arXiv subjects

Wei Li

Publications and source records attributed to Wei Li.

At least 325 records · Page 18Linked to original sources

arXiv2025

TIME: A Multi-level Benchmark for Temporal Reasoning of LLMs in Real-World Scenarios

Temporal reasoning is pivotal for Large Language Models (LLMs) to comprehend the real world. However, existing works neglect the real-world challenges for temporal reasoning: (1) intensive temporal information, (2) fast-changing event dynamics, and (3) complex temporal dependencies in social interactions. To bridge this gap, we propose a multi-level benchmark TIME, designed for temporal reasoning in real-world scenarios. TIME consists of 38,522 QA pairs, covering 3 levels with 11 fine-grained sub-tasks. This benchmark encompasses 3 sub-datasets reflecting different real-world challenges: TIME-Wiki, TIME-News, and TIME-Dial. We conduct extensive experiments on reasoning models and non-reasoning models. And we conducted an in-depth analysis of temporal reasoning performance across diverse real-world scenarios and tasks, and summarized the impact of test-time scaling on temporal reasoning capabilities. Additionally, we release TIME-Lite, a human-annotated subset to foster future research and standardized evaluation in temporal reasoning. The code is available at https://github.com/sylvain-wei/TIME , the dataset is available at https://huggingface.co/datasets/SylvainWei/TIME , and the project page link is https://sylvain-wei.github.io/TIME/ .

cs.AI↗

arXiv2025

Model Predictive Path Integral Control for Roll-to-Roll Manufacturing

Roll-to-roll (R2R) manufacturing is a continuous processing technology essential for scalable production of thin-film materials and printed electronics, but precise control remains challenging due to subsystem interactions, nonlinearities, and process disturbances. This paper proposes a Model Predictive Path Integral (MPPI) control formulation for R2R systems, leveraging a GPU-based Monte-Carlo sampling approach to efficiently approximate optimal controls online. Crucially, MPPI easily handles non-differentiable cost functions, enabling the incorporation of complex performance criteria relevant to advanced manufacturing processes. A case study is presented that demonstrates that MPPI significantly improves tension regulation performance compared to conventional model predictive control (MPC), highlighting its suitability for real-time control in advanced manufacturing.

eess.SY↗

arXiv2025

Compact Multi-level-prior Tensor Representation for Hyperspectral Image Super-resolution

Fusing a hyperspectral image with a multispectral image acquired over the same scene, \textit{i.e.}, hyperspectral image super-resolution, has become a popular computational way to access the latent high-spatial-spectral-resolution image. To date, a variety of fusion methods have been proposed, among which the tensor-based ones have testified that multiple priors, such as multidimensional low-rankness and spatial total variation at multiple levels, effectively drive the fusion process. However, existing tensor-based models can only effectively leverage one or two priors at one or two levels, since simultaneously incorporating multi-level priors inevitably increases model complexity. This introduces challenges in both balancing the weights of different priors and optimizing multi-block structures. Concerning this, we present a novel hyperspectral super-resolution model compactly characterizing these multi-level priors of hyperspectral images within the tensor framework. Firstly, the proposed model decouples the spectral low-rankness and spatial priors by casting the latent high-spatial-spectral-resolution image into spectral subspace and spatial maps via block term decomposition. Secondly, these spatial maps are stacked as the spatial tensor encoding the high-order spatial low-rankness and smoothness priors, which are co-modeled via the proposed non-convex mode-shuffled tensor correlated total variation. Finally, we draw inspiration from the linearized alternating direction method of multipliers to design an efficient algorithm to optimize the resulting model, theoretically proving its Karush-Kuhn-Tucker convergence under mild conditions. Experiments on multiple datasets demonstrate the effectiveness of the proposed algorithm. The code implementation will be available from https://github.com/WongYinJ.

cs.CV↗

arXiv2025

Spin Seebeck Effect of Triangular-lattice Spin Supersolid

Using thermal tensor-network approach, we investigate the spin Seebeck effect (SSE) of the triangular-lattice quantum antiferromagnet hosting spin supersolid phase. We focus on the low-temperature scaling behaviors of the normalized spin current across the interface. For the 1D Heisenberg chain, we find a negative spinon spin in the bulk current with algebraic temperature scaling; at low fields, boundary effects induce a second sign reversal at lower temperatures. These benchmark results are consistent with field-theoretical analysis. On the triangular lattice, spin frustration dramatically enhances the low-temperature SSE, with distinct spin-current signatures -- particularly the sign reversal and characteristic temperature dependence -- distinguishing different spin states. Remarkably, we discover a persistent, negative spin current in the spin supersolid phase, which saturates to a non-zero value in the low-temperature limit and can be ascribed to the Goldstone-mode-mediated spin supercurrents. Moreover, a universal scaling $T^{d/z}$ is found at the U(1)-symmetric polarization quantum critical points. These distinct quantum spin transport traits provide sensitive spin current probes for spin supersolid states in quantum magnets such as Na$_2$BaCo(PO$_4$)$_2$. Furthermore, our results also establish spin supersolids as a tunable quantum platform for spin caloritronics in the ultralow-temperature regime.

cond-mat.str-el↗

arXiv2025

Assumption-lean Inference for Network-linked Data

We consider statistical inference for network-linked regression problems, where covariates may include network summary statistics computed for each node. In settings involving network data, it is often natural to posit that latent variables govern connection probabilities in the graph. Since the presence of these latent features makes classical regression assumptions even less tenable, we propose an assumption-lean framework for linear regression with jointly exchangeable regression arrays. We establish an analog of the Aldous-Hoover representation for such arrays, which may be of independent interest. Moreover, we consider two different projection parameters as potential targets and establish conditions under which asymptotic normality and bootstrap consistency hold when commonly used network statistics, including local subgraph frequencies and spectral embeddings, are used as covariates. In the case of linear regression with local count statistics, we show that a bias-corrected estimator allows one to target a more natural inferential target under weaker sparsity conditions compared to the OLS estimator. Our inferential tools are illustrated using both simulated data and real data related to the academic climate of elementary schools.

stat.ME↗

arXiv2025

MinerU2.5: A Decoupled Vision-Language Model for Efficient High-Resolution Document Parsing

We introduce MinerU2.5, a 1.2B-parameter document parsing vision-language model that achieves state-of-the-art recognition accuracy while maintaining exceptional computational efficiency. Our approach employs a coarse-to-fine, two-stage parsing strategy that decouples global layout analysis from local content recognition. In the first stage, the model performs efficient layout analysis on downsampled images to identify structural elements, circumventing the computational overhead of processing high-resolution inputs. In the second stage, guided by the global layout, it performs targeted content recognition on native-resolution crops extracted from the original image, preserving fine-grained details in dense text, complex formulas, and tables. To support this strategy, we developed a comprehensive data engine that generates diverse, large-scale training corpora for both pretraining and fine-tuning. Ultimately, MinerU2.5 demonstrates strong document parsing ability, achieving state-of-the-art performance on multiple benchmarks, surpassing both general-purpose and domain-specific models across various recognition tasks, while maintaining significantly lower computational overhead.

cs.CV↗

arXiv2025

NeoWorld: Neural Simulation of Explorable Virtual Worlds via Progressive 3D Unfolding

We introduce NeoWorld, a deep learning framework for generating interactive 3D virtual worlds from a single input image. Inspired by the on-demand worldbuilding concept in the science fiction novel Simulacron-3 (1964), our system constructs expansive environments where only the regions actively explored by the user are rendered with high visual realism through object-centric 3D representations. Unlike previous approaches that rely on global world generation or 2D hallucination, NeoWorld models key foreground objects in full 3D, while synthesizing backgrounds and non-interacted regions in 2D to ensure efficiency. This hybrid scene structure, implemented with cutting-edge representation learning and object-to-3D techniques, enables flexible viewpoint manipulation and physically plausible scene animation, allowing users to control object appearance and dynamics using natural language commands. As users interact with the environment, the virtual world progressively unfolds with increasing 3D detail, delivering a dynamic, immersive, and visually coherent exploration experience. NeoWorld significantly outperforms existing 2D and depth-layered 2.5D methods on the WorldScore benchmark.

cs.CV↗

arXiv2025

Audio-centric Video Understanding Benchmark without Text Shortcut

Audio often serves as an auxiliary modality in video understanding tasks of audio-visual large language models (LLMs), merely assisting in the comprehension of visual information. However, a thorough understanding of videos significantly depends on auditory information, as audio offers critical context, emotional cues, and semantic meaning that visual data alone often lacks. This paper proposes an audio-centric video understanding benchmark (AVUT) to evaluate the video comprehension capabilities of multimodal LLMs with a particular focus on auditory information. AVUT introduces a suite of carefully designed audio-centric tasks, holistically testing the understanding of both audio content and audio-visual interactions in videos. Moreover, this work points out the text shortcut problem that largely exists in other benchmarks where the correct answer can be found from question text alone without needing videos. AVUT addresses this problem by proposing a answer permutation-based filtering mechanism. A thorough evaluation across a diverse range of open-source and proprietary multimodal LLMs is performed, followed by the analyses of deficiencies in audio-visual LLMs. Demos and data are available at https://github.com/lark-png/AVUT.

cs.CV↗

arXiv2025

ABC-Eval: Benchmarking Large Language Models on Symbolic Music Understanding and Instruction Following

As large language models continue to develop, the feasibility and significance of text-based symbolic music tasks have become increasingly prominent. While symbolic music has been widely used in generation tasks, LLM capabilities in understanding and reasoning about symbolic music remain largely underexplored. To address this gap, we propose ABC-Eval, the first open-source benchmark dedicated to the understanding and instruction-following capabilities in text-based ABC notation scores. It comprises 1,086 test samples spanning 10 sub-tasks, covering scenarios from basic musical syntax comprehension to complex sequence-level reasoning. Such a diverse scope poses substantial challenges to models' ability to handle symbolic music tasks. We evaluated seven state-of-the-art LLMs on ABC-Eval, and the results reveal notable limitations in existing models' symbolic music processing capabilities. Furthermore, the consistent performance of individual baselines across different sub-tasks supports the reliability of our benchmark.

cs.SD↗

arXiv2025

Voyager: Real-Time Splatting City-Scale Gaussians on Resource-Constrained Devices

3D Gaussian splatting (3DGS) is an emerging technique for photorealistic 3D scene rendering. However, rendering city-scale 3DGS scenes on resource-constrained mobile devices in real-time remains a significant challenge due to two compute-intensive stages: level-of-detail (LoD) search and rasterization. In this paper, we propose Voyager, an effective solution to accelerate city-scale 3DGS rendering on mobile devices. Our key insight is that, under normal user motion, the number of newly visible Gaussians within the view frustum remains roughly constant. Leveraging this temporal correlation, we propose a temporal-aware LoD search to identify the necessary Gaussians for the remaining rendering stages. For the remaining rendering process, we accelerate the bottleneck stage, rasterization, via preemptive $α$-filtering. With all optimizations above, our system can deliver low-latency, city-scale 3DGS rendering on mobile devices. Compared to existing solutions, Voyager achieves up to 6.6$\times$ speedup and 85\% energy savings with superior rendering quality.

cs.GR↗

arXiv2025

video-SALMONN 2: Caption-Enhanced Audio-Visual Large Language Models

We present video-SALMONN 2, a family of audio-visual large language models that set new state-of-the-art (SOTA) results in video description and question answering (QA). Our core contribution is multi-round direct preference optimisation (MrDPO), paired with a caption-quality objective that jointly rewards completeness and factual accuracy. Unlike standard DPO with a fixed reference policy, MrDPO periodically refreshes the reference by bootstrapping from a newly re-initialised lightweight adapter trained on the latest preferences, avoiding reference staleness and enabling continual improvement. This strategy produces captions that are consistently more detailed and accurate than those from proprietary systems such as GPT-4o and Gemini-1.5 Pro. We further distil these gains by using our model to generate a high-quality video-caption corpus for supervised fine-tuning of new models, transferring benefits beyond captioning to strong performance on complex video-QA tasks. Across widely used audio-visual and visual-only understanding benchmarks (including Video-MME, WorldSense, AVUT, Video-Holmes, DailyOmni, MLVU, and LVBench), our 3B and 7B models achieve SOTA results at comparable scales, while the 72B model surpasses all other open-source systems. Our source code, models, and data are released at \href{https://github.com/bytedance/video-SALMONN-2}{https://github.com/bytedance/video-SALMONN-2}.

cs.CV↗

arXiv2025

Semantic Reformulation Entropy for Robust Hallucination Detection in QA Tasks

Reliable question answering with large language models (LLMs) is challenged by hallucinations, fluent but factually incorrect outputs arising from epistemic uncertainty. Existing entropy-based semantic-level uncertainty estimation methods are limited by sampling noise and unstable clustering of variable-length answers. We propose Semantic Reformulation Entropy (SRE), which improves uncertainty estimation in two ways. First, input-side semantic reformulations produce faithful paraphrases, expand the estimation space, and reduce biases from superficial decoder tendencies. Second, progressive, energy-based hybrid clustering stabilizes semantic grouping. Experiments on SQuAD and TriviaQA show that SRE outperforms strong baselines, providing more robust and generalizable hallucination detection. These results demonstrate that combining input diversification with multi-signal clustering substantially enhances semantic-level uncertainty estimation.

cs.CL↗

arXiv2025

Subdiffraction confinement and non-diffractive propagation of optical Stokes skyrmions enabled by a super-oscillatory metalens

Optical Stokes skyrmions have garnered extensive interest due to their intrinsic topological robustness and potential in informatics.However, most research remains confined to paraxial, low-numerical-aperture (low-NA) regimes, where their large transverse dimensions restrict broader applications.Under high-NA focusing, the polarization texture typically degrades or transforms abruptly as the beam traverses the focal region, hindering topology-preserving transport.In this work, we propose a strategy to generate a skyrmion needle field that maintains both subdiffraction confinement and non-diffractive propagation under high-NA conditions, thus preserving their topological characteristsics. Leveraging the polarization invariance of conventional optical needles, we realize the Stokes skyrmion needle using a single plasmonic metalens,designed to function as both a polarization filter and a super-resolving focusing element.Experimental and simulation results verify non-diffractive propagation over an extended depth of focus (up to 5 lambda), while the Stokes-vector texture retained at subdiffraction scales throughout propagation. This skyrmion needle not only addresses previous propagation constraints but also opens new avenues for diffraction-unlimited information transport. Such skyrmion needles exhibit substantial potential in fields including light-matter interaction, optical metrology, and informatics.

physics.optics↗

arXiv2025

Gender and Agricultural Commercialization in Sub-Saharan Africa: Evidence from Three Panel Surveys

Agricultural commercialization is often promoted as a key driver of development in Sub-Saharan Africa, yet its benefits may not extend equally to all farmers. Using longitudinal household data from the LSMS-ISA and a two-way Mundlak fixed effects estimator, we examine the relationship between farmers' gender and agricultural commercialization in Ethiopia, Nigeria, and Tanzania. In Ethiopia and Nigeria, women-headed households and those with a higher share of women-managed land face substantial disadvantages in market engagement, particularly in households oriented towards self-consumption. Interestingly, in both countries, women-headed households that do engage in sales are more likely to sell to market buyers and less likely to sell to individual buyers compared to men-headed households. In contrast, in Tanzania, the negative associations between gender and commercialization are weaker and less robust across outcomes. Overall, these findings demonstrate that gender gaps in commercialization are highly context-specific rather than universal, highlighting the need for country-tailored policies that address the institutional and market constraints faced by women farmers.

econ.GN↗

arXiv2025

On the Suitability of Reinforcement Fine-Tuning to Visual Tasks

Reinforcement Fine-Tuning (RFT) is proved to be greatly valuable for enhancing the reasoning ability of LLMs. Researchers have been starting to apply RFT to MLLMs, hoping it will also enhance the capabilities of visual understanding. However, these works are at a very early stage and have not examined how suitable RFT actually is for visual tasks. In this work, we endeavor to understand the suitabilities and limitations of RFT for visual tasks, through experimental analysis and observations. We start by quantitative comparisons on various tasks, which shows RFT is generally better than SFT on visual tasks. %especially when the number of training samples are limited. To check whether such advantages are brought up by the reasoning process, we design a new reward that encourages the model to ``think'' more, whose results show more thinking can be beneficial for complicated tasks but harmful for simple tasks. We hope this study can provide more insight for the rapid advancements on this topic.

cs.CV↗

arXiv2025

Universal Scaling Functions of the Gr{ü}neisen Ratio near Quantum Critical Points

The Grüneisen ratio, defined as $Γ_g \equiv (1/T) (\partial T/\partial g)_S$, serves as a highly sensitive probe for detecting quantum critical points (QCPs) driven by an external feild $g$ and for characterizing the magnetocaloric effect (MCE). Near a QCP, the Grüneisen ratio displays a universal divergence which is governed by a universality-class-dependent scaling function stemming from the scale invariance. In this work, we systematically investigate the universal scaling functions of Grüneisen ratio in both one-dimensional (1D) and two-dimensional (2D) quantum spin systems, including the transverse-field Ising model, the spin-1/2 Heisenberg model, the quantum $q$-state Potts model ($q=3,4$) and the $J_1$-$J_2$ columnar dimer model. Our approach employs the thermal tensor-network method for infinite-size 1D systems and the stochastic series expansion quantum Monte Carlo (SSE QMC) simulations for 2D systems, enabling precise calculations of the Grüneisen ratio near QCPs. Through data collapse analysis, we extract the corresponding scaling functions, which establish quantitative frameworks to interpret magnetocaloric experiments and guide the development of ultralow-temperature refrigeration.

cond-mat.str-el↗

arXiv2025

Cloud-Edge Collaborative Data Anomaly Detection in Industrial Sensor Networks

Existing research on sensor data anomaly detection for industrial sensor networks still has several inherent limitations. First, most detection models usually consider centralized detection. Thus, all sensor data have to be uploaded to the control center for analysis, leading to a heavy traffic load. However, industrial sensor networks have high requirements for reliable and real-time communication. The heavy traffic load may cause communication delays or packets lost by corruption. Second, there are complex spatial and temporal features in industrial sensor data. The full extraction of such features plays a key role in improving detection performance.To solve the limitations above, this paper develops a cloud-edge collaborative data anomaly detection approach for industrial sensor networks that mainly consists of a sensor data detection model deployed at individual edges and a sensor data analysis model deployed in the cloud. The former is implemented using Gaussian and Bayesian algorithms, which effectively filter the substantial volume of sensor data generated during the normal operation of the industrial sensor network, thereby reducing traffic load. It only uploads all the sensor data to the sensor data analysis model for further analysis when the network is in an anomalous state. The latter based on GCRL is developed by inserting Long Short-Term Memory network (LSTM) into Graph Convolutional Network (GCN), which can effectively extract the spatial and temporal features of the sensor data for anomaly detection.

cs.CR↗

arXiv2025

Cross-Modal Deep Metric Learning for Time Series Anomaly Detection

To effectively address the issues of low sensitivity and high time consumption in time series anomaly detection, we propose an anomaly detection method based on cross-modal deep metric learning. A cross-modal deep metric learning feature clustering model is constructed, composed of an input layer, a triplet selection layer, and a loss function computation layer. The squared Euclidean distances between cluster centers are calculated, and a stochastic gradient descent strategy is employed to optimize the model and classify different time series features. The inner product of principal component direction vectors is used as a metric for anomaly measurement. The von Mises-Fisher (vMF) distribution is applied to describe the directional characteristics of time series data, and historical data is used to train and obtain evaluation parameters. By comparing the principal component direction vector of actual time series data with the threshold, anomaly detection is performed. Experimental results demonstrate that the proposed method accurately classifies time series data with different attributes, exhibits high sensitivity to anomalies, and achieves high detection accuracy, fast detection speed, and strong robustness.

cs.LG↗