Shaofeng Li

Dr. Shaofeng Li


Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
22 July 2025 in arXiv
Verified
Preprint
Open access
PDF file
TDTian Dong
Tian Dong
YMYan Meng
Yan Meng
SLShaofeng Li
Shaofeng Li
GCGuoxing Chen
Guoxing Chen
ZLZhen Liu
Zhen Liu
HZHaojin Zhu
Haojin Zhu
Large Language Models (LLMs) are increasingly integrated into daily routines, yet they raise significant privacy and safety concerns. Recent research proposes collaborative inference, which outsources the early-layer inference to ensure data locality, and introduces model safety auditing based on inner neuron patterns. Both techniques expose the LLM's Internal States (ISs), which are traditionally considered irreversible to inputs due to optimization challenges and the highly abstract representations in deep layers. In this work, we challenge this assumption by proposing four inversion attacks that significantly improve the semantic similarity and token matching rate of inverted inputs. Specifically, we first develop two white-box optimization-based attacks tailored for low-depth and high-depth ISs. These attacks avoid local minima convergence, a limitation observed in prior work, through a two-phase inversion process. Then, we extend our optimization attack under more practical black-box weight access by leveraging the transferability between the source and the derived LLMs. Additionally, we introduce a generation-based attack that treats inversion as a translation task, employing an inversion model to reconstruct inputs. Extensive evaluation of short and long prompts from medical consulting and coding assistance datasets and 6 LLMs validates the effectiveness of our inversion attacks. Notably, a 4,112-token long medical consulting prompt can be nearly perfectly inverted with 86.88 F1 token matching from the middle layer of Llama-3 model. Finally, we evaluate four practical defenses that we found cannot perfectly prevent ISs inversion and draw conclusions for future mitigation design.

10 January 2025 in arXiv
Verified
Preprint
Open access
PDF file
YSYunmeng Shu
Yunmeng Shu
SLShaofeng Li
Shaofeng Li
TDTian Dong
Tian Dong
YMYan Meng
Yan Meng
HZHaojin Zhu
Haojin Zhu
Personalized Large Language Models (LLMs) have become increasingly prevalent, showcasing the impressive capabilities of models like GPT-4. This trend has also catalyzed extensive research on deploying LLMs on mobile devices. Feasible approaches for such edge-cloud deployment include using split learning. However, previous research has largely overlooked the privacy leakage associated with intermediate representations transmitted from devices to servers. This work is the first to identify model inversion attacks in the split learning framework for LLMs, emphasizing the necessity of secure defense. For the first time, we introduce mutual information entropy to understand the information propagation of Transformer-based LLMs and assess privacy attack performance for LLM blocks. To address the issue of representations being sparser and containing less information than embeddings, we propose a two-stage attack system in which the first part projects representations into the embedding space, and the second part uses a generative model to recover text from these embeddings. This design breaks down the complexity and achieves attack scores of 38%-75% in various scenarios, with an over 60% improvement over the SOTA. This work comprehensively highlights the potential privacy risks during the deployment of personalized LLMs on the edge side.

15 June 2025 in arXiv
Verified
Preprint
Open access
PDF file
LCLu Chen
Lu Chen
HYHan Yang
Han Yang
HWHu Wang
Hu Wang
YCYuxin Cao
Yuxin Cao
SLShaofeng Li
Shaofeng Li
YLYuan Luo
Yuan Luo
Adversarial examples have attracted significant attention over the years, yet understanding their frequency-based characteristics remains insufficient. In this paper, we investigate the intriguing properties of adversarial examples in the frequency domain for the image classification task, with the following key findings. (1) As the high-frequency components increase, the performance gap between adversarial and natural examples becomes increasingly pronounced. (2) The model performance against filtered adversarial examples initially increases to a peak and declines to its inherent robustness. (3) In Convolutional Neural Networks, mid- and high-frequency components of adversarial examples exhibit their attack capabilities, while in Transformers, low- and mid-frequency components of adversarial examples are particularly effective. These results suggest that different network architectures have different frequency preferences and that differences in frequency components between adversarial and natural examples may directly influence model robustness. Based on our findings, we further conclude with three useful proposals that serve as a valuable reference to the AI model security community.

11 April 2025 in Proceedings of the AAAI Conference on Artificial Intelligence
Verified
Journal Article
Open access
PDF file
LCLu Chen
Lu Chen
SLShaofeng Li
Shaofeng Li
BHBenhao Huang
Benhao Huang
FYFan Yang
Fan Yang
ZLZheng Li
Zheng Li
YLYuan Luo
Yuan Luo
Existing works have extensively studied adversarial examples, which are minimal perturbations that can mislead the output of deep neural networks (DNNs) while remaining imperceptible to humans. However, in this work, we reveal the existence of a harmless perturbation space, in which perturbations drawn from this space, regardless of their magnitudes, leave the network output unchanged when applied to inputs. Essentially, the harmless perturbation space emerges from the usage of non-injective functions (linear or non-linear layers) within DNNs, enabling multiple distinct inputs to be mapped to the same output. For linear layers with input dimensions exceeding output dimensions, any linear combination of the orthogonal bases of the nullspace of the parameter consistently yields no change in their output. For non-linear layers, the harmless perturbation space may expand, depending on the properties of the layers and input samples. Inspired by this property of DNNs, we solve for a family of general perturbation spaces that are redundant for the DNN's decision, and can be used to hide sensitive data and serve as a means of model identification. Our work highlights the distinctive robustness of DNNs (i.e., consistency under large magnitude perturbations) in contrast to adversarial examples (vulnerability for small noises).

11 December 2025 in arXiv
Verified
Preprint
Open access
PDF file
HYHan Yang
Han Yang
SLShaofeng Li
Shaofeng Li
TDTian Dong
Tian Dong
XXXiangyu Xu
Xiangyu Xu
GLGuangchi Liu
Guangchi Liu
ZLZhen Ling
Zhen Ling
Deep Neural Networks (DNNs), as valuable intellectual property, face unauthorized use. Existing protections, such as digital watermarking, are largely passive; they provide only post-hoc ownership verification and cannot actively prevent the illicit use of a stolen model. This work proposes a proactive protection scheme, dubbed ``Authority Backdoor," which embeds access constraints directly into the model. In particular, the scheme utilizes a backdoor learning framework to intrinsically lock a model's utility, such that it performs normally only in the presence of a specific trigger (e.g., a hardware fingerprint). But in its absence, the DNN's performance degrades to be useless. To further enhance the security of the proposed authority scheme, the certifiable robustness is integrated to prevent an adaptive attacker from removing the implanted backdoor. The resulting framework establishes a secure authority mechanism for DNNs, combining access control with certifiable robustness against adversarial attacks. Extensive experiments on diverse architectures and datasets validate the effectiveness and certifiable robustness of the proposed framework.

9 September 2025 in arXiv
Verified
Preprint
Open access
PDF file
CLChongqing Lei
Chongqing Lei
ZLZhen Ling
Zhen Ling
XXXiangyu Xu
Xiangyu Xu
SLShaofeng Li
Shaofeng Li
GLGuangchi Liu
Guangchi Liu
JLJunzhou Luo
Junzhou Luo
Microcontroller units (MCUs) are widely used in embedded devices due to their low power consumption and cost-effectiveness. MCU firmware controls these devices and is vital to the security of embedded systems. However, performing dynamic security analyses for MCU firmware has remained challenging due to the lack of usable execution environments -- existing dynamic analyses cannot run on physical devices (e.g., insufficient computational resources), while building emulators is costly due to the massive amount of heterogeneous hardware, especially peripherals. Our work is based on the insight that MCU peripherals can be modeled in a two-fold manner. At the structural level, peripherals have diverse implementations but we can use a limited set of primitives to abstract peripherals because their hardware implementations are based on common hardware concepts. At the semantic level, peripherals have diverse functionalities. However, we can use a single unified semantic model to describe the same kind of peripherals because they exhibit similar functionalities. Building on this, we propose FlexEmu, a flexible MCU peripheral emulation framework. Once semantic models are created, FlexEmu automatically extracts peripheral-specific details to instantiate models and generate emulators accordingly. We have successfully applied FlexEmu to model 12 kinds of MCU peripherals. Our evaluation on 90 firmware samples across 15 different MCU platforms shows that the automatically generated emulators can faithfully replicate hardware behaviors and achieve a 98.48% unit test passing rate, outperforming state-of-the-art approaches. To demonstrate the implications of FlexEmu on firmware security, we use the generated emulators to fuzz three popular RTOSes and uncover 10 previously unknown bugs.