arXiv subjects

Ting Liu

Publications and source records attributed to Ting Liu.

At least 109 records · Page 6Linked to original sources

arXiv2025

Anchoring Refusal Direction: Mitigating Safety Risks in Tuning via Projection Constraint

Instruction Fine-Tuning (IFT) has been widely adopted as an effective post-training strategy to enhance various abilities of Large Language Models (LLMs). However, prior studies have shown that IFT can significantly compromise LLMs' safety, particularly their ability to refuse malicious instructions, raising significant concerns. Recent research into the internal mechanisms of LLMs has identified the refusal direction (r-direction) in the hidden states, which plays a pivotal role in governing refusal behavior. Building on this insight, our study reveals that the r-direction tends to drift during training, which we identify as one of the causes of the associated safety risks. To mitigate such drift, our proposed ProCon method introduces a projection-constrained loss term that regularizes the projection magnitude of each training sample's hidden state onto the r-direction. Our initial analysis shows that applying an appropriate constraint can effectively mitigate the refusal direction drift and associated safety risks, but remains limited by overall performance barriers. To overcome this barrier, informed by our observation of early-stage sharp drift and a data-driven perspective, we introduce a warm-up strategy that emphasizes early-stage strong constraints and broaden the data distribution to strengthen constraint signals, leading to an enhanced ProCon method. Experimental results under various datasets, scenarios, and LLMs demonstrate that our method can significantly mitigate safety risks posed by IFT while preserving task performance gains. Even compared with strong baselines, our method consistently delivers superior overall performance. Crucially, our analysis indicates that ProCon can contribute to stabilizing the r-direction during training, while such an interpretability-driven exploration of LLMs' internal mechanisms lays a solid foundation for future safety research.

cs.CL↗

arXiv2025

MoGU V2: Toward a Higher Pareto Frontier Between Model Usability and Security

As Large Language Models (LLMs) increasingly permeate human life, their security has emerged as a critical concern, particularly their ability to maintain harmless responses to malicious instructions. Although extensive methods have improved LLMs' security, they often lead to conservative, rejection-oriented responses that compromise practical usability. This presents a key challenge: how to advance the Pareto frontier between LLMs' usability and security, rather than necessitate a trade-off between them. To address this, we propose the MoGU framework, in which the intra-layer router dynamically allocates weights by sensing hidden states, thereby balancing the contributions of security-optimized and usability-optimized variants. Despite its initial potential, the MoGU framework faces limitations such as parameter redundancy and performance bottlenecks. To overcome these, we further propose an improved MoGU_v2 framework that establishes a tighter coupling between the routers and hidden states. In MoGU_v2, routers are embedded only in layers encoding highly classifiable security features, and backbone modules are activated during router optimization to enable bidirectional adaptation. MoGU_V2 exhibits strong adaptability and stable improvements across various series of LLMs, including mainstream LLMs serving as brains in various applications, on-device LLMs optimized for resource-constrained scenarios, and reasoning LLMs tailored for user interpretability. Meanwhile, even facing risks introduced by Instruction Fine-tuning, MoGU_v2 can easily restore security without compromising the task performance gains via a simple data-mix strategy. These comprehensive improvements highlight MoGU_V2 as a robust and versatile solution for mitigating security risks in real-world applications.

cs.CL↗

arXiv2025

Direct measurement of the 103Rh(n,gamma) and 103Rh(gamma,n) cross section up to stellar temperatures at the CSNS Back-n and SSRF SLEGS

The cross sections of 103Rh(n,gamma) and 103Rh(gamma,n) play a crucial role in the stellar nucleosynthesis, rhodium-based self-powered neutron detectors, and nuclear medicine. The cross sections of 103Rh(n,gamma) was measured by the time-of-flight(TOF) method from 1 eV to 1000 keV at the Back-n facility of the Chinese Spallation Neutron Source. In the resolved resonance region, the data reported multiple new resonance structures for the first time. And some discrepancies were observed, offering valuable insights into the differences between the evaluated libraries. Maxwellian-averaged cross sections (MACSs) were calculated within the temperature range of the s process nucleosynthesis model, based on the averaged cross sections in the unresolved resonance region. Meanwhile the cross sections of 103Rh(gamma,n) within the range of p process nucleosynthesis were measured using laser Compton scattering (LCS) gamma rays and a new neutron flat efficiency detector (FED) array at the Shanghai Laser Electron Gamma Source (SLEGS), Shanghai Synchrotron Radiation Facility (SSRF). Using an unfolding iteration method, 103Rh(gamma,n) data were obtained with uncertainty less than 5%, and the inconsistencies between the available experimental data and the evaluated libraries were discussed. This study provides a reliable benchmark for nuclear data evaluation and model optimization, and lays a solid foundation for Rh medical isotope applications and astrophysical research.

nucl-ex↗

arXiv2025

GoAI: Enhancing AI Students' Learning Paths and Idea Generation via Graph of AI Ideas

With the rapid advancement of artificial intelligence technology, AI students are confronted with a significant "information-to-innovation" gap: they must navigate through the rapidly expanding body of literature, trace the development of a specific research field, and synthesize various techniques into feasible innovative concepts. An additional critical step for students is to identify the necessary prerequisite knowledge and learning paths. Although many approaches based on large language models (LLMs) can summarize the content of papers and trace the development of a field through citations, these methods often overlook the prerequisite knowledge involved in the papers and the rich semantic information embedded in the citation relationships between papers. Such information reveals how methods are interrelated, built upon, extended, or challenged. To address these limitations, we propose GoAI, a tool for constructing educational knowledge graphs from AI research papers that leverages these graphs to plan personalized learning paths and support creative ideation. The nodes in the knowledge graph we have built include papers and the prerequisite knowledge, such as concepts, skills, and tools, that they involve; the edges record the semantic information of citations. When a student queries a specific paper, a beam search-based path search method can trace the current development trends of the field from the queried paper and plan a learning path toward cutting-edge objectives. The integrated Idea Studio guides students to clarify problem statements, compare alternative designs, and provide formative feedback on novelty, clarity, feasibility, and alignment with learning objectives.

cs.AI↗

arXiv2025

MME-SCI: A Comprehensive and Challenging Science Benchmark for Multimodal Large Language Models

Recently, multimodal large language models (MLLMs) have achieved significant advancements across various domains, and corresponding evaluation benchmarks have been continuously refined and improved. In this process, benchmarks in the scientific domain have played an important role in assessing the reasoning capabilities of MLLMs. However, existing benchmarks still face three key challenges: 1) Insufficient evaluation of models' reasoning abilities in multilingual scenarios; 2) Inadequate assessment of MLLMs' comprehensive modality coverage; 3) Lack of fine-grained annotation of scientific knowledge points. To address these gaps, we propose MME-SCI, a comprehensive and challenging benchmark. We carefully collected 1,019 high-quality question-answer pairs, which involve 3 distinct evaluation modes. These pairs cover four subjects, namely mathematics, physics, chemistry, and biology, and support five languages: Chinese, English, French, Spanish, and Japanese. We conducted extensive experiments on 16 open-source models and 4 closed-source models, and the results demonstrate that MME-SCI is widely challenging for existing MLLMs. For instance, under the Image-only evaluation mode, o4-mini achieved accuracy of only 52.11%, 24.73%, 36.57%, and 29.80% in mathematics, physics, chemistry, and biology, respectively, indicating a significantly higher difficulty level compared to existing benchmarks. More importantly, using MME-SCI's multilingual and fine-grained knowledge attributes, we analyzed existing models' performance in depth and identified their weaknesses in specific domains. The Data and Evaluation Code are available at https://github.com/JCruan519/MME-SCI.

cs.CL↗

arXiv2025

Wisdom of the Crowd: Reinforcement Learning from Coevolutionary Collective Feedback

Reinforcement learning (RL) has significantly enhanced the reasoning capabilities of large language models (LLMs), but its reliance on expensive human-labeled data or complex reward models severely limits scalability. While existing self-feedback methods aim to address this problem, they are constrained by the capabilities of a single model, which can lead to overconfidence in incorrect answers, reward hacking, and even training collapse. To this end, we propose Reinforcement Learning from Coevolutionary Collective Feedback (RLCCF), a novel RL framework that enables multi-model collaborative evolution without external supervision. Specifically, RLCCF optimizes the ability of a model collective by maximizing its Collective Consistency (CC), which jointly trains a diverse ensemble of LLMs and provides reward signals by voting on collective outputs. Moreover, each model's vote is weighted by its Self-Consistency (SC) score, ensuring that more confident models contribute more to the collective decision. Benefiting from the diverse output distributions and complementary abilities of multiple LLMs, RLCCF enables the model collective to continuously enhance its reasoning ability through coevolution. Experiments on four mainstream open-source LLMs across four mathematical reasoning benchmarks demonstrate that our framework yields significant performance gains, achieving an average relative improvement of 16.72\% in accuracy. Notably, RLCCF not only improves the performance of individual models but also enhances the group's majority-voting accuracy by 4.51\%, demonstrating its ability to extend the collective capability boundary of the model collective.

cs.AI↗

arXiv2025

MedAtlas: Evaluating LLMs for Multi-Round, Multi-Task Medical Reasoning Across Diverse Imaging Modalities and Clinical Text

Artificial intelligence has demonstrated significant potential in clinical decision-making; however, developing models capable of adapting to diverse real-world scenarios and performing complex diagnostic reasoning remains a major challenge. Existing medical multi-modal benchmarks are typically limited to single-image, single-turn tasks, lacking multi-modal medical image integration and failing to capture the longitudinal and multi-modal interactive nature inherent to clinical practice. To address this gap, we introduce MedAtlas, a novel benchmark framework designed to evaluate large language models on realistic medical reasoning tasks. MedAtlas is characterized by four key features: multi-turn dialogue, multi-modal medical image interaction, multi-task integration, and high clinical fidelity. It supports four core tasks: open-ended multi-turn question answering, closed-ended multi-turn question answering, multi-image joint reasoning, and comprehensive disease diagnosis. Each case is derived from real diagnostic workflows and incorporates temporal interactions between textual medical histories and multiple imaging modalities, including CT, MRI, PET, ultrasound, and X-ray, requiring models to perform deep integrative reasoning across images and clinical texts. MedAtlas provides expert-annotated gold standards for all tasks. Furthermore, we propose two novel evaluation metrics: Round Chain Accuracy and Error Propagation Resistance. Benchmark results with existing multi-modal models reveal substantial performance gaps in multi-stage clinical reasoning. MedAtlas establishes a challenging evaluation platform to advance the development of robust and trustworthy medical AI.

cs.CV↗

arXiv2025

CookBench: A Long-Horizon Embodied Planning Benchmark for Complex Cooking Scenarios

Embodied Planning is dedicated to the goal of creating agents capable of executing long-horizon tasks in complex physical worlds. However, existing embodied planning benchmarks frequently feature short-horizon tasks and coarse-grained action primitives. To address this challenge, we introduce CookBench, a benchmark for long-horizon planning in complex cooking scenarios. By leveraging a high-fidelity simulation environment built upon the powerful Unity game engine, we define frontier AI challenges in a complex, realistic environment. The core task in CookBench is designed as a two-stage process. First, in Intention Recognition, an agent needs to accurately parse a user's complex intent. Second, in Embodied Interaction, the agent should execute the identified cooking goal through a long-horizon, fine-grained sequence of physical actions. Unlike existing embodied planning benchmarks, we refine the action granularity to a spatial level that considers crucial operational information while abstracting away low-level robotic control. Besides, We provide a comprehensive toolset that encapsulates the simulator. Its unified API supports both macro-level operations, such as placing orders and purchasing ingredients, and a rich set of fine-grained embodied actions for physical interaction, enabling researchers to focus on high-level planning and decision-making. Furthermore, we present an in-depth analysis of state-of-the-art, closed-source Large Language Model and Vision-Language Model, revealing their major shortcomings and challenges posed by complex, long-horizon tasks. The full benchmark will be open-sourced to facilitate future research.

cs.RO↗

arXiv2025

ETrace:Event-Driven Vulnerability Detection in Smart Contracts via LLM-Based Trace Analysis

With the advance application of blockchain technology in various fields, ensuring the security and stability of smart contracts has emerged as a critical challenge. Current security analysis methodologies in vulnerability detection can be categorized into static analysis and dynamic analysis methods.However, these existing traditional vulnerability detection methods predominantly rely on analyzing original contract code, not all smart contracts provide accessible code.We present ETrace, a novel event-driven vulnerability detection framework for smart contracts, which uniquely identifies potential vulnerabilities through LLM-powered trace analysis without requiring source code access. By extracting fine-grained event sequences from transaction logs, the framework leverages Large Language Models (LLMs) as adaptive semantic interpreters to reconstruct event analysis through chain-of-thought reasoning. ETrace implements pattern-matching to establish causal links between transaction behavior patterns and known attack behaviors. Furthermore, we validate the effectiveness of ETrace through preliminary experimental results.

cs.CR↗

arXiv2025

ReviewAgents: Bridging the Gap Between Human and AI-Generated Paper Reviews

Academic paper review is a critical yet time-consuming task within the research community. With the increasing volume of academic publications, automating the review process has become a significant challenge. The primary issue lies in generating comprehensive, accurate, and reasoning-consistent review comments that align with human reviewers' judgments. In this paper, we address this challenge by proposing ReviewAgents, a framework that leverages large language models (LLMs) to generate academic paper reviews. We first introduce a novel dataset, Review-CoT, consisting of 142k review comments, designed for training LLM agents. This dataset emulates the structured reasoning process of human reviewers-summarizing the paper, referencing relevant works, identifying strengths and weaknesses, and generating a review conclusion. Building upon this, we train LLM reviewer agents capable of structured reasoning using a relevant-paper-aware training method. Furthermore, we construct ReviewAgents, a multi-role, multi-LLM agent review framework, to enhance the review comment generation process. Additionally, we propose ReviewBench, a benchmark for evaluating the review comments generated by LLMs. Our experimental results on ReviewBench demonstrate that while existing LLMs exhibit a certain degree of potential for automating the review process, there remains a gap when compared to human-generated reviews. Moreover, our ReviewAgents framework further narrows this gap, outperforming advanced LLMs in generating review comments.

cs.CL↗

arXiv2025

RPHunter: Unveiling Rug Pull Schemes in Crypto Token via Code-and-Transaction Fusion Analysis

Rug pull scams have emerged as a persistent threat to cryptocurrency, causing significant financial losses. A typical scenario involves scammers deploying honeypot contracts to attract investments, restricting token sales, and draining the funds, which leaves investors with worthless tokens. Current methods either rely on predefined patterns to detect code risks or utilize statistical transaction data to train detection models. However, real-world Rug Pull schemes often involve a complex interplay between malicious code and suspicious transaction behaviors. These methods, which solely focus on one aspect, fall short in detecting such schemes effectively. In this paper, we propose RPHunter, a novel technique that integrates code and transaction for Rug Pull detection. First, RPHunter establishes declarative rules and performs flow analysis to extract code risk information, further constructing a semantic risk code graph (SRCG). Meanwhile, to leverage transaction information, RPHunter formulates dynamic token transaction activities as a token flow behavior graph (TFBG) in which nodes and edges are characterized from network structure and market manipulation perspectives. Finally, RPHunter employs graph neural networks to extract complementary features from SRCG and TFBG, integrating them through an attention fusion model to enhance the detection of Rug Pull. We manually analyzed 645 Rug Pull incidents from code and transaction aspects and constructed a ground-truth dataset. We evaluated RPHunter on our dataset, achieving a precision of 95.3%, a recall of 93.8% and an F1 score of 94.5%, which highlights superior performance compared to existing methods. Furthermore, when applied to the real-world scenarios, RPHunter has identified 4801 Rug Pull tokens, achieving a precision of 90.7%.

cs.SE↗

arXiv2025

The Hidden Life of Tokens: Reducing Hallucination of Large Vision-Language Models via Visual Information Steering

Large Vision-Language Models (LVLMs) can reason effectively over both textual and visual inputs, but they tend to hallucinate syntactically coherent yet visually ungrounded contents. In this paper, we investigate the internal dynamics of hallucination by examining the tokens logits ranking throughout the generation process, revealing three key patterns in how LVLMs process information: (1) gradual visual information loss - visually grounded tokens gradually become less favored throughout generation, and (2) early excitation - semantically meaningful tokens achieve peak activation in the layers earlier than the final layer. (3) hidden genuine information - visually grounded tokens though not being eventually decoded still retain relatively high rankings at inference. Based on these insights, we propose VISTA (Visual Information Steering with Token-logit Augmentation), a training-free inference-time intervention framework that reduces hallucination while promoting genuine information. VISTA works by combining two complementary approaches: reinforcing visual information in activation space and leveraging early layer activations to promote semantically meaningful decoding. Compared to existing methods, VISTA requires no external supervision and is applicable to various decoding strategies. Extensive experiments show that VISTA on average reduces hallucination by about 40% on evaluated open-ended generation task, and it consistently outperforms existing methods on four benchmarks across four architectures under three decoding strategies. Code is available at https://github.com/LzVv123456/VISTA.

cs.CV↗

arXiv2025

MTADiffusion: Mask Text Alignment Diffusion Model for Object Inpainting

Advancements in generative models have enabled image inpainting models to generate content within specific regions of an image based on provided prompts and masks. However, existing inpainting methods often suffer from problems such as semantic misalignment, structural distortion, and style inconsistency. In this work, we present MTADiffusion, a Mask-Text Alignment diffusion model designed for object inpainting. To enhance the semantic capabilities of the inpainting model, we introduce MTAPipeline, an automatic solution for annotating masks with detailed descriptions. Based on the MTAPipeline, we construct a new MTADataset comprising 5 million images and 25 million mask-text pairs. Furthermore, we propose a multi-task training strategy that integrates both inpainting and edge prediction tasks to improve structural stability. To promote style consistency, we present a novel inpainting style-consistency loss using a pre-trained VGG network and the Gram matrix. Comprehensive evaluations on BrushBench and EditBench demonstrate that MTADiffusion achieves state-of-the-art performance compared to other methods.

cs.CV↗

arXiv2025

CAI: Caption-Sensitive Attention Intervention for Mitigating Object Hallucination in Large Vision-Language Models

Although Large Vision-Language Models (LVLMs) have demonstrated powerful capabilities in interpreting visual information, they frequently produce content that deviates from visual information, leading to object hallucination. To tackle this, recent works mostly depend on expensive manual annotations and training cost, or significantly increase inference time. In this work, we observe that LVLMs' attention to visual information is significantly stronger when answering caption queries compared to non-caption queries. Inspired by this phenomenon, we propose Caption-sensitive Attention Intervention (CAI), a training-free, plug-and-play hallucination mitigation method that leverages the attention activation pattern in response to caption queries to enhance LVLMs' visual perception capability. Extensive experimental results across four benchmarks covering both discriminative and generative tasks, demonstrate that CAI achieves state-of-the-art (SOTA) hallucination mitigating performance only with minimal additional inference cost.

cs.CV↗

arXiv2025

Beyond Fixed Length: Bucket Pre-training is All You Need

Large Language Models (LLMs) have demonstrated exceptional performance across various tasks, with pre-training stage serving as the cornerstone of their capabilities. However, the conventional fixed-length data composition strategy for pre-training presents several practical challenges. When using shorter sequences, documents are often truncated, potentially leading to information loss and affecting the model's ability to capture long-range dependencies. Conversely, longer sequences require concatenation of multiple documents, which can introduce noise and affect the natural document boundaries and semantic coherence as well as require substantial computational overhead. To address these challenges, we first establish three quantitative metrics for evaluating data composition quality: padding ratio, truncation ratio, and concatenation ratio. Building upon these metrics, we propose a novel multi-bucket data composition method that transcends the fixed-length paradigm. Our approach adaptively organizes training data to achieve optimal composition quality as measured by the proposed metrics, offering a more flexible and efficient approach for pre-training. We conduct extensive experiments and the results demonstrate that our proposed method significantly enhances both the efficiency and effectiveness of LLM pre-training.

cs.CL↗

arXiv2025

MaPPER: Multimodal Prior-guided Parameter Efficient Tuning for Referring Expression Comprehension

Referring Expression Comprehension (REC), which aims to ground a local visual region via natural language, is a task that heavily relies on multimodal alignment. Most existing methods utilize powerful pre-trained models to transfer visual/linguistic knowledge by full fine-tuning. However, full fine-tuning the entire backbone not only breaks the rich prior knowledge embedded in the pre-training, but also incurs significant computational costs. Motivated by the recent emergence of Parameter-Efficient Transfer Learning (PETL) methods, we aim to solve the REC task in an effective and efficient manner. Directly applying these PETL methods to the REC task is inappropriate, as they lack the specific-domain abilities for precise local visual perception and visual-language alignment. Therefore, we propose a novel framework of Multimodal Prior-guided Parameter Efficient Tuning, namely MaPPER. Specifically, MaPPER comprises Dynamic Prior Adapters guided by an aligned prior, and Local Convolution Adapters to extract precise local semantics for better visual perception. Moreover, the Prior-Guided Text module is proposed to further utilize the prior for facilitating the cross-modal alignment. Experimental results on three widely-used benchmarks demonstrate that MaPPER achieves the best accuracy compared to the full fine-tuning and other PETL methods with only 1.41% tunable backbone parameters. Our code is available at https://github.com/liuting20/MaPPER.

cs.CV↗

arXiv2025

Exploring and Exploiting the Inherent Efficiency within Large Reasoning Models for Self-Guided Efficiency Enhancement

Recent advancements in large reasoning models (LRMs) have significantly enhanced language models' capabilities in complex problem-solving by emulating human-like deliberative thinking. However, these models often exhibit overthinking (i.e., the generation of unnecessarily verbose and redundant content), which hinders efficiency and inflates inference cost. In this work, we explore the representational and behavioral origins of this inefficiency, revealing that LRMs inherently possess the capacity for more concise reasoning. Empirical analyses show that correct reasoning paths vary significantly in length, and the shortest correct responses often suffice, indicating untapped efficiency potential. Exploiting these findings, we propose two lightweight methods to enhance LRM efficiency. First, we introduce Efficiency Steering, a training-free activation steering technique that modulates reasoning behavior via a single direction in the model's representation space. Second, we develop Self-Rewarded Efficiency RL, a reinforcement learning framework that dynamically balances task accuracy and brevity by rewarding concise correct solutions. Extensive experiments on seven LRM backbones across multiple mathematical reasoning benchmarks demonstrate that our methods significantly reduce reasoning length while preserving or improving task performance. Our results highlight that reasoning efficiency can be improved by leveraging and guiding the intrinsic capabilities of existing models in a self-guided manner.

cs.AI↗

arXiv2025

Detecting State Manipulation Vulnerabilities in Smart Contracts Using LLM and Static Analysis

An increasing number of DeFi protocols are gaining popularity, facilitating transactions among multiple anonymous users. State Manipulation is one of the notorious attacks in DeFi smart contracts, with price variable being the most commonly exploited state variable-attackers manipulate token prices to gain illicit profits. In this paper, we propose PriceSleuth, a novel method that leverages the Large Language Model (LLM) and static analysis to detect Price Manipulation (PM) attacks proactively. PriceSleuth firstly identifies core logic function related to price calculation in DeFi contracts. Then it guides LLM to locate the price calculation code statements. Secondly, PriceSleuth performs backward dependency analysis of price variables, instructing LLM in detecting potential price manipulation. Finally, PriceSleuth utilizes propagation analysis of price variables to assist LLM in detecting whether these variables are maliciously exploited. We presented preliminary experimental results to substantiate the effectiveness of PriceSleuth . And we outline future research directions for PriceSleuth.

cs.SE↗