arXiv subjects

Meng Zhang

Publications and source records attributed to Meng Zhang.

At least 91 records · Page 5Linked to original sources

arXiv2025

A Physics-informed Deep Operator for Real-Time Freeway Traffic State Estimation

Traffic state estimation (TSE) falls methodologically into three categories: model-driven, data-driven, and model-data dual-driven. Model-driven TSE relies on macroscopic traffic flow models originated from hydrodynamics. Data-driven TSE leverages historical sensing data and employs statistical models or machine learning methods to infer traffic state. Model-data dual-driven traffic state estimation attempts to harness the strengths of both aspects to achieve more accurate TSE. From the perspective of mathematical operator theory, TSE can be viewed as a type of operator that maps available measurements of inerested traffic state into unmeasured traffic state variables in real time. For the first time this paper proposes to study real-time freeway TSE in the idea of physics-informed deep operator network (PI-DeepONet), which is an operator-oriented architecture embedding traffic flow models based on deep neural networks. The paper has developed an extended architecture from the original PI-DeepONet. The extended architecture is featured with: (1) the acceptance of 2-D data input so as to support CNN-based computations; (2) the introduction of a nonlinear expansion layer, an attention mechanism, and a MIMO mechanism; (3) dedicated neural network design for adaptive identification of traffic flow model parameters. A traffic state estimator built on the basis of this extended PI-DeepONet architecture was evaluated with respect to a short freeway stretch of NGSIM and a large-scale urban expressway in China, along with other four baseline TSE methods. The evaluation results demonstrated that this novel TSE method outperformed the baseline methods with high-precision estimation results of flow and mean speed.

cs.LG↗

arXiv2025

Spectroscopic ages for 4 million main-sequence dwarf stars from LAMOST DR10 estimated with data-driven approach

Stellar age determination for large samples of stars opens new avenues for a broad range of astronomical sciences. While precise stellar ages for evolved stars have been derived from large ground- and space-based stellar surveys, reliable age determination for cool main-sequence dwarf stars remains a challenge. In this work, we set out to estimate the age of dwarf stars from the LAMOST spectra with a data-driven approach. We build a training set by using wide binaries that the primary component has reliable isochrone age estimate thus gives the age of the secondary. This training set is further supplemented with field stars and cluster stars whose ages are known. We then train a data-driven model for inferring age from their spectra with the XGBoost algorithm. Given a spectral signal-to-noise ratio greater than 50, the age estimation precise to 10% to 25% for K-type stars, as younger stars have larger relative errors. Validations suggest that the underlying information used for our age estimation is largely attributed to the LAMOST spectral features of chemical abundances. It means our result is a manifestation of stellar chemical clock effectively acted on LAMOST spectra ($R\simeq1800$). Applying our model to the LAMOST DR10 yields a massive age catalog for $\sim4$ million dwarf stars. Statistical properties, such as the age distribution, age-abundance and age-stellar activity relations of the sample stars are discussed. The catalog is publicly accessible and can be helpful for extensive sciences from detection and characterization of Earth-like planets to Galactic archaeology.

astro-ph.SR↗

arXiv2025

MagicVL-2B: Empowering Vision-Language Models on Mobile Devices with Lightweight Visual Encoders via Curriculum Learning

Vision-Language Models (VLMs) have achieved remarkable breakthroughs in recent years, enabling a diverse array of applications in everyday life. However, the substantial computational and storage demands of VLMs pose significant challenges for their efficient deployment on mobile devices, which represent the most ubiquitous and accessible computing platforms today. In this work, we introduce MagicVL-2B, a novel VLM meticulously optimized for flagship smartphones. MagicVL-2B leverages a lightweight visual encoder with fewer than 100M parameters and features a redesigned dynamic resolution scheme that adaptively generates image tokens without excessive modification of image dimensions. To further enhance the performance of this compact encoder within VLMs, we propose a multimodal curriculum learning strategy that incrementally increases task difficulty and data information density throughout training. This approach substantially improves the model's performance across a variety of sub-tasks. Extensive evaluations on standard VLM benchmarks demonstrate that MagicVL-2B matches the accuracy of current state-of-the-art models while reducing on-device power consumption by 41.1%. These results establish MagicVL-2B as a practical and robust solution for real-world mobile vision-language applications, enabling advanced multimodal intelligence to run directly on smartphones.

cs.CV↗

arXiv2025

Robust Low-light Scene Restoration via Illumination Transition

Synthesizing normal-light novel views from low-light multiview images is an important yet challenging task, given the low visibility and high ISO noise present in the input images. Existing low-light enhancement methods often struggle to effectively preprocess such low-light inputs, as they fail to consider correlations among multiple views. Although other state-of-the-art methods have introduced illumination-related components offering alternative solutions to the problem, they often result in drawbacks such as color distortions and artifacts, and they provide limited denoising effectiveness. In this paper, we propose a novel Robust Low-light Scene Restoration framework (RoSe), which enables effective synthesis of novel views in normal lighting conditions from low-light multiview image inputs, by formulating the task as an illuminance transition estimation problem in 3D space, conceptualizing it as a specialized rendering task. This multiview-consistent illuminance transition field establishes a robust connection between low-light and normal-light conditions. By further exploiting the inherent low-rank property of illumination to constrain the transition representation, we achieve more effective denoising without complex 2D techniques or explicit noise modeling. To implement RoSe, we design a concise dual-branch architecture and introduce a low-rank denoising module. Experiments demonstrate that RoSe significantly outperforms state-of-the-art models in both rendering quality and multiview consistency on standard benchmarks. The codes and data are available at https://pegasus2004.github.io/RoSe.

cs.CV↗

arXiv2025

Digital Salon: An AI and Physics-Driven Tool for 3D Hair Grooming and Simulation

We introduce Digital Salon, a comprehensive hair authoring system that supports real-time 3D hair generation, simulation, and rendering. Unlike existing methods that focus on isolated parts of 3D hair modeling and involve a heavy computation process or network training, Digital Salon offers a holistic and interactive system that lowers the technical barriers of 3D hair modeling through natural language-based interaction. The system guides users through four key stages: text-guided hair retrieval, real-time hair simulation, interactive hair refinement, and hair-conditioned image generation. This cohesive workflow makes advanced hair design accessible to users of varying skill levels and dramatically streamlines the creative process in digital media with an intuitive, versatile, and efficient solution for hair modeling. User studies show that our system can outperform traditional hair modeling workflows for rapid prototyping. Furthermore, we provide insights into the benefits of our system with future potential of deploying our system in real salon environments. More details can be found on our project page: https://digital-salon.github.io/.

cs.GR↗

arXiv2025

An Information-Theoretic Analysis for Federated Learning under Concept Drift

Recent studies in federated learning (FL) commonly train models on static datasets. However, real-world data often arrives as streams with shifting distributions, causing performance degradation known as concept drift. This paper analyzes FL performance under concept drift using information theory and proposes an algorithm to mitigate the performance degradation. We model concept drift as a Markov chain and introduce the \emph{Stationary Generalization Error} to assess a model's capability to capture characteristics of future unseen data. Its upper bound is derived using KL divergence and mutual information. We study three drift patterns (periodic, gradual, and random) and their impact on FL performance. Inspired by this, we propose an algorithm that regularizes the empirical risk minimization approach with KL divergence and mutual information, thereby enhancing long-term performance. We also explore the performance-cost tradeoff by identifying a Pareto front. To validate our approach, we build an FL testbed using Raspberry Pi4 devices. Experimental results corroborate with theoretical findings, confirming that drift patterns significantly affect performance. Our method consistently outperforms existing approaches for these three patterns, demonstrating its effectiveness in adapting concept drift in FL.

cs.LG↗

arXiv2025

Universally enhanced superconductivity and coexisting ferroelectricity at oxide interfaces

The coexistence of superconductivity and ferroelectricity is rare due to their conflicting requirements: superconductivity relies on free charge carriers, whereas ferroelectricity typically occurs in insulating systems. At LaAlO3/KTaO3 interfaces, we demonstrate the coexistence of two-dimensional superconductivity and ferroelectricity, enabled by the unique properties of KTaO3 as a quantum paraelectric. Systematic gating and poling experiments reveal a universal enhancement of the superconducting transition temperature (Tc) by 0.2-0.6 K and bistable transport properties, including hysteresis, strongly suggesting the existence of switchable ferroelectric polarization in the interfacial conducting layer. Hysteresis loops indicate robust ferroelectricity below 50 K. The Tc enhancement is attributed to ferroelectric polarization-induced reduction in dielectric constant, which narrows the interfacial potential well, confining carriers closer to the interface. The bistability arises from switchable ferroelectric polarization, which modulates the potential well depending on polarization direction. These findings establish a straightforward mechanism coupling ferroelectricity and superconductivity, providing a promising platform for exploring their interplay.

cond-mat.supr-con↗

arXiv2025

Homogeneous Stellar Atmospheric Parameters and 22 Elemental Abundances for FGK Stars Derived From LAMOST Low-resolution Spectra with DD-Payne

A deep understanding of our Galaxy desires detailed decomposition of its stellar populations via their chemical fingerprints. This requires precise stellar abundances of many elements for a large number of stars. Here we present an updated catalog of stellar labels derived from LAMOST low-resolution spectra in a physics-sensible and rigorous manner with DD-Payne, taking labels from high-resolution spectroscopy as training set. The catalog contains atmospheric parameters for 6.4 million stars released in LAMOST DR9, and abundances for 22 elements, namely, C, N, O, Na, Mg, Al, Si, Ca, Ti, Cr, Mn, Fe, Ni, Sr, Y, Zr, Ba, La, Ce, Nd, Sm, and Eu, for nearly 3.6 million stars with spectral signal-to-noise ratio (SNR) higher than 20. The [Fe/H] is valid down to $\sim$-4.0, while elemental abundance ratios [X/Fe] are mostly valid for stars with [Fe/H] $\gtrsim-2.0$. Measurement errors in these labels are sensitive to and almost inversely proportional with SNR. For stars with S/N>50, we achieved a typical error of 30 K in Teff, 0.07 dex in $\log g$, $\sim0.05$ dex in abundances for most elements with atomic number smaller than Sr, and 0.1--0.2 dex for heavier elements. Homogenization to the label estimates is carried out via dedicated internal and external calibration. In particular, the non-local thermal equilibrium effect is corrected for the [Fe/H] estimates, the Teff is calibrated to the infrared flux method scale, and the $\log~g$ is validated with asteroseismic measurements. The elemental abundances are internally calibrated using wide binaries, eliminating systematic trend with effective temperature. The catalog is publicly available.

astro-ph.SR↗

arXiv2025

VideoCap-R1: Enhancing MLLMs for Video Captioning via Structured Thinking

While recent advances in reinforcement learning have significantly enhanced reasoning capabilities in large language models (LLMs), these techniques remain underexplored in multi-modal LLMs for video captioning. This paper presents the first systematic investigation of GRPO-based RL post-training for video MLLMs, with the goal of enhancing video MLLMs' capability of describing actions in videos. Specifically, we develop the VideoCap-R1, which is prompted to first perform structured thinking that analyzes video subjects with their attributes and actions before generating complete captions, supported by two specialized reward mechanisms: a LLM-free think scorer evaluating the structured thinking quality and a LLM-assisted caption scorer assessing the output quality. The RL training framework effectively establishes the connection between structured reasoning and comprehensive description generation, enabling the model to produce captions with more accurate actions. Our experiments demonstrate that VideoCap-R1 achieves substantial improvements over the Qwen2VL-7B baseline using limited samples (1.5k) across multiple video caption benchmarks (DREAM1K: +4.4 event F1, VDC: +4.2 Acc, CAREBENCH: +3.1 action F1, +6.9 object F1) while consistently outperforming the SFT-trained counterparts, confirming GRPO's superiority in enhancing MLLMs' captioning capabilities.

cs.CV↗

arXiv2025

HMPC-assisted Adversarial Inverse Reinforcement Learning for Smart Home Energy Management

This letter proposes an Adversarial Inverse Reinforcement Learning (AIRL)-based energy management method for a smart home, which incorporates an implicit thermal dynamics model. In the proposed method, historical optimal decisions are first generated using a neural network-assisted Hierarchical Model Predictive Control (HMPC) framework. These decisions are then used as expert demonstrations in the AIRL module, which aims to train a discriminator to distinguish expert demonstrations from transitions generated by a reinforcement learning agent policy, while simultaneously updating the agent policy that can produce transitions to confuse the discriminator. The proposed HMPC-AIRL method eliminates the need for explicit thermal dynamics models, prior or predictive knowledge of uncertain parameters, or manually designed reward functions. Simulation results based on real-world traces demonstrate the effectiveness and data efficiency of the proposed method.

eess.SY↗

arXiv2025

Pangu Light: Weight Re-Initialization for Pruning and Accelerating LLMs

Large Language Models (LLMs) deliver state-of-the-art capabilities across numerous tasks, but their immense size and inference costs pose significant computational challenges for practical deployment. While structured pruning offers a promising avenue for model compression, existing methods often struggle with the detrimental effects of aggressive, simultaneous width and depth reductions, leading to substantial performance degradation. This paper argues that a critical, often overlooked, aspect in making such aggressive joint pruning viable is the strategic re-initialization and adjustment of remaining weights to improve the model post-pruning training accuracies. We introduce Pangu Light, a framework for LLM acceleration centered around structured pruning coupled with novel weight re-initialization techniques designed to address this ``missing piece''. Our framework systematically targets multiple axes, including model width, depth, attention heads, and RMSNorm, with its effectiveness rooted in novel re-initialization methods like Cross-Layer Attention Pruning (CLAP) and Stabilized LayerNorm Pruning (SLNP) that mitigate performance drops by providing the network a better training starting point. Further enhancing efficiency, Pangu Light incorporates specialized optimizations such as absorbing Post-RMSNorm computations and tailors its strategies to Ascend NPU characteristics. The Pangu Light models consistently exhibit a superior accuracy-efficiency trade-off, outperforming prominent baseline pruning methods like Nemotron and established LLMs like Qwen3 series. For instance, on Ascend NPUs, Pangu Light-32B's 81.6 average score and 2585 tokens/s throughput exceed Qwen3-32B's 80.9 average score and 2225 tokens/s.

cs.CL↗

arXiv2025

Soft Weighted Machine Unlearning

Machine unlearning, as a post-hoc processing technique, has gained widespread adoption in addressing challenges like bias mitigation and robustness enhancement, colloquially, machine unlearning for fairness and robustness. However, existing non-privacy unlearning-based solutions persist in using binary data removal framework designed for privacy-driven motivation, leading to significant information loss, a phenomenon known as over-unlearning. While over-unlearning has been largely described in many studies as primarily causing utility degradation, we investigate its fundamental causes and provide deeper insights in this work through counterfactual leave-one-out analysis. In this paper, we introduce a weighted influence function that assigns tailored weights to each sample by solving a convex quadratic programming problem analytically. Building on this, we propose a soft-weighted framework enabling fine-grained model adjustments to address the over-unlearning challenge. We demonstrate that the proposed soft-weighted scheme is versatile and can be seamlessly integrated into most existing unlearning algorithms. Extensive experiments show that in fairness- and robustness-driven tasks, the soft-weighted scheme significantly outperforms hard-weighted schemes in fairness/robustness metrics and alleviates the decline in utility metric, thereby enhancing machine unlearning algorithm as an effective correction solution.

cs.LG↗

arXiv2025

Timely CPU Scheduling for Computation-intensive Status Updates

The proliferation of mobile devices and real-time status updating applications has motivated the optimization of data freshness in the context of age of information (AoI). Meanwhile, increasing computational demands have inspired research on CPU scheduling. Since prior CPU scheduling strategies have ignored data freshness and prior age-minimization strategies have considered only constant CPU speed, we formulate the first CPU scheduling problem as a constrained semi-Markov decision process (SMDP) problem with uncountable space, which aims to minimize the long-term average age of information, subject to an average CPU power constraint. We optimize strategies that specify when the CPU sleeps and adapt the CPU speed (clock frequency) during the execution of update-processing tasks. We consider the age-minimal CPU scheduling problem for both predictable task size (PTS) and unpredictable task size (UTS) cases, where the task size is realized at the start (PTS) or at the completion (UTS) of the task, respectively. To address the non-convex objective, we employ Dinkelbach's fractional programming method to transform our problem into an average cost SMDP. We develop a value-iteration-based algorithm and prove its convergence to obtain optimal policies and structural results for both the PTS and UTS systems. Compared to constant CPU speed, numerical results show that our proposed scheme can reduce the AoI by 50\% or more, with increasing benefits under tighter power constraints. Further, for a given AoI target, the age-minimal CPU scheduling policy can reduce the energy consumption by 50\% or more, with greater AoI reductions when the task size distribution exhibits higher variance.

cs.IT↗

arXiv2025

Perm: A Parametric Representation for Multi-Style 3D Hair Modeling

We present Perm, a learned parametric representation of human 3D hair designed to facilitate various hair-related applications. Unlike previous work that jointly models the global hair structure and local curl patterns, we propose to disentangle them using a PCA-based strand representation in the frequency domain, thereby allowing more precise editing and output control. Specifically, we leverage our strand representation to fit and decompose hair geometry textures into low- to high-frequency hair structures, termed guide textures and residual textures, respectively. These decomposed textures are later parameterized with different generative models, emulating common stages in the hair grooming process. We conduct extensive experiments to validate the architecture design of Perm, and finally deploy the trained model as a generic prior to solve task-agnostic problems, further showcasing its flexibility and superiority in tasks such as single-view hair reconstruction, hairstyle editing, and hair-conditioned image generation. More details can be found on our project page: https://cs.yale.edu/homes/che/projects/perm/.

cs.CV↗

arXiv2025

Biomed-DPT: Dual Modality Prompt Tuning for Biomedical Vision-Language Models

Prompt learning is one of the most effective paradigms for adapting pre-trained vision-language models (VLMs) to the biomedical image classification tasks in few shot scenarios. However, most of the current prompt learning methods only used the text prompts and ignored the particular structures (such as the complex anatomical structures and subtle pathological features) in the biomedical images. In this work, we propose Biomed-DPT, a knowledge-enhanced dual modality prompt tuning technique. In designing the text prompt, Biomed-DPT constructs a dual prompt including the template-driven clinical prompts and the large language model (LLM)-driven domain-adapted prompts, then extracts the clinical knowledge from the domain-adapted prompts through the knowledge distillation technique. In designing the vision prompt, Biomed-DPT introduces the zero vector as a soft prompt to leverage attention re-weighting so that the focus on non-diagnostic regions and the recognition of non-critical pathological features are avoided. Biomed-DPT achieves an average classification accuracy of 66.14\% across 11 biomedical image datasets covering 9 modalities and 10 organs, with performance reaching 78.06\% in base classes and 75.97\% in novel classes, surpassing the Context Optimization (CoOp) method by 6.20\%, 3.78\%, and 8.04\%, respectively. Our code are available at \underline{https://github.com/Kanyooo/Biomed-DPT}.

cs.CV↗

arXiv2025

Metallicities of 20 Million Giant Stars Based on Gaia XP spectra

We design an uncertainty-aware cost-sensitive neural network (UA-CSNet) to estimate metallicities from dereddened and corrected Gaia BP/RP (XP) spectra for giant stars. This method accounts for both stochastic errors in the input spectra and the imbalanced density distribution in [Fe/H] values. With a specialized architecture and training strategy, the UA-CSNet improves the precision of the predicted metallicities, especially for very metal-poor (VMP; $\rm [Fe/H] \leq -2.0$) stars. With the PASTEL catalog as the training sample, our model can estimate metallicities down to $\rm [Fe/H] \sim -4$. We compare our estimates with a number of external catalogs and conduct tests using star clusters, finding overall good agreement. We also confirm that our estimates for VMP stars are unaffected by carbon enhancement. Applying the UA-CSNet, we obtain reliable and precise metallicity estimates for approximately 20 million giant stars, including 360,000 VMP stars and 50,000 extremely metal-poor (EMP; $\rm [Fe/H] \leq -3.0$) stars. The resulting catalog is publicly available at https://doi.org/10.12149/101604. This work highlights the potential of low-resolution spectra for metallicity estimation and provides a valuable dataset for studying the formation and chemo-dynamical evolution of our Galaxy.

astro-ph.SR↗

arXiv2025

Graph Attention Reinforcement Learning for Multicast Routing and Age-Optimal Scheduling

Multicast routing is essential for real-time group applications, such as video streaming, virtual reality, and metaverse platforms, where the Age of Information (AoI) acts as a crucial metric to assess information timeliness. This paper studies dynamic multicast networks with the objective of minimizing the expected average Age of Information (AoI) by jointly optimizing multicast routing and scheduling. The main challenges stem from the intricate coupling between routing and scheduling decisions, the inherent complexity of multicast operations, and the graph representation. We first decompose the original problem into two subtasks amenable to hierarchical reinforcement learning (RL) methods. We propose the first RL framework to address the multicast routing problem, also known as the Steiner Tree problem, by incorporating graph embedding and the successive addition of nodes and links. For graph embedding, we propose the Normalized Graph Attention mechanism (NGAT) framework with a proven contraction mapping property, enabling effective graph information capture and superior generalization within the hierarchical RL framework. We validate our framework through experiments on four datasets, including the real-world AS-733 dataset. The results demonstrate that our proposed scheme can be up to 9.85 times more computationally efficient than traditional multicast routing algorithms, achieving approximation ratios of 1.1-1.3 that are not only comparable to state-of-the-art (SOTA) methods but also highlight its superior generalization capabilities, performing effectively on unseen and more complex tasks. Additionally, our age-optimal TGMS algorithm reduces the average weighted Age of Information (AoI) by 25.6% and the weighted peak age by 29.2% under low-energy scenarios.

cs.NI↗

arXiv2025

Age of Information in Random Access Networks with Energy Harvesting

We study the age of information (AoI) in a random access network consisting of multiple source-destination pairs, where each source node is empowered by energy harvesting capability. Every source node transmits a sequence of data packets to its destination using only the harvested energy. Each data packet is encoded with finite-length codewords, characterizing the nature of short codeword transmissions in random access networks. By combining tools from bulk-service Markov chains with stochastic geometry, we derive an analytical expression for the network average AoI and obtain closed-form results in two special cases, i.e., the small and large energy buffer size scenarios. Our analysis reveals the trade-off between energy accumulation time and transmission success probability. We then optimize the network average AoI by jointly adjusting the update rate and the blocklength of the data packet. Our findings indicate that the optimal update rate should be set to one in the energy-constrained regime where the energy consumption rate exceeds the energy arrival rate. This also means if the optimal blocklength of the data packet is pre-configured, an energy buffer size supporting only one transmission is sufficient.

cs.NI↗