arXiv subjects

Rui Liu

Publications and source records attributed to Rui Liu.

At least 145 records · Page 8Linked to original sources

arXiv2026

Save the Good Prefix: Precise Error Penalization via Process-Supervised RL to Enhance LLM Reasoning

Reinforcement learning (RL) has emerged as a powerful framework for improving the reasoning capabilities of large language models (LLMs). However, most existing RL approaches rely on sparse outcome rewards, which fail to credit correct intermediate steps in partially successful solutions. Process reward models (PRMs) offer fine-grained step-level supervision, but their scores are often noisy and difficult to evaluate. As a result, recent PRM benchmarks focus on a more objective capability: detecting the first incorrect step in a reasoning path. However, this evaluation target is misaligned with how PRMs are typically used in RL, where their step-wise scores are treated as raw rewards to maximize. To bridge this gap, we propose Verifiable Prefix Policy Optimization (VPPO), which uses PRMs only to localize the first error during RL. Given an incorrect rollout, VPPO partitions the trajectory into a verified correct prefix and an erroneous suffix based on the first error, rewarding the former while applying targeted penalties only after the detected mistake. This design yields stable, interpretable learning signals and improves credit assignment. Across multiple reasoning benchmarks, VPPO consistently outperforms sparse-reward RL and prior PRM-guided baselines on both Pass@1 and Pass@K.

cs.LG↗

arXiv2026

SafeThinker: Reasoning about Risk to Deepen Safety Beyond Shallow Alignment

Despite the intrinsic risk-awareness of Large Language Models (LLMs), current defenses often result in shallow safety alignment, rendering models vulnerable to disguised attacks (e.g., prefilling) while degrading utility. To bridge this gap, we propose SafeThinker, an adaptive framework that dynamically allocates defensive resources via a lightweight gateway classifier. Based on the gateway's risk assessment, inputs are routed through three distinct mechanisms: (i) a Standardized Refusal Mechanism for explicit threats to maximize efficiency; (ii) a Safety-Aware Twin Expert (SATE) module to intercept deceptive attacks masquerading as benign queries; and (iii) a Distribution-Guided Think (DDGT) component that adaptively intervenes during uncertain generation. Experiments show that SafeThinker significantly lowers attack success rates across diverse jailbreak strategies without compromising utility, demonstrating that coordinating intrinsic judgment throughout the generation process effectively balances robustness and practicality.

cs.CR↗

arXiv2026

Emotion and Acoustics Should Agree: Cross-Level Inconsistency Analysis for Audio Deepfake Detection

Audio Deepfake Detection (ADD) aims to detect spoof speech from bonafide speech. Most prior studies assume that stronger correlations within or across acoustic and emotional features imply authenticity, and thus focus on enhancing or measuring such correlations. However, existing methods often treat acoustic and emotional features in isolation or rely on correlation metrics, which overlook subtle desynchronization between them and smooth out abrupt discontinuities. To address these issues, we propose EAI-ADD, which treats cross level emotion acoustic inconsistency as the primary detection signal. We first project emotional and acoustic representations into a comparable space. Then we progressively integrate frame level and utterance level emotion features with acoustic features to capture cross level emotion acoustic inconsistencies across different temporal granularities. Experimental results on the ASVspoof 2019LA and 2021LA datasets demonstrate that the proposed EAI-ADD outperforms baselines, providing a more effective solution for audio anti spoofing detection.

cs.SD↗

arXiv2026

Panacea: Mitigating Harmful Fine-tuning for Large Language Models via Post-fine-tuning Perturbation

Harmful fine-tuning attack introduces significant security risks to the fine-tuning services. Main-stream defenses aim to vaccinate the model such that the later harmful fine-tuning attack is less effective. However, our evaluation results show that such defenses are fragile--with a few fine-tuning steps, the model still can learn the harmful knowledge. To this end, we do further experiment and find that an embarrassingly simple solution--adding purely random perturbations to the fine-tuned model, can recover the model from harmful behaviors, though it leads to a degradation in the model's fine-tuning performance. To address the degradation of fine-tuning performance, we further propose Panacea, which optimizes an adaptive perturbation that will be applied to the model after fine-tuning. Panacea maintains model's safety alignment performance without compromising downstream fine-tuning performance. Comprehensive experiments are conducted on different harmful ratios, fine-tuning tasks and mainstream LLMs, where the average harmful scores are reduced by up-to 21.2%, while maintaining fine-tuning performance. As a by-product, we analyze the adaptive perturbation and show that different layers in various LLMs have distinct safety affinity, which coincide with finding from several previous study. Source code available at https://github.com/w-yibo/Panacea.

cs.CL↗

arXiv2026

Generative Modeling of Human-Computer Interfaces with Diffusion Processes and Conditional Control

This study investigates human-computer interface generation based on diffusion models to overcome the limitations of traditional template-based design and fixed rule-driven methods. It first analyzes the key challenges of interface generation, including the diversity of interface elements, the complexity of layout logic, and the personalization of user needs. A generative framework centered on the diffusion-reverse diffusion process is then proposed, with conditional control introduced in the reverse diffusion stage to integrate user intent, contextual states, and task constraints, enabling unified modeling of visual presentation and interaction logic. In addition, regularization constraints and optimization objectives are combined to ensure the rationality and stability of the generated interfaces. Experiments are conducted on a public interface dataset with systematic evaluations, including comparative experiments, hyperparameter sensitivity tests, environmental sensitivity tests, and data sensitivity tests. Results show that the proposed method outperforms representative models in mean squared error, structural similarity, peak signal-to-noise ratio, and mean absolute error, while maintaining strong robustness under different parameter settings and environmental conditions. Overall, the diffusion model framework effectively improves the diversity, rationality, and intelligence of interface generation, providing a feasible solution for automated interface generation in complex interaction scenarios.

cs.HC↗

arXiv2026

MoE Adapter for Large Audio Language Models: Sparsity, Disentanglement, and Gradient-Conflict-Free

Extending the input modality of Large Language Models~(LLMs) to the audio domain is essential for achieving comprehensive multimodal perception. However, it is well-known that acoustic information is intrinsically \textit{heterogeneous}, entangling attributes such as speech, music, and environmental context. Existing research is limited to a dense, parameter-shared adapter to model these diverse patterns, which induces \textit{gradient conflict} during optimization, as parameter updates required for distinct attributes contradict each other. To address this limitation, we introduce the \textit{\textbf{MoE-Adapter}}, a sparse Mixture-of-Experts~(MoE) architecture designed to decouple acoustic information. Specifically, it employs a dynamic gating mechanism that routes audio tokens to specialized experts capturing complementary feature subspaces while retaining shared experts for global context, thereby mitigating gradient conflicts and enabling fine-grained feature learning. Comprehensive experiments show that the MoE-Adapter achieves superior performance on both audio semantic and paralinguistic tasks, consistently outperforming dense linear baselines with comparable computational costs. Furthermore, we will release the related code and models to facilitate future research.

cs.SD↗

arXiv2026

AEGIS: Exploring the Limit of World Knowledge Capabilities for Unified Mulitmodal Models

The capability of Unified Multimodal Models (UMMs) to apply world knowledge across diverse tasks remains a critical, unresolved challenge. Existing benchmarks fall short, offering only siloed, single-task evaluations with limited diagnostic power. To bridge this gap, we propose AEGIS (\emph{i.e.}, \textbf{A}ssessing \textbf{E}diting, \textbf{G}eneration, \textbf{I}nterpretation-Understanding for \textbf{S}uper-intelligence), a comprehensive multi-task benchmark covering visual understanding, generation, editing, and interleaved generation. AEGIS comprises 1,050 challenging, manually-annotated questions spanning 21 topics (including STEM, humanities, daily life, etc.) and 6 reasoning types. To concretely evaluate the performance of UMMs in world knowledge scope without ambiguous metrics, we further propose Deterministic Checklist-based Evaluation (DCE), a protocol that replaces ambiguous prompt-based scoring with atomic ``Y/N'' judgments, to enhance evaluation reliability. Our extensive experiments reveal that most UMMs exhibit severe world knowledge deficits and that performance degrades significantly with complex reasoning. Additionally, simple plug-in reasoning modules can partially mitigate these vulnerabilities, highlighting a promising direction for future research. These results highlight the importance of world-knowledge-based reasoning as a critical frontier for UMMs.

cs.CV↗

arXiv2025

Metallic solid-state hydrogen storage crystals achieved through chemical precompression under ambient conditions

Improving hydrogen storage density is essential for reducing the extreme conditions required in applications such as nuclear fusion. However, the recognition of metallic hydrogen as the "Holy Grail" of high-pressure science highlights the difficulty of high-density hydrogen aggregation. Here, we report a solid-state crystal H9@C20 formed by embedding hydrogen atoms into C20 fullerene cages and utilizing chemical precompression, which remains stable under ambient pressure and temperature conditions and exhibits metallic properties. This precompression effect is reflected in the formation of C-H bonds within the cage and C-C bonds between cages, resulting in the transformation of all C atoms from sp2 to sp3 hybridization with inward and outward distortions, while promoting delocalized multicenter bonding within the H9 aggregate. In particular, the hydrogen density inside the C20 cage exceeds that of solid hydrogen, achieving a uniform discrete distribution with H9 as monomers. Further study reveals that filling hydrogen molecules into voids between H9@C20 primitive cells can increase hydrogen content while maintaining structural stability, forming a solid-gas mixed hydrogen storage crystal. Our findings provide a basis for developing high-density hydrogen storage materials under ambient conditions.

cond-mat.mtrl-sci↗

arXiv2025

CAML: Collaborative Auxiliary Modality Learning for Multi-Agent Systems

Multi-modal learning has emerged as a key technique for improving performance across domains such as autonomous driving, robotics, and reasoning. However, in certain scenarios, particularly in resource-constrained environments, some modalities available during training may be absent during inference. While existing frameworks effectively utilize multiple data sources during training and enable inference with reduced modalities, they are primarily designed for single-agent settings. This poses a critical limitation in dynamic environments such as connected autonomous vehicles (CAV), where incomplete data coverage can lead to decision-making blind spots. Conversely, some works explore multi-agent collaboration but without addressing missing modality at test time. To overcome these limitations, we propose Collaborative Auxiliary Modality Learning (CAML), a novel multi-modal multi-agent framework that enables agents to collaborate and share multi-modal data during training, while allowing inference with reduced modalities during testing. Experimental results in collaborative decision-making for CAV in accident-prone scenarios demonstrate that CAML achieves up to a 58.1% improvement in accident detection. Additionally, we validate CAML on real-world aerial-ground robot data for collaborative semantic segmentation, achieving up to a 10.6% improvement in mIoU.

cs.RO↗

arXiv2025

Scaling Reinforcement Learning for Content Moderation with Large Language Models

Content moderation at scale remains one of the most pressing challenges in today's digital ecosystem, where billions of user- and AI-generated artifacts must be continuously evaluated for policy violations. Although recent advances in large language models (LLMs) have demonstrated strong potential for policy-grounded moderation, the practical challenges of training these systems to achieve expert-level accuracy in real-world settings remain largely unexplored, particularly in regimes characterized by label sparsity, evolving policy definitions, and the need for nuanced reasoning beyond shallow pattern matching. In this work, we present a comprehensive empirical investigation of scaling reinforcement learning (RL) for content classification, systematically evaluating multiple RL training recipes and reward-shaping strategies-including verifiable rewards and LLM-as-judge frameworks-to transform general-purpose language models into specialized, policy-aligned classifiers across three real-world content moderation tasks. Our findings provide actionable insights for industrial-scale moderation systems, demonstrating that RL exhibits sigmoid-like scaling behavior in which performance improves smoothly with increased training data, rollouts, and optimization steps before gradually saturating. Moreover, we show that RL substantially improves performance on tasks requiring complex policy-grounded reasoning while achieving up to 100x higher data efficiency than supervised fine-tuning, making it particularly effective in domains where expert annotations are scarce or costly.

cs.AI↗

arXiv2025

Height and Energy Evolution of X-ray Double Sources in a Solar Flare

In the standard model, magnetic reconnection at a vertical current sheet above the flare arcade is key to explaining many aspects of solar eruptions. The supra-arcade region is where the vertical current sheet is supposedly located, and X-ray/EUV emission therein reflects underlying energy release and transport processes, therefore providing valuable insight into the structure and evolution of the current sheet. Previous studies have focused primarily on the impulsive phase of flares, but phenomena in the decay phase are also crucial for understanding the complete flaring scenario. In this paper, we investigated an M6.7-class limb flare that occurred on August 28, 2022, combining observations from the Solar Orbiter (SolO) and Solar Dynamics Observatory (SDO). Coronal X-ray sources are continually observed by the Spectrometer/Telescope for Imaging X-rays (STIX) onboard SolO for over two hours, revealing a multi-phase evolution with varying velocities and multiple substructures, with higher-energy components consistently appearing at higher altitudes. Such a height-energy relation is notably observed in a double coronal source during the decay-phase, which is dominated by thermal emission. The energy distribution of the double source distinguish itself from previous studies that showed a symmetric distribution, with the higher-energy components being closer to the center of the double source during the impulsive phase. Obtained from two opposite side-on perspectives on the supra-arcade region, these findings reveal the spatio-temporal complexity of the energy release process in the post-flare current sheet during the decay phase.

astro-ph.SR↗

arXiv2025

Stable and Efficient Single-Rollout RL for Multimodal Reasoning

Reinforcement Learning with Verifiable Rewards (RLVR) has become a key paradigm to improve the reasoning capabilities of Multimodal Large Language Models (MLLMs). However, prevalent group-based algorithms such as GRPO require multi-rollout sampling for each prompt. While more efficient single-rollout variants have recently been explored in text-only settings, we find that they suffer from severe instability in multimodal contexts, often leading to training collapse. To address this training efficiency-stability trade-off, we introduce $\textbf{MSSR}$ (Multimodal Stabilized Single-Rollout), a group-free RLVR framework that achieves both stable optimization and effective multimodal reasoning performance. MSSR achieves this via an entropy-based advantage-shaping mechanism that adaptively regularizes advantage magnitudes, preventing collapse and maintaining training stability. While such mechanisms have been used in group-based RLVR, we show that in the multimodal single-rollout setting they are not merely beneficial but essential for stability. In in-distribution evaluations, MSSR demonstrates superior training compute efficiency, achieving similar validation accuracy to the group-based baseline with half the training steps. When trained for the same number of steps, MSSR's performance surpasses the group-based baseline and shows consistent generalization improvements across five diverse reasoning-intensive benchmarks. Together, these results demonstrate that MSSR enables stable, compute-efficient, and effective RLVR for complex multimodal reasoning tasks.

cs.LG↗

arXiv2025

Transformer-Based Modeling of User Interaction Sequences for Dwell Time Prediction in Human-Computer Interfaces

This study investigates the task of dwell time prediction and proposes a Transformer framework based on interaction behavior modeling. The method first represents user interaction sequences on the interface by integrating dwell duration, click frequency, scrolling behavior, and contextual features, which are mapped into a unified latent space through embedding and positional encoding. On this basis, a multi-head self-attention mechanism is employed to capture long-range dependencies, while a feed-forward network performs deep nonlinear transformations to model the dynamic patterns of dwell time. Multiple comparative experiments are conducted with BILSTM, DRFormer, FedFormer, and iTransformer as baselines under the same conditions. The results show that the proposed method achieves the best performance in terms of MSE, RMSE, MAPE, and RMAE, and more accurately captures the complex patterns in interaction behavior. In addition, sensitivity experiments are carried out on hyperparameters and environments to examine the impact of the number of attention heads, sequence window length, and device environment on prediction performance, which further demonstrates the robustness and adaptability of the method. Overall, this study provides a new solution for dwell time prediction from both theoretical and methodological perspectives and verifies its effectiveness in multiple aspects.

cs.HC↗

arXiv2025

Scene-aware SAR ship detection guided by unsupervised sea-land segmentation

DL based Synthetic Aperture Radar (SAR) ship detection has tremendous advantages in numerous areas. However, it still faces some problems, such as the lack of prior knowledge, which seriously affects detection accuracy. In order to solve this problem, we propose a scene-aware SAR ship detection method based on unsupervised sea-land segmentation. This method follows a classical two-stage framework and is enhanced by two models: the unsupervised land and sea segmentation module (ULSM) and the land attention suppression module (LASM). ULSM and LASM can adaptively guide the network to reduce attention on land according to the type of scenes (inshore scene and offshore scene) and add prior knowledge (sea land segmentation information) to the network, thereby reducing the network's attention to land directly and enhancing offshore detection performance relatively. This increases the accuracy of ship detection and enhances the interpretability of the model. Specifically, in consideration of the lack of land sea segmentation labels in existing deep learning-based SAR ship detection datasets, ULSM uses an unsupervised approach to classify the input data scene into inshore and offshore types and performs sea-land segmentation for inshore scenes. LASM uses the sea-land segmentation information as prior knowledge to reduce the network's attention to land. We conducted our experiments using the publicly available SSDD dataset, which demonstrated the effectiveness of our network.

cs.CV↗

arXiv2025

MMMamba: A Versatile Cross-Modal In Context Fusion Framework for Pan-Sharpening and Zero-Shot Image Enhancement

Pan-sharpening aims to generate high-resolution multispectral (HRMS) images by integrating a high-resolution panchromatic (PAN) image with its corresponding low-resolution multispectral (MS) image. To achieve effective fusion, it is crucial to fully exploit the complementary information between the two modalities. Traditional CNN-based methods typically rely on channel-wise concatenation with fixed convolutional operators, which limits their adaptability to diverse spatial and spectral variations. While cross-attention mechanisms enable global interactions, they are computationally inefficient and may dilute fine-grained correspondences, making it difficult to capture complex semantic relationships. Recent advances in the Multimodal Diffusion Transformer (MMDiT) architecture have demonstrated impressive success in image generation and editing tasks. Unlike cross-attention, MMDiT employs in-context conditioning to facilitate more direct and efficient cross-modal information exchange. In this paper, we propose MMMamba, a cross-modal in-context fusion framework for pan-sharpening, with the flexibility to support image super-resolution in a zero-shot manner. Built upon the Mamba architecture, our design ensures linear computational complexity while maintaining strong cross-modal interaction capacity. Furthermore, we introduce a novel multimodal interleaved (MI) scanning mechanism that facilitates effective information exchange between the PAN and MS modalities. Extensive experiments demonstrate the superior performance of our method compared to existing state-of-the-art (SOTA) techniques across multiple tasks and benchmarks.

cs.CV↗

arXiv2025

Confirmatory Biomarker Identification with k-FWER Control Using Derandomized Knockoffs with Cox Regression

Selecting important features in high-dimensional survival analysis is critical for identifying confirmatory biomarkers while maintaining rigorous error control. In this paper, we propose a derandomized knockoffs procedure for Cox regression that enhances stability in feature selection while maintaining rigorous control over the k-familywise error rate (k-FWER). By aggregating across multiple randomized knockoff realizations, our approach mitigates the instability commonly observed with conventional knockoffs. Through extensive simulations, we demonstrate that our method consistently outperforms standard knockoffs in both selection power and error control. Moreover, we apply our procedure to a clinical dataset on primary biliary cirrhosis (PBC) to identify key prognostic biomarkers associated with patient survival. The results confirm the superior stability of the derandomized knockoffs method, allowing for a more reliable identification of important clinical variables. Additionally, our approach is applicable to datasets containing both continuous and categorical covariates, broadening its utility in real-world biomedical studies. This framework provides a robust and interpretable solution for high-dimensional survival analysis, making it particularly suitable for applications requiring precise and stable variable selection.

stat.ME↗

arXiv2025

OCCDiff: Occupancy Diffusion Model for High-Fidelity 3D Building Reconstruction from Noisy Point Clouds

A major challenge in reconstructing buildings from LiDAR point clouds lies in accurately capturing building surfaces under varying point densities and noise interference. To flexibly gather high-quality 3D profiles of the building in diverse resolution, we propose OCCDiff applying latent diffusion in the occupancy function space. Our OCCDiff combines a latent diffusion process with a function autoencoder architecture to generate continuous occupancy functions evaluable at arbitrary locations. Moreover, a point encoder is proposed to provide condition features to diffusion learning, constraint the final occupancy prediction for occupancy decoder, and insert multi-modal features for latent generation to latent encoder. To further enhance the model performance, a multi-task training strategy is employed, ensuring that the point encoder learns diverse and robust feature representations. Empirical results show that our method generates physically consistent samples with high fidelity to the target distribution and exhibits robustness to noisy data.

cs.CV↗

arXiv2025

DraCo: Draft as CoT for Text-to-Image Preview and Rare Concept Generation

Recent unified multimodal large language models (MLLMs) have shown impressive capabilities, incorporating chain-of-thought (CoT) reasoning for enhanced text-to-image generation. However, existing approaches remain limited, either treating the model merely as a standalone generator or relying on abstract textual planning. To this end, we propose Draft-as-CoT (DraCo), a novel interleaved reasoning paradigm that fully leverages both textual and visual contents in CoT for better planning and verification. Our method first generates a low-resolution draft image as preview, providing more concrete and structural visual planning and guidance. Then, we employ the model's inherent understanding capability to verify potential semantic misalignments between the draft and input prompt, and performs refinement through selective corrections with super-resolution. In this way, our approach addresses two fundamental challenges: the coarse-grained nature of textual planning and the difficulty in generating rare attribute combinations. To support training, we curate DraCo-240K, aiming to enhance three atomic capabilities spanning general correction, instance manipulation, and layout reorganization. Supported by DraCo-CFG, a specialized classifier-free guidance (CFG) strategy for interleaved reasoning, DraCo achieves a tremendous increase on GenEval (+8%), Imagine-Bench (+0.91), and GenEval++ (+3%), significantly outperforming direct generation and other generation methods empowered by CoT.

cs.CV↗