arXiv subjects

Zhen Sun

Publications and source records attributed to Zhen Sun.

At least 55 records · Page 3Linked to original sources

arXiv2025

FC-Attack: Jailbreaking Multimodal Large Language Models via Auto-Generated Flowcharts

Multimodal Large Language Models (MLLMs) have become powerful and widely adopted in some practical applications. However, recent research has revealed their vulnerability to multimodal jailbreak attacks, whereby the model can be induced to generate harmful content, leading to safety risks. Although most MLLMs have undergone safety alignment, recent research shows that the visual modality is still vulnerable to jailbreak attacks. In our work, we discover that by using flowcharts with partially harmful information, MLLMs can be induced to provide additional harmful details. Based on this, we propose a jailbreak attack method based on auto-generated flowcharts, FC-Attack. Specifically, FC-Attack first fine-tunes a pre-trained LLM to create a step-description generator based on benign datasets. The generator is then used to produce step descriptions corresponding to a harmful query, which are transformed into flowcharts in 3 different shapes (vertical, horizontal, and S-shaped) as visual prompts. These flowcharts are then combined with a benign textual prompt to execute the jailbreak attack on MLLMs. Our evaluations on Advbench show that FC-Attack attains an attack success rate of up to 96% via images and up to 78% via videos across multiple MLLMs. Additionally, we investigate factors affecting the attack performance, including the number of steps and the font styles in the flowcharts. We also find that FC-Attack can improve the jailbreak performance from 4% to 28% in Claude-3.5 by changing the font style. To mitigate the attack, we explore several defenses and find that AdaShield can largely reduce the jailbreak performance but with the cost of utility drop.

cs.CV↗

arXiv2025

ZPD-SCA: Unveiling the Blind Spots of LLMs in Assessing Students' Cognitive Abilities

Large language models (LLMs) have demonstrated potential in educational applications, yet their capacity to accurately assess the cognitive alignment of reading materials with students' developmental stages remains insufficiently explored. This gap is particularly critical given the foundational educational principle of the Zone of Proximal Development (ZPD), which emphasizes the need to match learning resources with Students' Cognitive Abilities (SCA). Despite the importance of this alignment, there is a notable absence of comprehensive studies investigating LLMs' ability to evaluate reading comprehension difficulty across different student age groups, especially in the context of Chinese language education. To fill this gap, we introduce ZPD-SCA, a novel benchmark specifically designed to assess stage-level Chinese reading comprehension difficulty. The benchmark is annotated by 60 Special Grade teachers, a group that represents the top 0.15% of all in-service teachers nationwide. Experimental results reveal that LLMs perform poorly in zero-shot learning scenarios, with Qwen-max and GLM even falling below the probability of random guessing. When provided with in-context examples, LLMs performance improves substantially, with some models achieving nearly double the accuracy of their zero-shot baselines. These results reveal that LLMs possess emerging abilities to assess reading difficulty, while also exposing limitations in their current training for educationally aligned judgment. Notably, even the best-performing models display systematic directional biases, suggesting difficulties in accurately aligning material difficulty with SCA. Furthermore, significant variations in model performance across different genres underscore the complexity of task. We envision that ZPD-SCA can provide a foundation for evaluating and improving LLMs in cognitively aligned educational applications.

cs.CL↗

arXiv2025

Thought Manipulation: External Thought Can Be Efficient for Large Reasoning Models

Recent advancements in large reasoning models (LRMs) have demonstrated the effectiveness of scaling test-time computation to enhance reasoning capabilities on various tasks. However, LRMs often suffer from an ``overthinking'' problem, where the model generates excessively redundant reasoning steps with limited performance gains. In this work, we empirically reveal an important characteristic of LRM behaviors that placing external CoTs generated by smaller models between the thinking token (\texttt{ } and \texttt{ }) can effectively manipulate the model to generate fewer thoughts. Building on this finding, we propose a simple yet efficient pipeline, \Method, to enable LRMs to bypass unnecessary intermediate steps, thereby significantly reducing computational costs. We conduct extensive experiments to evaluate the utility and efficiency of \Method. For instance, when applied to QwQ-32B on the LiveBench/Code dataset, \Method keeps the original performance while reducing output token counts by approximately 30\%, with minimal overhead introduced by the CoT generator. Furthermore, we identify two suboptimal modes, blindly following flawed external thoughts and unnecessary rethinking, and show that simple mitigations, such as difficulty-aware fallbacks, can further improve performance. Overall, \Method offers a practical, general, and efficient way to optimize LRM inference, making powerful reasoning models more accessible and scalable for real-world applications.

cs.CL↗

arXiv2025

Unsafe LLM-Based Search: Quantitative Analysis and Mitigation of Safety Risks in AI Web Search

Recent advancements in Large Language Models (LLMs) have significantly enhanced the capabilities of AI-Powered Search Engines (AIPSEs), offering precise and efficient responses by integrating external databases with pre-existing knowledge. However, we observe that these AIPSEs raise risks such as quoting malicious content or citing malicious websites, leading to harmful or unverified information dissemination. In this study, we conduct the first safety risk quantification on seven production AIPSEs by systematically defining the threat model, risk type, and evaluating responses to various query types. With data collected from PhishTank, ThreatBook, and LevelBlue, our findings reveal that AIPSEs frequently generate harmful content that contains malicious URLs even with benign queries (e.g., with benign keywords). We also observe that directly querying a URL will increase the number of main risk-inclusive responses, while querying with natural language will slightly mitigate such risk. Compared to traditional search engines, AIPSEs outperform in both utility and safety. We further perform two case studies on online document spoofing and phishing to show the ease of deceiving AIPSEs in the real-world setting. To mitigate these risks, we develop an agent-based defense with a GPT-4.1-based content refinement tool and a URL detector. Our evaluation shows that our defense can effectively reduce the risk, with only a minor cost of reducing available information by approximately 10.7%. Our research highlights the urgent need for robust safety measures in AIPSEs.

cs.CR↗

arXiv2025

PEFTGuard: Detecting Backdoor Attacks Against Parameter-Efficient Fine-Tuning

Fine-tuning is an essential process to improve the performance of Large Language Models (LLMs) in specific domains, with Parameter-Efficient Fine-Tuning (PEFT) gaining popularity due to its capacity to reduce computational demands through the integration of low-rank adapters. These lightweight adapters, such as LoRA, can be shared and utilized on open-source platforms. However, adversaries could exploit this mechanism to inject backdoors into these adapters, resulting in malicious behaviors like incorrect or harmful outputs, which pose serious security risks to the community. Unfortunately, few current efforts concentrate on analyzing the backdoor patterns or detecting the backdoors in the adapters. To fill this gap, we first construct and release PADBench, a comprehensive benchmark that contains 13,300 benign and backdoored adapters fine-tuned with various datasets, attack strategies, PEFT methods, and LLMs. Moreover, we propose PEFTGuard, the first backdoor detection framework against PEFT-based adapters. Extensive evaluation upon PADBench shows that PEFTGuard outperforms existing detection methods, achieving nearly perfect detection accuracy (100%) in most cases. Notably, PEFTGuard exhibits zero-shot transferability on three aspects, including different attacks, PEFT methods, and adapter ranks. In addition, we consider various adaptive attacks to demonstrate the high robustness of PEFTGuard. We further explore several possible backdoor mitigation defenses, finding fine-mixing to be the most effective method. We envision that our benchmark and method can shed light on future LLM backdoor detection research.

cs.CR↗

arXiv2025

Backdoor Attack on Vision Language Models with Stealthy Semantic Manipulation

Vision Language Models (VLMs) have shown remarkable performance, but are also vulnerable to backdoor attacks whereby the adversary can manipulate the model's outputs through hidden triggers. Prior attacks primarily rely on single-modality triggers, leaving the crucial cross-modal fusion nature of VLMs largely unexplored. Unlike prior work, we identify a novel attack surface that leverages cross-modal semantic mismatches as implicit triggers. Based on this insight, we propose BadSem (Backdoor Attack with Semantic Manipulation), a data poisoning attack that injects stealthy backdoors by deliberately misaligning image-text pairs during training. To perform the attack, we construct SIMBad, a dataset tailored for semantic manipulation involving color and object attributes. Extensive experiments across four widely used VLMs show that BadSem achieves over 98% average ASR, generalizes well to out-of-distribution datasets, and can transfer across poisoning modalities. Our detailed analysis using attention visualization shows that backdoored models focus on semantically sensitive regions under mismatched conditions while maintaining normal behavior on clean inputs. To mitigate the attack, we try two defense strategies based on system prompt and supervised fine-tuning but find that both of them fail to mitigate the semantic backdoor. Our findings highlight the urgent need to address semantic vulnerabilities in VLMs for their safer deployment.

cs.CV↗

arXiv2025

Are We in the AI-Generated Text World Already? Quantifying and Monitoring AIGT on Social Media

Social media platforms are experiencing a growing presence of AI-Generated Texts (AIGTs). However, the misuse of AIGTs could have profound implications for public opinion, such as spreading misinformation and manipulating narratives. Despite its importance, it remains unclear how prevalent AIGTs are on social media. To address this gap, this paper aims to quantify and monitor the AIGTs on online social media platforms. We first collect a dataset (SM-D) with around 2.4M posts from 3 major social media platforms: Medium, Quora, and Reddit. Then, we construct a diverse dataset (AIGTBench) to train and evaluate AIGT detectors. AIGTBench combines popular open-source datasets and our AIGT datasets generated from social media texts by 12 LLMs, serving as a benchmark for evaluating mainstream detectors. With this setup, we identify the best-performing detector (OSM-Det). We then apply OSM-Det to SM-D to track AIGTs across social media platforms from January 2022 to October 2024, using the AI Attribution Rate (AAR) as the metric. Specifically, Medium and Quora exhibit marked increases in AAR, rising from 1.77% to 37.03% and 2.06% to 38.95%, respectively. In contrast, Reddit shows slower growth, with AAR increasing from 1.31% to 2.45% over the same period. Our further analysis indicates that AIGTs on social media differ from human-written texts across several dimensions, including linguistic patterns, topic distributions, engagement levels, and the follower distribution of authors. We envision our analysis and findings on AIGTs in social media can shed light on future research in this domain.

cs.AI↗

arXiv2025

Evaluation Hallucination in Multi-Round Incomplete Information Lateral-Driven Reasoning Tasks

Multi-round incomplete information tasks are crucial for evaluating the lateral thinking capabilities of large language models (LLMs). Currently, research primarily relies on multiple benchmarks and automated evaluation metrics to assess these abilities. However, our study reveals novel insights into the limitations of existing methods, as they often yield misleading results that fail to uncover key issues, such as shortcut-taking behaviors, rigid patterns, and premature task termination. These issues obscure the true reasoning capabilities of LLMs and undermine the reliability of evaluations. To address these limitations, we propose a refined set of evaluation standards, including inspection of reasoning paths, diversified assessment metrics, and comparative analyses with human performance.

cs.CL↗

arXiv2025

SynDec: A Synthesize-then-Decode Approach for Arbitrary Textual Style Transfer via Large Language Models

Large Language Models (LLMs) are emerging as dominant forces for textual style transfer. However, for arbitrary style transfer, LLMs face two key challenges: (1) considerable reliance on manually-constructed prompts and (2) rigid stylistic biases inherent in LLMs. In this paper, we propose a novel Synthesize-then-Decode (SynDec) approach, which automatically synthesizes high-quality prompts and amplifies their roles during decoding process. Specifically, our approach synthesizes prompts by selecting representative few-shot samples, conducting a four-dimensional style analysis, and reranking the candidates. At LLM decoding stage, the TST effect is amplified by maximizing the contrast in output probabilities between scenarios with and without the synthesized prompt, as well as between prompts and negative samples. We conduct extensive experiments and the results show that SynDec outperforms existing state-of-the-art LLM-based methods on five out of six benchmarks (e.g., achieving up to a 9\% increase in accuracy for modern-to-Elizabethan English transfer). Detailed ablation studies further validate the effectiveness of SynDec.

cs.CL↗

arXiv2025

Humanizing LLMs: A Survey of Psychological Measurements with Tools, Datasets, and Human-Agent Applications

As large language models (LLMs) are increasingly used in human-centered tasks, assessing their psychological traits is crucial for understanding their social impact and ensuring trustworthy AI alignment. While existing reviews have covered some aspects of related research, several important areas have not been systematically discussed, including detailed discussions of diverse psychological tests, LLM-specific psychological datasets, and the applications of LLMs with psychological traits. To address this gap, we systematically review six key dimensions of applying psychological theories to LLMs: (1) assessment tools; (2) LLM-specific datasets; (3) evaluation metrics (consistency and stability); (4) empirical findings; (5) personality simulation methods; and (6) LLM-based behavior simulation. Our analysis highlights both the strengths and limitations of current methods. While some LLMs exhibit reproducible personality patterns under specific prompting schemes, significant variability remains across tasks and settings. Recognizing methodological challenges such as mismatches between psychological tools and LLMs' capabilities, as well as inconsistencies in evaluation practices, this study aims to propose future directions for developing more interpretable, robust, and generalizable psychological assessment frameworks for LLMs.

cs.CY↗

arXiv2025

ATHENA: An In-vehicle CAN Intrusion Detection Framework Based on Physical Characteristics of Vehicle Systems

With the growing interconnection between In-Vehicle Networks (IVNs) and external environments, intelligent vehicles are increasingly vulnerable to sophisticated external network attacks. This paper proposes ATHENA, the first IVN intrusion detection framework that adopts a vehicle-cloud integrated architecture to achieve better security performance for the resource-constrained vehicular environment. Specifically, in the cloud with sufficient resources, ATHENA uses the clustering method of multi-distribution mixture model combined with deep data mining technology to generate the raw Payload Rule Bank of IVN CAN messages, and then improves the rule quality with the help of exploitation on the first-principled physical knowledge of the vehicle system, after which the payload rules are periodically sent to the vehicle terminal. At the vehicle terminal, a simple LSTM component is used to generate the Time Rule Bank representing the long-term time series dependencies and the periodic characteristics of CAN messages, but not for any detection tasks as in traditional usage scenarios, where only the generated time rules are the candidates for further IVN intrusion detection tasks. Based on both the payload and time rules generated from cloud and vehicle terminal, ATHENA can achieve efficient intrusion detection capability by simple rule-base matching operations, rather than using complex black-box reasoning of resource-intensive neural network models, which is in fact only used for rule logic generation phase instead of the actual intrusion detection phase in our framework. Comparative experimental results on the ROAD dataset, which is current the most outstanding real-world in-vehicle CAN dataset covering new instances of sophisticated and stealthy masquerade attacks, demonstrate ATHENA significantly outperforms the state-of-the-art IVN intrusion detection methods in detecting complex attacks.

cs.CR↗

arXiv2025

TH-Bench: Evaluating Evading Attacks via Humanizing AI Text on Machine-Generated Text Detectors

As Large Language Models (LLMs) advance, Machine-Generated Texts (MGTs) have become increasingly fluent, high-quality, and informative. Existing wide-range MGT detectors are designed to identify MGTs to prevent the spread of plagiarism and misinformation. However, adversaries attempt to humanize MGTs to evade detection (named evading attacks), which requires only minor modifications to bypass MGT detectors. Unfortunately, existing attacks generally lack a unified and comprehensive evaluation framework, as they are assessed using different experimental settings, model architectures, and datasets. To fill this gap, we introduce the Text-Humanization Benchmark (TH-Bench), the first comprehensive benchmark to evaluate evading attacks against MGT detectors. TH-Bench evaluates attacks across three key dimensions: evading effectiveness, text quality, and computational overhead. Our extensive experiments evaluate 6 state-of-the-art attacks against 13 MGT detectors across 6 datasets, spanning 19 domains and generated by 11 widely used LLMs. Our findings reveal that no single evading attack excels across all three dimensions. Through in-depth analysis, we highlight the strengths and limitations of different attacks. More importantly, we identify a trade-off among three dimensions and propose two optimization insights. Through preliminary experiments, we validate their correctness and effectiveness, offering potential directions for future research.

cs.CR↗

arXiv2025

A2I-Calib: An Anti-noise Active Multi-IMU Spatial-temporal Calibration Framework for Legged Robots

Recently, multi-node inertial measurement unit (IMU)-based odometry for legged robots has gained attention due to its cost-effectiveness, power efficiency, and high accuracy. However, the spatial and temporal misalignment between foot-end motion derived from forward kinematics and foot IMU measurements can introduce inconsistent constraints, resulting in odometry drift. Therefore, accurate spatial-temporal calibration is crucial for the multi-IMU systems. Although existing multi-IMU calibration methods have addressed passive single-rigid-body sensor calibration, they are inadequate for legged systems. This is due to the insufficient excitation from traditional gaits for calibration, and enlarged sensitivity to IMU noise during kinematic chain transformations. To address these challenges, we propose A$^2$I-Calib, an anti-noise active multi-IMU calibration framework enabling autonomous spatial-temporal calibration for arbitrary foot-mounted IMUs. Our A$^2$I-Calib includes: 1) an anti-noise trajectory generator leveraging a proposed basis function selection theorem to minimize the condition number in correlation analysis, thus reducing noise sensitivity, and 2) a reinforcement learning (RL)-based controller that ensures robust execution of calibration motions. Furthermore, A$^2$I-Calib is validated on simulation and real-world quadruped robot platforms with various multi-IMU settings, which demonstrates a significant reduction in noise sensitivity and calibration errors, thereby improving the overall multi-IMU odometry performance.

cs.RO↗

arXiv2025

THE-SEAN: A Heart Rate Variation-Inspired Temporally High-Order Event-Based Visual Odometry with Self-Supervised Spiking Event Accumulation Networks

Event-based visual odometry has recently gained attention for its high accuracy and real-time performance in fast-motion systems. Unlike traditional synchronous estimators that rely on constant-frequency (zero-order) triggers, event-based visual odometry can actively accumulate information to generate temporally high-order estimation triggers. However, existing methods primarily focus on adaptive event representation after estimation triggers, neglecting the decision-making process for efficient temporal triggering itself. This oversight leads to the computational redundancy and noise accumulation. In this paper, we introduce a temporally high-order event-based visual odometry with spiking event accumulation networks (THE-SEAN). To the best of our knowledge, it is the first event-based visual odometry capable of dynamically adjusting its estimation trigger decision in response to motion and environmental changes. Inspired by biological systems that regulate hormone secretion to modulate heart rate, a self-supervised spiking neural network is designed to generate estimation triggers. This spiking network extracts temporal features to produce triggers, with rewards based on block matching points and Fisher information matrix (FIM) trace acquired from the estimator itself. Finally, THE-SEAN is evaluated across several open datasets, thereby demonstrating average improvements of 13\% in estimation accuracy, 9\% in smoothness, and 38\% in triggering efficiency compared to the state-of-the-art methods.

cs.RO↗

arXiv2025

On the Generalization and Adaptation Ability of Machine-Generated Text Detectors in Academic Writing

The rising popularity of large language models (LLMs) has raised concerns about machine-generated text (MGT), particularly in academic settings, where issues like plagiarism and misinformation are prevalent. As a result, developing a highly generalizable and adaptable MGT detection system has become an urgent priority. Given that LLMs are most commonly misused in academic writing, this work investigates the generalization and adaptation capabilities of MGT detectors in three key aspects specific to academic writing: First, we construct MGT-Acedemic, a large-scale dataset comprising over 336M tokens and 749K samples. MGT-Acedemic focuses on academic writing, featuring human-written texts (HWTs) and MGTs across STEM, Humanities, and Social Sciences, paired with an extensible code framework for efficient benchmarking. Second, we benchmark the performance of various detectors for binary classification and attribution tasks in both in-domain and cross-domain settings. This benchmark reveals the often-overlooked challenges of attribution tasks. Third, we introduce a novel attribution task where models have to adapt to new classes over time without (or with very limited) access to prior training data in both few-shot and many-shot scenarios. We implement eight different adapting techniques to improve the performance and highlight the inherent complexity of the task. Our findings provide insights into the generalization and adaptation ability of MGT detectors across diverse scenarios and lay the foundation for building robust, adaptive detection systems. The code framework is available at https://github.com/Y-L-LIU/MGTBench-2.0.

cs.AI↗

arXiv2024

Quantized Delta Weight Is Safety Keeper

Recent advancements in fine-tuning proprietary language models enable customized applications across various domains but also introduce two major challenges: high resource demands and security risks. Regarding resource demands, recent work proposes novel partial compression, such as BitDelta, to quantize the delta weights between the fine-tuned model and base model. Regarding the security risks, user-defined fine-tuning can introduce security vulnerabilities, such as alignment issues, backdoor attacks, and hallucinations. However, most of the current efforts in security assessment focus on the full-precision or full-compression models, it is not well-discussed how the partial compression methods affect security concerns. To bridge this gap, we evaluate the robustness of delta-weight quantization against these security threats. In this paper, we uncover a "free lunch" phenomenon: partial compression can enhance model security against fine-tuning-based attacks with bearable utility loss. Using Llama-2-7b-chat as a case study, we show that, with under 10% utility degradation, the partial compression mitigates alignment-breaking risks by up to 66.17%, harmful backdoor vulnerabilities by 64.46%, and targeted output manipulation risks by up to 90.53%. We further apply LogitLens to visualize internal state transformations during forward passes, suggesting mechanisms for both security failure and recovery in standard versus compressed fine-tuning. This work offers new insights into selecting effective delta compression methods for secure, resource-efficient multi-tenant services.

cs.CR↗

arXiv2024

IMOST: Incremental Memory Mechanism with Online Self-Supervision for Continual Traversability Learning

Traversability estimation is the foundation of path planning for a general navigation system. However, complex and dynamic environments pose challenges for the latest methods using self-supervised learning (SSL) technique. Firstly, existing SSL-based methods generate sparse annotations lacking detailed boundary information. Secondly, their strategies focus on hard samples for rapid adaptation, leading to forgetting and biased predictions. In this work, we propose IMOST, a continual traversability learning framework composed of two key modules: incremental dynamic memory (IDM) and self-supervised annotation (SSA). By mimicking human memory mechanisms, IDM allocates novel data samples to new clusters according to information expansion criterion. It also updates clusters based on diversity rule, ensuring a representative characterization of new scene. This mechanism enhances scene-aware knowledge diversity while maintaining a compact memory capacity. The SSA module, integrating FastSAM, utilizes point prompts to generate complete annotations in real time which reduces training complexity. Furthermore, IMOST has been successfully deployed on the quadruped robot, with performance evaluated during the online learning process. Experimental results on both public and self-collected datasets demonstrate that our IMOST outperforms current state-of-the-art method, maintains robust recognition capabilities and adaptability across various scenarios. The code is available at https://github.com/SJTU-MKH/OCLTrav.

cs.RO↗

arXiv2024

Learning-based Traversability Costmap for Autonomous Off-road Navigation

Traversability estimation in off-road terrains is an essential procedure for autonomous navigation. However, creating reliable labels for complex interactions between the robot and the surface is still a challenging problem in learning-based costmap generation. To address this, we propose a method that predicts traversability costmaps by leveraging both visual and geometric information of the environment. To quantify the surface properties like roughness and bumpiness, we introduce a novel way of risk-aware labelling with proprioceptive information for network training. We validate our method in costmap prediction and navigation tasks for complex off-road scenarios. Our results demonstrate that our costmap prediction method excels in terms of average accuracy and MSE. The navigation results indicate that using our learned costmaps leads to safer and smoother driving, outperforming previous methods in terms of the highest success rate, lowest normalized trajectory length, lowest time cost, and highest mean stability across two scenarios.

cs.RO↗