Chen

Deep learning has shown its power in many applications, including object detection in images, natural-language understanding, and speech recognition. To make it more accessible to end users, many deep learning models are now embedded in mobile apps. Compared to offloading deep learning from smartphones to the cloud, performing machine learning on-device can help improve latency, connectivity, and power consumption. However, most deep learning models within Android apps can easily be obtained via mature reverse engineering, while the models' exposure may invite adversarial attacks. In this study, we propose a simple but effective approach to hacking deep learning models using adversarial attacks by identifying highly similar pre-trained models from TensorFlow Hub. All 10 real-world Android apps in the experiment are successfully attacked by our approach. Apart from the feasibility of the model attack, we also carry out an empirical study that investigates the characteristics of deep learning models used by hundreds of Android apps on Google Play. The results show that many of them are similar to each other and widely use fine-tuning techniques to pre-trained models on the Internet.