arXiv subjects

Chen Wang

Publications and source records attributed to Chen Wang.

At least 271 records · Page 15Linked to original sources

arXiv2025

Video SimpleQA: Towards Factuality Evaluation in Large Video Language Models

Recent advancements in Large Video Language Models (LVLMs) have highlighted their potential for multi-modal understanding, yet evaluating their factual grounding in videos remains a critical unsolved challenge. To address this gap, we introduce Video SimpleQA, the first comprehensive benchmark tailored for factuality evaluation in video contexts. Our work differs from existing video benchmarks through the following key features: 1) Knowledge required: demanding integration of external knowledge beyond the video's explicit narrative; 2) Multi-hop fact-seeking question: Each question involves multiple explicit facts and requires strict factual grounding without hypothetical or subjective inferences. We also include per-hop single-fact-based sub-QAs alongside final QAs to enable fine-grained, stepby-step evaluation; 3) Short-form definitive answer: Answers are crafted as unambiguous and definitively correct in a short format with minimal scoring variance; 4) Temporal grounded required: Requiring answers to rely on one or more temporal segments in videos, rather than single frames. We extensively evaluate 33 state-of-the-art LVLMs and summarize key findings as follows: 1) Current LVLMs exhibit notable deficiencies in factual adherence, with the best-performing model o3 merely achieving an F-score of 66.3%; 2) Most LVLMs are overconfident in what they generate, with self-stated confidence exceeding actual accuracy; 3) Retrieval-augmented generation demonstrates consistent improvements at the cost of additional inference time overhead; 4) Multi-hop QA demonstrates substantially degraded performance compared to single-hop sub-QAs, with first-hop object or event recognition emerging as the primary bottleneck. We position Video SimpleQA as the cornerstone benchmark for video factuality assessment, aiming to steer LVLM development toward verifiable grounding in real-world contexts.

cs.CV↗

arXiv2025

BridgeDepth: Bridging Monocular and Stereo Reasoning with Latent Alignment

Monocular and stereo depth estimation offer complementary strengths: monocular methods capture rich contextual priors but lack geometric precision, while stereo approaches leverage epipolar geometry yet struggle with ambiguities such as reflective or textureless surfaces. Despite post-hoc synergies, these paradigms remain largely disjoint in practice. We introduce a unified framework that bridges both through iterative bidirectional alignment of their latent representations. At its core, a novel cross-attentive alignment mechanism dynamically synchronizes monocular contextual cues with stereo hypothesis representations during stereo reasoning. This mutual alignment resolves stereo ambiguities (e.g., specular surfaces) by injecting monocular structure priors while refining monocular depth with stereo geometry within a single network. Extensive experiments demonstrate state-of-the-art results: \textbf{it reduces zero-shot generalization error by $\!>\!40\%$ on Middlebury and ETH3D}, while addressing longstanding failures on transparent and reflective surfaces. By harmonizing multi-view geometry with monocular context, our approach enables robust 3D perception that transcends modality-specific limitations. Codes available at https://github.com/aeolusguan/BridgeDepth.

cs.CV↗

arXiv2025

CRAFTS for HI cosmology: I. data processing pipeline and validation tests

We present the calibration procedures and validation of source measurement with the data of the Commensal Radio Astronomy FAST Survey (CRAFTS) for \HI intensity mapping by the Five-hundred-meter Aperture Spherical Radio Telescope (FAST). Using 70-hour drift-scan observation with the L-band (1.05-1.45GHz) 19-beam receiver, we obtain the data covering $270\,\rm deg^2$ sky area. We employ both the pulsar backend and the spectrum backend to calibrate the spectral time-ordered-data (TOD) before projecting them onto HEALPix maps. We produce calibrated TOD with frequency resolution of 30kHz and time resolution of 1s and the map data-cube with frequency resolution of 30kHz and spatial resolution of $2.95\,\rm arcmin^2$. We examine the pointing errors, noise overflow, RFI contamination and their effect on the data quality. The resulting noise level is $\sim$ 5.7mJy for the calibrated TOD and 1.6mJy for the map, consistent with the theoretical predictions within 5\% at RFI-free channels. We also validate the data by Principal Components Analysis (PCA) and find the residual map looks thermal noise dominated after removing 30 modes. We identify 447 isolated bright continuum sources in our data matching the NRAO-VLA Sky Survey (NVSS) catalog, with relative flux error of 8.3\% for TOD and 6.6\% for the map-level. We also measure the \HI emission of 90 galaxies with redshift $z<0.07$ and compare with \HI-MaNGA spectra, yielding an overall relative \HI integral flux error of 16.7\%. These results provide an important first step in assessing the feasibility of conducting cosmological \HI detection with CRAFTS.

astro-ph.CO↗

arXiv2025

Towards Pareto Optimal Throughput in Small Language Model Serving

Large language models (LLMs) have revolutionized the state-of-the-art of many different natural language processing tasks. Although serving LLMs is computationally and memory demanding, the rise of Small Language Models (SLMs) offers new opportunities for resource-constrained users, who now are able to serve small models with cutting-edge performance. In this paper, we present a set of experiments designed to benchmark SLM inference at performance and energy levels. Our analysis provides a new perspective in serving, highlighting that the small memory footprint of SLMs allows for reaching the Pareto-optimal throughput within the resource capacity of a single accelerator. In this regard, we present an initial set of findings demonstrating how model replication can effectively improve resource utilization for serving SLMs.

cs.CL↗

arXiv2025

Formula-Guided Machine Learning for Ground Vibration Propagation and Attenuation Modeling

Understanding the propagation and attenuation patterns of ground vibrations is critical for evaluating the impact of environmental disturbances on large-scale scientific facilities. However, complex site conditions often result in intricate vibration behaviors, limiting the accuracy of traditional predictive methods. This study proposes a hybrid iterative fitting method that integrates machine learning with the Bornitz formula through an intelligent formula generation model. The method enables the automatic derivation of high-precision, interpretable ground vibration attenuation formulas from experimental data. A case study was conducted at the High Energy Photon Source in Beijing, where field tests were performed to collect vibration data. Using the proposed approach, an attenuation formula describing ground vibration propagation was derived. The physical validity of the model was further verified via finite element simulations. A probabilistic analysis was then employed to estimate computational errors. Comparative evaluations with black-box machine learning models and empirical formulas from previous studies demonstrate that the proposed method offers significant advantages in both interpretability and accuracy. These findings provide a valuable framework for vibration impact assessment and mitigation in other large-scale scientific infrastructure projects.

physics.app-ph↗

arXiv2025

Testing High-dimensional Nonstationary Time Series

In this article, we first establish the joint central limit theorem (CLT) for the extreme eigenvalues of the sample correlation matrix of high-dimensional random walks with cross-sectional dependence. We further investigate the asymptotic spectral properties of the sample correlation matrix of high-dimensional autoregressive processes. To apply our theoretical results, we propose a novel high-dimensional unit root test and develop a forward sequential test to determine the number of unit roots in high-dimensional time series data. Finally, we conduct an empirical study of the purchasing power parity (PPP) hypothesis in high-dimensional settings.

stat.ME↗

arXiv2025

AI paradigm for solving differential equations: first-principles data generation and scale-dilation operator AI solver

Many problems are governed by differential equations (DEs). Artificial intelligence (AI) is a new path for solving DEs. However, data is very scarce and existing AI solvers struggle with approximation of high frequency components (AHFC). We propose an AI paradigm for solving diverse DEs, including DE-ruled first-principles data generation methodology and scale-dilation operator (SDO) AI solver. Using either prior knowledge or random fields, we generate solutions and then substitute them into the DEs to derive the sources and initial/boundary conditions through balancing DEs, thus producing arbitrarily vast amount of, first-principles-consistent training datasets at extremely low computational cost. We introduce a reversible SDO that leverages the Fourier transform of the multiscale solutions to fix AHFC, and design a spatiotemporally coupled, attention-based Transformer AI solver of DEs with SDO. An upper bound on the Hessian condition number of the loss function is proven to be proportional to the squared 2-norm of the solution gradient, revealing that SDO yields a smoother loss landscape, consequently fixing AHFC with efficient training. Extensive tests on diverse DEs demonstrate that our AI paradigm achieves consistently superior accuracy over state-of-the-art methods. This work makes AI solver of DEs to be truly usable in broad nature and engineering fields.

cs.LG↗

arXiv2025

Latent Swap Joint Diffusion for 2D Long-Form Latent Generation

This paper introduces Swap Forward (SaFa), a modality-agnostic and efficient method to generate seamless and coherence long spectrum and panorama through latent swap joint diffusion across multi-views. We first investigate the spectrum aliasing problem in spectrum-based audio generation caused by existing joint diffusion methods. Through a comparative analysis of the VAE latent representation of Mel-spectra and RGB images, we identify that the failure arises from excessive suppression of high-frequency components during the spectrum denoising process due to the averaging operator. To address this issue, we propose Self-Loop Latent Swap, a frame-level bidirectional swap applied to the overlapping region of adjacent views. Leveraging stepwise differentiated trajectories of adjacent subviews, this swap operator adaptively enhances high-frequency components and avoid spectrum distortion. Furthermore, to improve global cross-view consistency in non-overlapping regions, we introduce Reference-Guided Latent Swap, a unidirectional latent swap operator that provides a centralized reference trajectory to synchronize subview diffusions. By refining swap timing and intervals, we can achieve a cross-view similarity-diversity balance in a forward-only manner. Quantitative and qualitative experiments demonstrate that SaFa significantly outperforms existing joint diffusion methods and even training-based methods in audio generation using both U-Net and DiT models, along with effective longer length adaptation. It also adapts well to panorama generation, achieving comparable performance with 2 $\sim$ 20 $\times$ faster speed and greater model generalizability. More generation demos are available at https://swapforward.github.io/

cs.SD↗

arXiv2025

Advancing Multimodal Reasoning via Reinforcement Learning with Cold Start

Recent advancements in large language models (LLMs) have demonstrated impressive chain-of-thought reasoning capabilities, with reinforcement learning (RL) playing a crucial role in this progress. While "aha moment" patterns--where models exhibit self-correction through reflection--are often attributed to emergent properties from RL, we first demonstrate that these patterns exist in multimodal LLMs (MLLMs) prior to RL training but may not necessarily correlate with improved reasoning performance. Building on these insights, we present a comprehensive study on enhancing multimodal reasoning through a two-stage approach: (1) supervised fine-tuning (SFT) as a cold start with structured chain-of-thought reasoning patterns, followed by (2) reinforcement learning via GRPO to further refine these capabilities. Our extensive experiments show that this combined approach consistently outperforms both SFT-only and RL-only methods across challenging multimodal reasoning benchmarks. The resulting models achieve state-of-the-art performance among open-source MLLMs at both 3B and 7B scales, with our 7B model showing substantial improvements over base models (e.g., 66.3 %$\rightarrow$73.4 % on MathVista, 62.9 %$\rightarrow$70.4 % on We-Math) and our 3B model achieving performance competitive with several 7B models. Overall, this work provides practical guidance for building advanced multimodal reasoning models. Our code is available at https://github.com/waltonfuture/RL-with-Cold-Start.

cs.CL↗

arXiv2025

Voice-based AI Agents: Filling the Economic Gaps in Digital Health Delivery

The integration of voice-based AI agents in healthcare presents a transformative opportunity to bridge economic and accessibility gaps in digital health delivery. This paper explores the role of large language model (LLM)-powered voice assistants in enhancing preventive care and continuous patient monitoring, particularly in underserved populations. Drawing insights from the development and pilot study of Agent PULSE (Patient Understanding and Liaison Support Engine) -- a collaborative initiative between IBM Research, Cleveland Clinic Foundation, and Morehouse School of Medicine -- we present an economic model demonstrating how AI agents can provide cost-effective healthcare services where human intervention is economically unfeasible. Our pilot study with 33 inflammatory bowel disease patients revealed that 70\% expressed acceptance of AI-driven monitoring, with 37\% preferring it over traditional modalities. Technical challenges, including real-time conversational AI processing, integration with healthcare systems, and privacy compliance, are analyzed alongside policy considerations surrounding regulation, bias mitigation, and patient autonomy. Our findings suggest that AI-driven voice agents not only enhance healthcare scalability and efficiency but also improve patient engagement and accessibility. For healthcare executives, our cost-utility analysis demonstrates huge potential savings for routine monitoring tasks, while technologists can leverage our framework to prioritize improvements yielding the highest patient impact. By addressing current limitations and aligning AI development with ethical and regulatory frameworks, voice-based AI agents can serve as a critical entry point for equitable, sustainable digital healthcare solutions.

cs.AI↗

arXiv2025

MapEx: Indoor Structure Exploration with Probabilistic Information Gain from Global Map Predictions

Exploration is a critical challenge in robotics, centered on understanding unknown environments. In this work, we focus on robots exploring structured indoor environments which are often predictable and composed of repeating patterns. Most existing approaches, such as conventional frontier approaches, have difficulty leveraging the predictability and explore with simple heuristics such as `closest first'. Recent works use deep learning techniques to predict unknown regions of the map, using these predictions for information gain calculation. However, these approaches are often sensitive to the predicted map quality or do not reason over sensor coverage. To overcome these issues, our key insight is to jointly reason over what the robot can observe and its uncertainty to calculate probabilistic information gain. We introduce MapEx, a new exploration framework that uses predicted maps to form probabilistic sensor model for information gain estimation. MapEx generates multiple predicted maps based on observed information, and takes into consideration both the computed variances of predicted maps and estimated visible area to estimate the information gain of a given viewpoint. Experiments on the real-world KTH dataset showed on average 12.4% improvement than representative map-prediction based exploration and 25.4% improvement than nearest frontier approach. Website: mapex-explorer.github.io

cs.RO↗

arXiv2025

A Framework to Pinpoint Bottlenecks in Emerging Solar Cells and Disordered Devices via Differential Machine Learning

A key challenge in the development of materials for the next generation of solar cells, sensors and transistors is linking macroscopic device performance to underlying microscopic properties. For years, fabrication of devices has been faster than our ability to characterize them. This has led to a random walk of material development, with new materials being proposed faster than our understanding. We present two neural network-based methods for extracting key material parameters, including charge carrier mobility and trap state density, in optoelectronic devices such as solar cells. Our methods require solely measured light current--voltage curve and modest computational resources, making our approach applicable in even minimally equipped laboratories. Unlike traditional machine learning models, our methods place the final material values in a non-Gaussian likelihood distribution, allowing confidence assessment of each predicted parameter. We demonstrate these techniques using fresh PM6:Y12 and degraded PM6:BTP-eC9 organic solar cells.

physics.app-ph↗

arXiv2025

Mind the Memory Gap: Unveiling GPU Bottlenecks in Large-Batch LLM Inference

Large language models have been widely adopted across different tasks, but their auto-regressive generation nature often leads to inefficient resource utilization during inference. While batching is commonly used to increase throughput, performance gains plateau beyond a certain batch size, especially with smaller models, a phenomenon that existing literature typically explains as a shift to the compute-bound regime. In this paper, through an in-depth GPU-level analysis, we reveal that large-batch inference remains memory-bound, with most GPU compute capabilities underutilized due to DRAM bandwidth saturation as the primary bottleneck. To address this, we propose a Batching Configuration Advisor (BCA) that optimizes memory allocation, reducing GPU memory requirements with minimal impact on throughput. The freed memory and underutilized GPU compute capabilities can then be leveraged by concurrent workloads. Specifically, we use model replication to improve serving throughput and GPU utilization. Our findings challenge conventional assumptions about LLM inference, offering new insights and practical strategies for improving resource utilization, particularly for smaller language models. The code is publicly available at https://github.com/FerranAgulloLopez/vLLMBatchingMemoryGap.

cs.DC↗

arXiv2025

Modeling Uncertainties on the Z Boson Background in the Context of High Precision W Boson Mass Measurements

The precise determination of the $W$ boson mass, $m_W$, is a cornerstone of electroweak precision tests and a critical input to global fits of the Standard Model (SM). While most existing measurements of $m_W$ are in agreement with SM predictions, the recent result from the CDF collaboration exhibits a significant deviation from the SM prediction but also exceeding $4σ$ from the global average of all other high precision measurements. This prompts renewed scrutiny of potential sources of systematic bias. In this study, we investigate the modeling of the $Z$ boson background, which is particularly relevant in the muon decay channel of the $W$ boson, where undetected muons can mimic missing transverse momentum. Using updated theoretical predictions and high-precision simulations, we construct reweighted templates to evaluate the impact of possible mismodeling in the $Z$ background. Our analysis suggests that such effects can induce a shift in the extracted $m_W$ value of up to 8~MeV in the muon decay channel. While this shift is non-negligible and exceeds the originally quoted uncertainty from this source, it remains insufficient to explain the full discrepancy observed in the CDF measurement.

hep-ex↗

arXiv2025

Enhanced Influence-aware Group Recommendation for Online Media Propagation

Group recommendation over social media streams has attracted significant attention due to its wide applications in domains such as e-commerce, entertainment, and online news broadcasting. By leveraging social connections and group behaviours, group recommendation (GR) aims to provide more accurate and engaging content to a set of users rather than individuals. Recently, influence-aware GR has emerged as a promising direction, as it considers the impact of social influence on group decision-making. In earlier work, we proposed Influence-aware Group Recommendation (IGR) to solve this task. However, this task remains challenging due to three key factors: the large and ever-growing scale of social graphs, the inherently dynamic nature of influence propagation within user groups, and the high computational overhead of real-time group-item matching. To tackle these issues, we propose an Enhanced Influence-aware Group Recommendation (EIGR) framework. First, we introduce a Graph Extraction-based Sampling (GES) strategy to minimise redundancy across multiple temporal social graphs and effectively capture the evolving dynamics of both groups and items. Second, we design a novel DYnamic Independent Cascade (DYIC) model to predict how influence propagates over time across social items and user groups. Finally, we develop a two-level hash-based User Group Index (UG-Index) to efficiently organise user groups and enable real-time recommendation generation. Extensive experiments on real-world datasets demonstrate that our proposed framework, EIGR, consistently outperforms state-of-the-art baselines in both effectiveness and efficiency.

cs.IR↗

arXiv2025

Vision Technologies with Applications in Traffic Surveillance Systems: A Holistic Survey

Traffic Surveillance Systems (TSS) have become increasingly crucial in modern intelligent transportation systems, with vision technologies playing a central role for scene perception and understanding. While existing surveys typically focus on isolated aspects of TSS, a comprehensive analytical framework bridging low-level and high-level perception tasks, particularly considering emerging technologies, remains lacking. This paper presents a systematic review of vision technologies in TSS, examining both low-level perception tasks (object detection, classification, and tracking) and high-level perception tasks (parameter estimation, anomaly detection, and behavior understanding). Specifically, we first provide a detailed methodological categorization and comprehensive performance evaluation for each task. Our investigation reveals five fundamental limitations in current TSS: perceptual data degradation in complex scenarios, data-driven learning constraints, semantic understanding gaps, sensing coverage limitations and computational resource demands. To address these challenges, we systematically analyze five categories of current approaches and potential trends: advanced perception enhancement, efficient learning paradigms, knowledge-enhanced understanding, cooperative sensing frameworks and efficient computing frameworks, critically assessing their real-world applicability. Furthermore, we evaluate the transformative potential of foundation models in TSS, which exhibit remarkable zero-shot learning abilities, strong generalization, and sophisticated reasoning capabilities across diverse tasks. This review provides a unified analytical framework bridging low-level and high-level perception tasks, systematically analyzes current limitations and solutions, and presents a structured roadmap for integrating emerging technologies, particularly foundation models, to enhance TSS capabilities.

cs.CV↗

arXiv2025

AirLine: Efficient Learnable Line Detection with Local Edge Voting

Line detection is widely used in many robotic tasks such as scene recognition, 3D reconstruction, and simultaneous localization and mapping (SLAM). Compared to points, lines can provide both low-level and high-level geometrical information for downstream tasks. In this paper, we propose a novel learnable edge-based line detection algorithm, AirLine, which can be applied to various tasks. In contrast to existing learnable endpoint-based methods, which are sensitive to the geometrical condition of environments, AirLine can extract line segments directly from edges, resulting in a better generalization ability for unseen environments. To balance efficiency and accuracy, we introduce a region-grow algorithm and a local edge voting scheme for line parameterization. To the best of our knowledge, AirLine is one of the first learnable edge-based line detection methods. Our extensive experiments have shown that it retains state-of-the-art-level precision, yet with a 3 to 80 times runtime acceleration compared to other learning-based methods, which is critical for low-power robots.

cs.RO↗

arXiv2025

No Free Lunch: Rethinking Internal Feedback for LLM Reasoning

Reinforcement learning has emerged as a powerful paradigm for post-training large language models (LLMs) to improve reasoning. Approaches like Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning with Verifiable Rewards (RLVR) have shown strong results, but they require extensive external supervision. We investigate an alternative class of methods, Reinforcement Learning from Internal Feedback (RLIF), which relies solely on intrinsic model-derived signals instead of external rewards. In particular, we leverage unsupervised reward proxies such as token-level entropy, trajectory-level entropy, and self-certainty. Our theoretical analysis shows these internal objectives are partially equivalent, and we empirically evaluate various RLIF strategies on challenging math reasoning benchmarks. Experimental results demonstrate that RLIF can boost the reasoning performance of base LLMs at the beginning phase of the training, matching or surpassing RLVR techniques on these tasks. However, when training progresses, performance degrades even below the model before training. Moreover, we find that RLIF yields little improvement for instruction-tuned models, indicating diminishing returns of intrinsic feedback once an LLM is already instruction-tuned. We further analyze this limitation by mixing model weights and explain the reason of RLIF's training behaviors, providing practical guidelines for integrating internal feedback signals into LLM training. We hope our analysis of internal feedback will inform more principled and effective strategies for LLM post-training.

cs.LG↗