arXiv subjects

Haifeng Wang

Publications and source records attributed to Haifeng Wang.

At least 55 records · Page 3Linked to original sources

arXiv2026

Distributional Clarity: The Hidden Driver of RL-Friendliness in Large Language Models

Language model families exhibit striking disparity in their capacity to benefit from reinforcement learning: under identical training, models like Qwen achieve substantial gains, while others like Llama yield limited improvements. Complementing data-centric approaches, we reveal that this disparity reflects a hidden structural property: \textbf{distributional clarity} in probability space. Through a three-stage analysis-from phenomenon to mechanism to interpretation-we uncover that RL-friendly models exhibit intra-class compactness and inter-class separation in their probability assignments to correct vs. incorrect responses. We quantify this clarity using the \textbf{Silhouette Coefficient} ($S$) and demonstrate that (1) high $S$ correlates strongly with RL performance; (2) low $S$ is associated with severe logic errors and reasoning instability. To confirm this property, we introduce a Silhouette-Aware Reweighting strategy that prioritizes low-$S$ samples during training. Experiments across six mathematical benchmarks show consistent improvements across all model families, with gains up to 5.9 points on AIME24. Our work establishes distributional clarity as a fundamental, trainable property underlying RL-Friendliness.

cs.CL↗

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

Character-R1: Enhancing Role-Aware Reasoning in Role-Playing Agents via RLVR

Current role-playing agents (RPAs) are typically constructed by imitating surface-level behaviors, but this approach lacks internal cognitive consistency, often causing out-of-character errors in complex situations. To address this, we propose Character-R1, a framework designed to provide comprehensive verifiable reward signals for effective role-aware reasoning, which are missing in recent studies. Specifically, our framework comprises three core designs: (1) Cognitive Focus Reward, which enforces explicit label-based analysis of 10 character elements (e.g., worldview) to structure internal cognition; (2) Reference-Guided Reward, which utilizes overlap-based metrics with reference responses as optimization anchors to enhance exploration and performance; and (3) Character-Conditioned Reward Normalization, which adjusts reward distributions based on character categories to ensure robust optimization across heterogeneous roles. Extensive experiments demonstrate that Character-R1 significantly outperforms existing methods in knowledge, memory and others.

cs.CL↗

arXiv2025

Rapid prediction of cardiac activation in the left ventricle with geometric deep learning: a step towards cardiac resynchronization therapy planning

Cardiac resynchronization therapy (CRT) is a common intervention for patients with dyssynchronous heart failure, yet approximately one-third of recipients fail to respond, partly due to suboptimal lead placement. Identifying optimal pacing sites remains challenging, largely due to patient-specific anatomical variability and limitations of current individualized planning strategies. In a step toward an in-silico approach, we develop two geometric deep learning models, based on graph neural network (GNN) and geometry-informed neural operator (GINO), to predict activation time maps on left ventricular (LV) geometries in real time. Trained on a large dataset generated from finite-element simulations spanning a wide range of synthetic LV shapes, pacing site configurations, and tissue conductivities, the GINO model outperforms the GNN on synthetic cases (1.38% vs 2.44% error), while both demonstrate comparable performance on real-world LV geometries (GINO: 4.79% vs GNN: 4.07%). Using the trained models, we develop a workflow to identify an optimal pacing site on the LV from a given activation time map and show that both models can robustly recover ground-truth subject-specific parameters from noisy inputs. In conjunction with an interactive web-based interface (https://dcsim.egr.msu.edu/), this study shows potential and motivates future extension toward a clinical decision-support tool for personalized pre-procedural CRT optimization.

cs.CE↗

arXiv2025

Principal Component Analysis for Nonlinear Optical Microscopic Chemical Imaging of Nitrogen Gas

We have implemented principal component analysis for microscopic wide-field chemical imaging via coherent Raman spectroscopy. Microscopic imaging of nitrogen gas has been challenging due to extremely weak signals stemming from low order Raman interaction. Wide-field coherent Raman micro-spectroscopy has demonstrated the ability to chemically distinguish nitrogen gas although it has been difficult to quantify spatial-density information due to significant levels of background noise. By subtracting the Gaussian beam shape and removing contributions from uninformative noise simultaneously from the set of images, we can reconstruct the normalized intensity fluctuations. Our analysis demonstrates that nitrogen gas within microvolume can be rapidly monitored under ambient conditions in less than 0.2 seconds. We believe that our work has the potential to improve visualization of microscopic flows due to molecular dynamics of gases and/or liquids otherwise invisible to infrared optical techniques.

physics.optics↗

arXiv2025

V-ITI: Mitigating Hallucinations in Multimodal Large Language Models via Visual Inference-Time Intervention

Multimodal Large Language Models (MLLMs) excel in numerous vision-language tasks yet suffer from hallucinations, producing content inconsistent with input visuals, that undermine reliability in precision-sensitive domains. This issue stems from a fundamental problem of visual neglect, where models fail to adequately prioritize input images. Existing methods typically alleviate hallucinations by intervening in the attention score or output logits, focusing on "how to intervene" but overlooking the prerequisite "when to intervene", which leads to the "over-intervention" problem and subsequently introduces new hallucinations and unnecessary computational overhead. To address this gap, we first investigate the mechanism of visual neglect and reveal it can be accurately detected via head-level activation patterns in MLLMs. We thus propose V-ITI, a lightweight visual inference-time intervention framework integrating a Visual Neglect Detector that identifies visual neglect via head-level discriminative probes and a Visual Recall Intervenor that modulates activations with prestored visual activation information only when the visual neglect is detected. Extensive experiments across eight benchmarks and different MLLM families demonstrate that V-ITI consistently mitigates vision-related hallucinations while preserving general task performance.

cs.CV↗

arXiv2025

Semantic Energy: Detecting LLM Hallucination Beyond Entropy

Large Language Models (LLMs) are being increasingly deployed in real-world applications, but they remain susceptible to hallucinations, which produce fluent yet incorrect responses and lead to erroneous decision-making. Uncertainty estimation is a feasible approach to detect such hallucinations. For example, semantic entropy estimates uncertainty by considering the semantic diversity across multiple sampled responses, thus identifying hallucinations. However, semantic entropy relies on post-softmax probabilities and fails to capture the model's inherent uncertainty, causing it to be ineffective in certain scenarios. To address this issue, we introduce Semantic Energy, a novel uncertainty estimation framework that leverages the inherent confidence of LLMs by operating directly on logits of penultimate layer. By combining semantic clustering with a Boltzmann-inspired energy distribution, our method better captures uncertainty in cases where semantic entropy fails. Experiments across multiple benchmarks show that Semantic Energy significantly improves hallucination detection and uncertainty estimation, offering more reliable signals for downstream applications such as hallucination detection.

cs.LG↗

arXiv2025

BEE-RAG: Balanced Entropy Engineering for Retrieval-Augmented Generation

With the rapid advancement of large language models (LLMs), retrieval-augmented generation (RAG) has emerged as a critical approach to supplement the inherent knowledge limitations of LLMs. However, due to the typically large volume of retrieved information, RAG tends to operate with long context lengths. From the perspective of entropy engineering, we identify unconstrained entropy growth and attention dilution due to long retrieval context as significant factors affecting RAG performance. In this paper, we propose the balanced entropy-engineered RAG (BEE-RAG) framework, which improves the adaptability of RAG systems to varying context lengths through the principle of entropy invariance. By leveraging balanced context entropy to reformulate attention dynamics, BEE-RAG separates attention sensitivity from context length, ensuring a stable entropy level. Building upon this, we introduce a zero-shot inference strategy for multi-importance estimation and a parameter-efficient adaptive fine-tuning mechanism to obtain the optimal balancing factor for different settings. Extensive experiments across multiple RAG tasks demonstrate the effectiveness of BEE-RAG.

cs.CL↗

arXiv2025

Towards Large-Scale In-Context Reinforcement Learning by Meta-Training in Randomized Worlds

In-Context Reinforcement Learning (ICRL) enables agents to learn automatically and on-the-fly from their interactive experiences. However, a major challenge in scaling up ICRL is the lack of scalable task collections. To address this, we propose the procedurally generated tabular Markov Decision Processes, named AnyMDP. Through a carefully designed randomization process, AnyMDP is capable of generating high-quality tasks on a large scale while maintaining relatively low structural biases. To facilitate efficient meta-training at scale, we further introduce decoupled policy distillation and induce prior information in the ICRL framework. Our results demonstrate that, with a sufficiently large scale of AnyMDP tasks, the proposed model can generalize to tasks that were not considered in the training set through versatile in-context learning paradigms. The scalable task set provided by AnyMDP also enables a more thorough empirical investigation of the relationship between data distribution and ICRL performance. We further show that the generalization of ICRL potentially comes at the cost of increased task diversity and longer adaptation periods. This finding carries critical implications for scaling robust ICRL capabilities, highlighting the necessity of diverse and extensive task design, and prioritizing asymptotic performance over few-shot adaptation.

cs.LG↗

arXiv2025

Robust Cross-Domain Adaptation in Texture Features Transferring for Wood Chip Moisture Content Prediction

Accurate and quick prediction of wood chip moisture content is critical for optimizing biofuel production and ensuring energy efficiency. The current widely used direct method (oven drying) is limited by its longer processing time and sample destructiveness. On the other hand, existing indirect methods, including near-infrared spectroscopy-based, electrical capacitance-based, and image-based approaches, are quick but not accurate when wood chips come from various sources. Variability in the source material can alter data distributions, undermining the performance of data-driven models. Therefore, there is a need for a robust approach that effectively mitigates the impact of source variability. Previous studies show that manually extracted texture features have the potential to predict wood chip moisture class. Building on this, in this study, we conduct a comprehensive analysis of five distinct texture feature types extracted from wood chip images to predict moisture content. Our findings reveal that a combined feature set incorporating all five texture features achieves an accuracy of 95% and consistently outperforms individual texture features in predicting moisture content. To ensure robust moisture prediction, we propose a domain adaptation method named AdaptMoist that utilizes the texture features to transfer knowledge from one source of wood chip data to another, addressing variability across different domains. We also proposed a criterion for model saving based on adjusted mutual information. The AdaptMoist method improves prediction accuracy across domains by 23%, achieving an average accuracy of 80%, compared to 57% for non-adapted models. These results highlight the effectiveness of AdaptMoist as a robust solution for wood chip moisture content estimation across domains, making it a potential solution for wood chip-reliant industries.

cs.CV↗

arXiv2025

Accurate myocardial T1 mapping at 5T using an improved MOLLI method: A validation study

Background: Accurate myocardial T1 mapping at 5T remains a technical challenge due to field inhomogeneity and prolonged T1 values. The aim of this study is to develop an accurate and clinically applicable myocardial T1 mapping technique for 5T magnetic resonance imaging (MRI) systems and validate its performance in a multicenter study. Methods: The proposed method is based on a 5-(3)-3 MOLLI sequence, dubbed combined-correction MOLLI (coMOLLI), which corrects for both inversion efficiency and readout-induced signal disturbances. Specifically, coMOLLI employs a gradient recalled echo (GRE) readout rather than the commonly used bSSFP readout. Its signal evolution is modeled to estimate T1 values, which incorporates both inversion efficiency and readout disturbances to improve fitting accuracy. To further enhance accuracy, the inversion pulse was redesigned under hardware constraints and the observed B0 and B1 variations over the heart at 5T, using adiabatic hyperbolic secant (HSn) and tangent/hyperbolic tangent (Tan/Tanh) pulses. The method was validated in phantom experiments, as well as in 21 healthy volunteers and 9 patients. Results: The optimized inversion pulse at 5T is the Tan/Tanh pulse with A = 10 kHz, K_s = 4, k = 22, and T_p = 8 ms. In phantom studies, coMOLLI showed high accuracy versus reference IR-FSE, yielding relative errors within 5% for all nine vials. In vivo studies, the average native myocardial T1 values across 21 healthy volunteers were 1468 +/- 48 ms, 1514 +/- 39 ms, and 1545 +/- 50 ms, and blood T1 values were 2182 +/- 132 ms, 2124 +/- 153 ms, and 2131 +/- 158 ms for apical, middle, and base slices, respectively. Conclusion: The coMOLLI method demonstrated high accuracy in phantom studies and feasibility in vivo studies. By adopting the widely used 5-(3)-3 MOLLI acquisition scheme, it shows potential for clinical cardiac imaging at 5T.

physics.med-ph↗

arXiv2025

A Survey on Parallel Reasoning

With the increasing capabilities of Large Language Models (LLMs), parallel reasoning has emerged as a new inference paradigm that enhances reasoning robustness by concurrently exploring multiple lines of thought before converging on a final answer. It has become a significant trend to explore parallel reasoning to overcome the fragility of standard sequential methods and improve practical performance. In this paper, we aim to survey and summarize the progress and challenges of parallel reasoning. We first present a formal definition of parallel reasoning and clarify its distinction from related concepts like Chain-of-Thought. Then, we organize and discuss advanced techniques based on a novel taxonomy, including non-interactive reasoning, interactive reasoning, and efficiency-focused decoding strategies. Additionally, we explore various application scenarios, such as solving complex problems and enhancing the reliability of LLM outputs.Finally, we highlight the core challenges of parallel reasoning and suggest potential directions for future research. We hope that our work can provide a useful roadmap for beginners and encourage more research on improving parallel reasoning methods. Related source can be avaliable in https://github.com/PPPP-kaqiu/Awesome-Parallel-Reasoning.

cs.CL↗

arXiv2025

MatryoshkaThinking: Recursive Test-Time Scaling Enables Efficient Reasoning

Test-time scaling has emerged as a promising paradigm in language modeling, wherein additional computational resources are allocated during inference to enhance model performance. Recent approaches, such as DeepConf, have demonstrated the efficacy of this strategy, however, they often incur substantial computational overhead to achieve competitive results. In this work, we propose MatryoshkaThinking, a novel method that significantly reduces computational cost while maintaining state-of-the-art performance. Specifically, MatryoshkaThinking attains a score of 99.79 on AIME2025 using only 4% of the computation required by DeepConf. The core of our approach lies in the recursive exploitation of the model's intrinsic capabilities in reasoning, verification, and summarization, which collectively enhance the retention of correct solutions and reduce the disparity between Pass@k and Pass@1. Comprehensive evaluations across multiple open-source models and challenging multi-modal reasoning benchmarks validate the effectiveness and generality of our method. These findings offer new insights into the design of efficient and scalable test-time inference strategies for advanced language models.

cs.CL↗

arXiv2025

A Lightweight Large Language Model-Based Multi-Agent System for 2D Frame Structural Analysis

Large language models (LLMs) have recently been used to empower autonomous agents in engineering, significantly improving automation and efficiency in labor-intensive workflows. However, their potential remains underexplored in structural engineering, particularly for finite element modeling tasks requiring geometric modeling, complex reasoning, and domain knowledge. To bridge this gap, this paper develops a LLM-based multi-agent system to automate finite element modeling of 2D frames. The system decomposes structural analysis into subtasks, each managed by a specialized agent powered by the lightweight Llama-3.3 70B Instruct model. The workflow begins with a Problem Analysis Agent, which extracts geometry, boundary, and material parameters from the user input. Next, a Geometry Agent incrementally derives node coordinates and element connectivity by applying expert-defined rules. These structured outputs are converted into executable OpenSeesPy code by a Translation Agent and refined by a Model Validation Agent through consistency checks. Then, a Load Agent applies load conditions into the assembled structural model. Experimental evaluations on 20 benchmark problems demonstrate that the system achieves accuracy over 80% in most cases across 10 repeated trials, outperforming Gemini-2.5 Pro and ChatGPT-4o models.

cs.CL↗

arXiv2025

MMRQA: Signal-Enhanced Multimodal Large Language Models for MRI Quality Assessment

Magnetic resonance imaging (MRI) quality assessment is crucial for clinical decision-making, yet remains challenging due to data scarcity and protocol variability. Traditional approaches face fundamental trade-offs: signal-based methods like MRIQC provide quantitative metrics but lack semantic understanding, while deep learning approaches achieve high accuracy but sacrifice interpretability. To address these limitations, we introduce the Multimodal MRI Quality Assessment (MMRQA) framework, pioneering the integration of multimodal large language models (MLLMs) with acquisition-aware signal processing. MMRQA combines three key innovations: robust metric extraction via MRQy augmented with simulated artifacts, structured transformation of metrics into question-answer pairs using Qwen, and parameter-efficient fusion through Low-Rank Adaptation (LoRA) of LLaVA-OneVision. Evaluated on MR-ART, FastMRI, and MyConnectome benchmarks, MMRQA achieves state-of-the-art performance with strong zero-shot generalization, as validated by comprehensive ablation studies. By bridging quantitative analysis with semantic reasoning, our framework generates clinically interpretable outputs that enhance quality control in dynamic medical settings.

cs.CV↗

arXiv2025

A2R: An Asymmetric Two-Stage Reasoning Framework for Parallel Reasoning

Recent Large Reasoning Models have achieved significant improvements in complex task-solving capabilities by allocating more computation at the inference stage with a "thinking longer" paradigm. Even as the foundational reasoning capabilities of models advance rapidly, the persistent gap between a model's performance in a single attempt and its latent potential, often revealed only across multiple solution paths, starkly highlights the disparity between its realized and inherent capabilities. To address this, we present A2R, an Asymmetric Two-Stage Reasoning framework designed to explicitly bridge the gap between a model's potential and its actual performance. In this framework, an "explorer" model first generates potential solutions in parallel through repeated sampling. Subsequently,a "synthesizer" model integrates these references for a more refined, second stage of reasoning. This two-stage process allows computation to be scaled orthogonally to existing sequential methods. Our work makes two key innovations: First, we present A2R as a plug-and-play parallel reasoning framework that explicitly enhances a model's capabilities on complex questions. For example, using our framework, the Qwen3-8B-distill model achieves a 75% performance improvement compared to its self-consistency baseline. Second, through a systematic analysis of the explorer and synthesizer roles, we identify an effective asymmetric scaling paradigm. This insight leads to A2R-Efficient, a "small-to-big" variant that combines a Qwen3-4B explorer with a Qwen3-8B synthesizer. This configuration surpasses the average performance of a monolithic Qwen3-32B model at a nearly 30% lower cost. Collectively, these results show that A2R is not only a performance-boosting framework but also an efficient and practical solution for real-world applications.

cs.AI↗

arXiv2025

Stoichiometry preservation and generalization of Bilger mixture fraction for non-premixed combustion with differential molecular diffusion

The Bilger mixture fraction is a widely used parameter in non-premixed combustion when considering differential molecular diffusion, a prevalent phenomenon in hydrogen or hydrogen-blended fuel combustion. The property of stoichiometry preservation of mixture fractions is investigated. Two different Bilger mixture fraction formulations are clarified. It is found that they belong to a class of one-parameter generalized mixture fraction definitions discovered in this work. Specific definitions from the class of mixture fractions are compared for hydrocarbon fuels. The comparison shows that the difference can be significant. An optimal mixture fraction definition is sought from the general definitions by minimizing its deviation from the desired properties. The obtained optimal mixture fractions show overall better preservation of stoichiometry than Bilger's definitions. The extension of the generalized mixture fraction to other fuels that contain nitrogen (like ammonia $\mathrm{NH_3}$) or sulfur (like hydrogen sulfide $\mathrm{H_2S}$) is also demonstrated.

physics.flu-dyn↗

arXiv2025

An U-Net-Based Deep Neural Network for Cloud Shadow and Sun-Glint Correction of Unmanned Aerial System (UAS) Imagery

The use of unmanned aerial systems (UASs) has increased tremendously in the current decade. They have significantly advanced remote sensing with the capability to deploy and image the terrain as per required spatial, spectral, temporal, and radiometric resolutions for various remote sensing applications. One of the major advantages of UAS imagery is that images can be acquired in cloudy conditions by flying the UAS under the clouds. The limitation to the technology is that the imagery is often sullied by cloud shadows. Images taken over water are additionally affected by sun glint. These are two pose serious issues for estimating water quality parameters from the UAS images. This study proposes a novel machine learning approach first to identify and extract regions with cloud shadows and sun glint and separate such regions from non-obstructed clear sky regions and sun-glint unaffected regions. The data was extracted from the images at pixel level to train an U-Net based deep learning model and best settings for model training was identified based on the various evaluation metrics from test cases. Using this evaluation, a high-quality image correction model was determined, which was used to recover the cloud shadow and sun glint areas in the images.

cs.CV↗