Wenzhi Chen

Chuer Yu, Xuhong Zhang, Yuxuan Duan et al.

Despite the fact that DeepFake forgery detection algorithms have achieved impressive performance on known manipulations, they often face disastrous performance degradation when generalized to an unseen manipulation. Some recent works show improvement in generalization but rely on features fragile to image distortions such as compression. To this end, we propose Diff-ID, a concise and effective approach that explains and measures the identity loss induced by facial manipulations. When testing on an image of a specific person, Diff-ID utilizes an authentic image of that person as a reference and aligns them to the same identity-insensitive attribute feature space by applying a face-swapping generator. We then visualize the identity loss between the test and the reference image from the image differences of the aligned pairs, and design a custom metric to quantify the identity loss. The metric is then proved to be effective in distinguishing the forgery images from the real ones. Extensive experiments show that our approach achieves high detection performance on DeepFake images and state-of-the-art generalization ability to unknown forgery methods, while also being robust to image distortions.

Chengkun Wei, Wenlong Meng, Zhikun Zhang et al.

Prompt-tuning has emerged as an attractive paradigm for deploying large-scale language models due to its strong downstream task performance and efficient multitask serving ability. Despite its wide adoption, we empirically show that prompt-tuning is vulnerable to downstream task-agnostic backdoors, which reside in the pretrained models and can affect arbitrary downstream tasks. The state-of-the-art backdoor detection approaches cannot defend against task-agnostic backdoors since they hardly converge in reversing the backdoor triggers. To address this issue, we propose LMSanitator, a novel approach for detecting and removing task-agnostic backdoors on Transformer models. Instead of directly inverting the triggers, LMSanitator aims to invert the predefined attack vectors (pretrained models' output when the input is embedded with triggers) of the task-agnostic backdoors, which achieves much better convergence performance and backdoor detection accuracy. LMSanitator further leverages prompt-tuning's property of freezing the pretrained model to perform accurate and fast output monitoring and input purging during the inference phase. Extensive experiments on multiple language models and NLP tasks illustrate the effectiveness of LMSanitator. For instance, LMSanitator achieves 92.8% backdoor detection accuracy on 960 models and decreases the attack success rate to less than 1% in most scenarios.

Yi Shi, Wenlong Meng, Zhenyuan Guo et al.

With the rapid rise of social media and Internet culture, memes have become a popular medium for expressing emotional tendencies. This has sparked growing interest in Meme Emotion Understanding (MEU), which aims to classify the emotional intent behind memes by leveraging their multimodal contents. While existing efforts have achieved promising results, two major challenges remain: (1) a lack of fine-grained multimodal fusion strategies, and (2) insufficient mining of memes' implicit meanings and background knowledge. To address these challenges, we propose MemoDetector, a novel framework for advancing MEU. First, we introduce a four-step textual enhancement module that utilizes the rich knowledge and reasoning capabilities of Multimodal Large Language Models (MLLMs) to progressively infer and extract implicit and contextual insights from memes. These enhanced texts significantly enrich the original meme contents and provide valuable guidance for downstream classification. Next, we design a dual-stage modal fusion strategy: the first stage performs shallow fusion on raw meme image and text, while the second stage deeply integrates the enhanced visual and textual features. This hierarchical fusion enables the model to better capture nuanced cross-modal emotional cues. Experiments on two datasets, MET-MEME and MOOD, demonstrate that our method consistently outperforms state-of-the-art baselines. Specifically, MemoDetector improves F1 scores by 4.3\% on MET-MEME and 3.4\% on MOOD. Further ablation studies and in-depth analyses validate the effectiveness and robustness of our approach, highlighting its strong potential for advancing MEU. Our code is available at https://github.com/singing-cat/MemoDetector.

Peiyu Liu, Junming Liu, Lirong Fu et al.

Recently, ChatGPT has attracted great attention from the code analysis domain. Prior works show that ChatGPT has the capabilities of processing foundational code analysis tasks, such as abstract syntax tree generation, which indicates the potential of using ChatGPT to comprehend code syntax and static behaviors. However, it is unclear whether ChatGPT can complete more complicated real-world vulnerability management tasks, such as the prediction of security relevance and patch correctness, which require an all-encompassing understanding of various aspects, including code syntax, program semantics, and related manual comments. In this paper, we explore ChatGPT's capabilities on 6 tasks involving the complete vulnerability management process with a large-scale dataset containing 70,346 samples. For each task, we compare ChatGPT against SOTA approaches, investigate the impact of different prompts, and explore the difficulties. The results suggest promising potential in leveraging ChatGPT to assist vulnerability management. One notable example is ChatGPT's proficiency in tasks like generating titles for software bug reports. Furthermore, our findings reveal the difficulties encountered by ChatGPT and shed light on promising future directions. For instance, directly providing random demonstration examples in the prompt cannot consistently guarantee good performance in vulnerability management. By contrast, leveraging ChatGPT in a self-heuristic way -- extracting expertise from demonstration examples itself and integrating the extracted expertise in the prompt is a promising research direction. Besides, ChatGPT may misunderstand and misuse the information in the prompt. Consequently, effectively guiding ChatGPT to focus on helpful information rather than the irrelevant content is still an open problem.

Yuxuan Duan, Xuhong Zhang, Chuer Yu et al.

Recently, face swapping has been developing rapidly and achieved a surprising reality, raising concerns about fake content. As a countermeasure, various detection approaches have been proposed and achieved promising performance. However, most existing detectors struggle to maintain performance on unseen face swapping methods and low-quality images. Apart from the generalization problem, current detection approaches have been shown vulnerable to evasion attacks crafted by detection-aware manipulators. Lack of robustness under adversary scenarios leaves threats for applying face swapping detection in real world. In this paper, we propose a novel face swapping detection approach based on face identification probability distributions, coined as IdP_FSD, to improve the generalization and robustness. IdP_FSD is specially designed for detecting swapped faces whose identities belong to a finite set, which is meaningful in real-world applications. Compared with previous general detection methods, we make use of the available real faces with concerned identities and require no fake samples for training. IdP_FSD exploits face swapping's common nature that the identity of swapped face combines that of two faces involved in swapping. We reflect this nature with the confusion of a face identification model and measure the confusion with the maximum value of the output probability distribution. What's more, to defend our detector under adversary scenarios, an attention-based finetuning scheme is proposed for the face identification models used in IdP_FSD. Extensive experiments show that the proposed IdP_FSD not only achieves high detection performance on different benchmark datasets and image qualities but also raises the bar for manipulators to evade the detection.

Wenlong Meng, Zhenyuan Guo, Lenan Wu et al.

Large Language Models (LLMs) pose significant privacy risks, potentially leaking training data due to implicit memorization. Existing privacy attacks primarily focus on membership inference attacks (MIAs) or data extraction attacks, but reconstructing specific personally identifiable information (PII) in LLMs' training data remains challenging. In this paper, we propose R.R. (Recollect and Rank), a novel two-step privacy stealing attack that enables attackers to reconstruct PII entities from scrubbed training data where the PII entities have been masked. In the first stage, we introduce a prompt paradigm named recollection, which instructs the LLM to repeat a masked text but fill in masks. Then we can use PII identifiers to extract recollected PII candidates. In the second stage, we design a new criterion to score each PII candidate and rank them. Motivated by membership inference, we leverage the reference model as a calibration to our criterion. Experiments across three popular PII datasets demonstrate that the R.R. achieves better PII identification performance than baselines. These results highlight the vulnerability of LLMs to PII leakage even when training data has been scrubbed. We release our code and datasets at GitHub.

Zhenyuan Guo, Yi Shi, Wenlong Meng et al.

Model merging is a widespread technology in large language models (LLMs) that integrates multiple task-specific LLMs into a unified one, enabling the merged model to inherit the specialized capabilities of these LLMs. Most task-specific LLMs are sourced from open-source communities and have not undergone rigorous auditing, potentially imposing risks in model merging. This paper highlights an overlooked privacy risk: \textit{an unsafe model could compromise the privacy of other LLMs involved in the model merging.} Specifically, we propose PhiMM, a privacy attack approach that trains a phishing model capable of stealing privacy using a crafted privacy phishing instruction dataset. Furthermore, we introduce a novel model cloaking method that mimics a specialized capability to conceal attack intent, luring users into merging the phishing model. Once victims merge the phishing model, the attacker can extract personally identifiable information (PII) or infer membership information (MI) by querying the merged model with the phishing instruction. Experimental results show that merging a phishing model increases the risk of privacy breaches. Compared to the results before merging, PII leakage increased by 3.9\% and MI leakage increased by 17.4\% on average. We release the code of PhiMM through a link.

Wenlong Meng, Shuguo Fan, Chengkun Wei et al.

In this paper, we introduce GradEscape, the first gradient-based evader designed to attack AI-generated text (AIGT) detectors. GradEscape overcomes the undifferentiable computation problem, caused by the discrete nature of text, by introducing a novel approach to construct weighted embeddings for the detector input. It then updates the evader model parameters using feedback from victim detectors, achieving high attack success with minimal text modification. To address the issue of tokenizer mismatch between the evader and the detector, we introduce a warm-started evader method, enabling GradEscape to adapt to detectors across any language model architecture. Moreover, we employ novel tokenizer inference and model extraction techniques, facilitating effective evasion even in query-only access. We evaluate GradEscape on four datasets and three widely-used language models, benchmarking it against four state-of-the-art AIGT evaders. Experimental results demonstrate that GradEscape outperforms existing evaders in various scenarios, including with an 11B paraphrase model, while utilizing only 139M parameters. We have successfully applied GradEscape to two real-world commercial AIGT detectors. Our analysis reveals that the primary vulnerability stems from disparity in text expression styles within the training data. We also propose a potential defense strategy to mitigate the threat of AIGT evaders. We open-source our GradEscape for developing more robust AIGT detectors.

Ying Wang, Zhen Jin, Jiexiong Xu et al.

As augmented large language models (LLMs) with external tools become increasingly popular in web applications, improving augmented LLM inference serving efficiency and optimizing service-level objectives (SLOs) are critical for enhancing user experience. To achieve this, inference systems must maximize request handling within latency constraints, referred to as increasing effective throughput. However, existing systems face two major challenges: (i) reliance on first-come-first-served (FCFS) scheduling causes severe head-of-line blocking, leading to queuing delays exceeding the SLOs for many requests; and (ii) static batch token limit, which fails to adapt to fluctuating loads and hardware conditions. Both of these factors degrade effective throughput and service quality. This paper presents AugServe, an efficient inference framework designed to reduce queueing latency and enhance effective throughput for augmented LLM inference services. The core idea of AugServe is a two-stage adaptive request scheduling strategy. Specifically, AugServe combines the inference features of augmented LLM requests to optimize the order of scheduling decisions (stage I). These decisions are continuously refined with runtime information (stage II), adapting to both request characteristics and system capabilities. In addition, AugServe dynamically adjusts the token batching mechanism based on hardware status and real-time load, further enhancing throughput performance. Experimental results show that AugServe achieves 4.7-33.1x and 3.3-13.2x higher effective throughput than vLLM and InferCept, while reducing time-to-first-token (TTFT) by up to 96.3% and 95.0%, respectively.

Wenlong Meng, Fan Zhang, Wendao Yao et al.

Large language models (LLMs) have demonstrated significant utility in a wide range of applications; however, their deployment is plagued by security vulnerabilities, notably jailbreak attacks. These attacks manipulate LLMs to generate harmful or unethical content by crafting adversarial prompts. While much of the current research on jailbreak attacks has focused on single-turn interactions, it has largely overlooked the impact of historical dialogues on model behavior. In this paper, we introduce a novel jailbreak paradigm, Dialogue Injection Attack (DIA), which leverages the dialogue history to enhance the success rates of such attacks. DIA operates in a black-box setting, requiring only access to the chat API or knowledge of the LLM's chat template. We propose two methods for constructing adversarial historical dialogues: one adapts gray-box prefilling attacks, and the other exploits deferred responses. Our experiments show that DIA achieves state-of-the-art attack success rates on recent LLMs, including Llama-3.1 and GPT-4o. Additionally, we demonstrate that DIA can bypass 5 different defense mechanisms, highlighting its robustness and effectiveness.

Chengkun Wei, Weixian Li, Chen Gong et al.

Differentially Private Stochastic Gradient Descent (DP-SGD) is a widely adopted technique for privacy-preserving deep learning. A critical challenge in DP-SGD is selecting the optimal clipping threshold C, which involves balancing the trade-off between clipping bias and noise magnitude, incurring substantial privacy and computing overhead during hyperparameter tuning. In this paper, we propose Dynamic Clipping DP-SGD (DC-SGD), a framework that leverages differentially private histograms to estimate gradient norm distributions and dynamically adjust the clipping threshold C. Our framework includes two novel mechanisms: DC-SGD-P and DC-SGD-E. DC-SGD-P adjusts the clipping threshold based on a percentile of gradient norms, while DC-SGD-E minimizes the expected squared error of gradients to optimize C. These dynamic adjustments significantly reduce the burden of hyperparameter tuning C. The extensive experiments on various deep learning tasks, including image classification and natural language processing, show that our proposed dynamic algorithms achieve up to 9 times acceleration on hyperparameter tuning than DP-SGD. And DC-SGD-E can achieve an accuracy improvement of 10.62% on CIFAR10 than DP-SGD under the same privacy budget of hyperparameter tuning. We conduct rigorous theoretical privacy and convergence analyses, showing that our methods seamlessly integrate with the Adam optimizer. Our results highlight the robust performance and efficiency of DC-SGD, offering a practical solution for differentially private deep learning with reduced computational overhead and enhanced privacy guarantees.

Yuansheng Gao, Han Bao, Tong Zhang et al.

Mental manipulation is a subtle yet pervasive form of psychological abuse that poses serious threats to mental health. Nevertheless, detecting mental manipulation remains a largely underexplored research problem. The field faces three major challenges: (i) insufficient and hard-to-obtain training data; (ii) the covert nature of mental manipulation, which hinders detection; and (iii) the lack of real-world datasets. To address these challenges, we propose MentalMAC, a novel framework that enhances large language models' ability to detect elements of mental manipulation in multi-turn dialogue. Our approach consists of three key components: EvoSA, an annotation-free data augmentation method based on evolutionary operations and speech act theory; teacher-model-generated multi-task supervision; and progressive task-level anti-curriculum distillation. We then constructed the ReaMent dataset, comprising 5,000 real-world dialogue samples, utilizing MentalMAC-distilled models to aid in human annotation. Vast experiments show that MentalMAC achieves up to 25.9% improvement in F1mac and 8.1% in accuracy over the best-performing baseline, outperforming commercial LLMs such as GPT-4 and Claude-3.5-Sonnet. Warning: This paper contains content that may be offensive to the reader.

Zhe Ma, Qingming Li, Xuhong Zhang et al.

The past few years have witnessed substantial advances in image generation powered by diffusion models. However, it was shown that diffusion models are susceptible to training data memorization, raising significant concerns regarding copyright infringement and privacy invasion. This study delves into a rigorous analysis of memorization in diffusion models. We introduce InvMM, an inversion-based measure of memorization, which is based on inverting a sensitive latent noise distribution accounting for the replication of an image. For accurate estimation of the measure, we propose an adaptive algorithm that balances the normality and sensitivity of the noise distribution. Comprehensive experiments across four datasets, conducted on both unconditional and text-guided diffusion models, demonstrate that InvMM provides a reliable and complete quantification of memorization. Notably, InvMM is commensurable between samples, reveals the true extent of memorization from an adversarial standpoint and implies how memorization differs from membership. In practice, it serves as an auditing tool for developers to reliably assess the risk of memorization, thereby contributing to the enhancement of trustworthiness and privacy-preserving capabilities of diffusion models.

Chengkun Wei, Minghu Zhao, Zhikun Zhang et al.

Differential privacy (DP), as a rigorous mathematical definition quantifying privacy leakage, has become a well-accepted standard for privacy protection. Combined with powerful machine learning techniques, differentially private machine learning (DPML) is increasingly important. As the most classic DPML algorithm, DP-SGD incurs a significant loss of utility, which hinders DPML's deployment in practice. Many studies have recently proposed improved algorithms based on DP-SGD to mitigate utility loss. However, these studies are isolated and cannot comprehensively measure the performance of improvements proposed in algorithms. More importantly, there is a lack of comprehensive research to compare improvements in these DPML algorithms across utility, defensive capabilities, and generalizability. We fill this gap by performing a holistic measurement of improved DPML algorithms on utility and defense capability against membership inference attacks (MIAs) on image classification tasks. We first present a taxonomy of where improvements are located in the machine learning life cycle. Based on our taxonomy, we jointly perform an extensive measurement study of the improved DPML algorithms. We also cover state-of-the-art label differential privacy (Label DP) algorithms in the evaluation. According to our empirical results, DP can effectively defend against MIAs, and sensitivity-bounding techniques such as per-sample gradient clipping play an important role in defense. We also explore some improvements that can maintain model utility and defend against MIAs more effectively. Experiments show that Label DP algorithms achieve less utility loss but are fragile to MIAs. To support our evaluation, we implement a modular re-usable software, DPMLBench, which enables sensitive data owners to deploy DPML algorithms and serves as a benchmark tool for researchers and practitioners.

Jiaqi Yang, Xuequan Lu, Wenzhi Chen

Most existing 3D geometry copy detection research focused on 3D watermarking, which first embeds ``watermarks'' and then detects the added watermarks. However, this kind of methods is non-straightforward and may be less robust to attacks such as cropping and noise. In this paper, we focus on a fundamental and practical research problem: judging whether a point cloud is plagiarized or copied to another point cloud in the presence of several manipulations (e.g., similarity transformation, smoothing). We propose a novel method to address this critical problem. Our key idea is first to align the two point clouds and then calculate their similarity distance. We design three different measures to compute the similarity. We also introduce two strategies to speed up our method. Comprehensive experiments and comparisons demonstrate the effectiveness and robustness of our method in estimating the similarity of two given 3D point clouds.