arXiv subjects

Hao Liu

Publications and source records attributed to Hao Liu.

At least 271 records · Page 15Linked to original sources

arXiv2025

Dolphin: Document Image Parsing via Heterogeneous Anchor Prompting

Document image parsing is challenging due to its complexly intertwined elements such as text paragraphs, figures, formulas, and tables. Current approaches either assemble specialized expert models or directly generate page-level content autoregressively, facing integration overhead, efficiency bottlenecks, and layout structure degradation despite their decent performance. To address these limitations, we present \textit{Dolphin} (\textit{\textbf{Do}cument Image \textbf{P}arsing via \textbf{H}eterogeneous Anchor Prompt\textbf{in}g}), a novel multimodal document image parsing model following an analyze-then-parse paradigm. In the first stage, Dolphin generates a sequence of layout elements in reading order. These heterogeneous elements, serving as anchors and coupled with task-specific prompts, are fed back to Dolphin for parallel content parsing in the second stage. To train Dolphin, we construct a large-scale dataset of over 30 million samples, covering multi-granularity parsing tasks. Through comprehensive evaluations on both prevalent benchmarks and self-constructed ones, Dolphin achieves state-of-the-art performance across diverse page-level and element-level settings, while ensuring superior efficiency through its lightweight architecture and parallel parsing mechanism. The code and pre-trained models are publicly available at https://github.com/ByteDance/Dolphin

cs.CV↗

arXiv2025

MM-Agent: LLM as Agents for Real-world Mathematical Modeling Problem

Mathematical modeling is a cornerstone of scientific discovery and engineering practice, enabling the translation of real-world problems into formal systems across domains such as physics, biology, and economics. Unlike mathematical reasoning, which assumes a predefined formulation, modeling requires open-ended problem analysis, abstraction, and principled formalization. While Large Language Models (LLMs) have shown strong reasoning capabilities, they fall short in rigorous model construction, limiting their utility in real-world problem-solving. To this end, we formalize the task of LLM-powered real-world mathematical modeling, where agents must analyze problems, construct domain-appropriate formulations, and generate complete end-to-end solutions. We introduce MM-Bench, a curated benchmark of 111 problems from the Mathematical Contest in Modeling (MCM/ICM), spanning the years 2000 to 2025 and across ten diverse domains such as physics, biology, and economics. To tackle this task, we propose MM-Agent, an expert-inspired framework that decomposes mathematical modeling into four stages: open-ended problem analysis, structured model formulation, computational problem solving, and report generation. Experiments on MM-Bench show that MM-Agent significantly outperforms baseline agents, achieving an 11.88\% improvement over human expert solutions while requiring only 15 minutes and \$0.88 per task using GPT-4o. Furthermore, under official MCM/ICM protocols, MM-Agent assisted two undergraduate teams in winning the Finalist Award (\textbf{top 2.0\% among 27,456 teams}) in MCM/ICM 2025, demonstrating its practical effectiveness as a modeling copilot. Our code is available at https://github.com/usail-hkust/LLM-MM-Agent

cs.AI↗

arXiv2025

A Bounding Box is Worth One Token: Interleaving Layout and Text in a Large Language Model for Document Understanding

Recently, many studies have demonstrated that exclusively incorporating OCR-derived text and spatial layouts with large language models (LLMs) can be highly effective for document understanding tasks. However, existing methods that integrate spatial layouts with text have limitations, such as producing overly long text sequences or failing to fully leverage the autoregressive traits of LLMs. In this work, we introduce Interleaving Layout and Text in a Large Language Model (LayTextLLM)} for document understanding. LayTextLLM projects each bounding box to a single embedding and interleaves it with text, efficiently avoiding long sequence issues while leveraging autoregressive traits of LLMs. LayTextLLM not only streamlines the interaction of layout and textual data but also shows enhanced performance in KIE and VQA. Comprehensive benchmark evaluations reveal significant improvements of LayTextLLM, with a 15.2% increase on KIE tasks and 10.7% on VQA tasks compared to previous SOTA OCR-based LLMs. All resources are available at https://github.com/LayTextLLM/LayTextLLM.

cs.CL↗

arXiv2025

PhyDA: Physics-Guided Diffusion Models for Data Assimilation in Atmospheric Systems

Data Assimilation (DA) plays a critical role in atmospheric science by reconstructing spatially continous estimates of the system state, which serves as initial conditions for scientific analysis. While recent advances in diffusion models have shown great potential for DA tasks, most existing approaches remain purely data-driven and often overlook the physical laws that govern complex atmospheric dynamics. As a result, they may yield physically inconsistent reconstructions that impair downstream applications. To overcome this limitation, we propose PhyDA, a physics-guided diffusion framework designed to ensure physical coherence in atmospheric data assimilation. PhyDA introduces two key components: (1) a Physically Regularized Diffusion Objective that integrates physical constraints into the training process by penalizing deviations from known physical laws expressed as partial differential equations, and (2) a Virtual Reconstruction Encoder that bridges observational sparsity for structured latent representations, further enhancing the model's ability to infer complete and physically coherent states. Experiments on the ERA5 reanalysis dataset demonstrate that PhyDA achieves superior accuracy and better physical plausibility compared to state-of-the-art baselines. Our results emphasize the importance of combining generative modeling with domain-specific physical knowledge and show that PhyDA offers a promising direction for improving real-world data assimilation systems.

cs.LG↗

arXiv2025

SongEval: A Benchmark Dataset for Song Aesthetics Evaluation

Aesthetics serve as an implicit and important criterion in song generation tasks that reflect human perception beyond objective metrics. However, evaluating the aesthetics of generated songs remains a fundamental challenge, as the appreciation of music is highly subjective. Existing evaluation metrics, such as embedding-based distances, are limited in reflecting the subjective and perceptual aspects that define musical appeal. To address this issue, we introduce SongEval, the first open-source, large-scale benchmark dataset for evaluating the aesthetics of full-length songs. SongEval includes over 2,399 songs in full length, summing up to more than 140 hours, with aesthetic ratings from 16 professional annotators with musical backgrounds. Each song is evaluated across five key dimensions: overall coherence, memorability, naturalness of vocal breathing and phrasing, clarity of song structure, and overall musicality. The dataset covers both English and Chinese songs, spanning nine mainstream genres. Moreover, to assess the effectiveness of song aesthetic evaluation, we conduct experiments using SongEval to predict aesthetic scores and demonstrate better performance than existing objective evaluation metrics in predicting human-perceived musical quality.

eess.AS↗

arXiv2025

Machine learning bridging battery field data and laboratory data

Aiming at the dilemma that most laboratory data-driven diagnostic and prognostic methods cannot be applied to field batteries in passenger cars and energy storage systems, this paper proposes a method to bridge field data and laboratory data using machine learning. Only two field real impedances corresponding to a medium frequency and a high frequency are needed to predict laboratory real impedance curve, laboratory charge/discharge curve, and laboratory relaxation curve. Based on the predicted laboratory data, laboratory data-driven methods can be used for field battery diagnosis and prognosis. Compared with the field data-driven methods based on massive historical field data, the proposed method has the advantages of higher accuracy, lower cost, faster speed, readily available, and no use of private data. The proposed method is tested using two open-source datasets containing 249 NMC cells. For a test set containing 76 cells, the mean absolute percentage errors of laboratory real impedance curve, charge curve, and discharge curve prediction results are 0.85%, 4.72%, and 2.69%, respectively. This work fills the gap between laboratory data-driven diagnostic and prognostic methods and field battery applications, making all laboratory data-driven methods applicable to field battery diagnosis and prognosis. Furthermore, this work overturns the fixed path of developing field battery diagnostic and prognostic methods based on massive field historical data, opening up new research and breakthrough directions for field battery diagnosis and prognosis.

stat.AP↗

arXiv2025

CV-MP: Max-Pressure Control in Heterogeneously Distributed and Partially Connected Vehicle Environments

Max-pressure (MP) control has emerged as a prominent real-time network traffic signal control strategy due to its simplicity, decentralized structure, and theoretical guarantees of network queue stability. Meanwhile, advances in connected vehicle (CV) technology have sparked extensive research into CV-based traffic signal control. Despite these developments, few studies have investigated MP control in heterogeneously distributed and partially CV environments while ensuring network queue stability. To address these research gaps, we propose a CV-based MP control (CV-MP) method that leverages real-time CV travel time information to compute the pressure, thereby incorporating both the spatial distribution and temporal delays of vehicles, unlike existing approaches that utilized only spatial distribution or temporal delays. In particular, we establish sufficient conditions for road network queue stability that are compatible with most existing MP control methods. Moreover, we pioneered the proof of network queue stability even if the vehicles are only partially connected and heterogeneously distributed, and gave a necessary condition of CV observation for maintaining the stability. Evaluation results on an Amsterdam corridor show that CV-MP significantly reduces vehicle delays compared to both actuated control and conventional MP control across various CV penetration rates. Moreover, in scenarios with dynamic traffic demand, CV-MP achieves lower spillover peaks even with low and heterogeneous CV penetration rates, further highlighting its effectiveness and robustness.

eess.SY↗

arXiv2025

Machine learning accelerates fuel cell life testing

Accelerated life testing (ALT) can significantly reduce the economic, time, and labor costs of life testing in the process of equipment, device, and material research and development (R&D), and improve R&D efficiency. This paper proposes a performance characterization data prediction (PCDP) method and a life prediction-driven ALT (LP-ALT) method to accelerate the life test of polymer electrolyte membrane fuel cells (PEMFCs). The PCDP method can accurately predict different PCD using only four impedances (real and imaginary) corresponding to a high frequency and a medium frequency, greatly shortening the measurement time of offline PCD and reducing the difficulty of life testing. The test results on an open source life test dataset containing 42 PEMFCs show that compared with the determination coefficient (R^2) results of predicted aging indicators, including limiting current, total mass transport resistance, electrochemically active surface area, and crossover current, obtained based on the measured PCD, the R^2 results of predicted aging indicators based on the predicted PCD is only reduced by 0.04, 0.01, 0.05, and 0.06, respectively. The LP-ALT method can shorten the life test time through early life prediction. Test results on the same open-source life test dataset of PEMFCs show that the acceleration ratio of the LP-ALT method can reach 30 times under the premise of ensuring that the minimum R^2 of the prediction results of different aging indicators, including limiting current, total mass transport resistance, and electrochemically active surface area, is not less than 0.89. Combining the different performance characterization data predicted by the PCDP method and the life prediction of the LP-ALT method, the diagnosis and prognosis of PEMFCs and their components can be achieved.

stat.AP↗

arXiv2025

Stacked Intelligent Metasurfaces for Wireless Communications: Applications and Challenges

The rapid growth of wireless communications has created a significant demand for high throughput, seamless connectivity, and extremely low latency. To meet these goals, a novel technology -- stacked intelligent metasurfaces (SIMs) -- has been developed to perform signal processing by directly utilizing electromagnetic waves, thus achieving incredibly fast computing speed while reducing hardware requirements. In this article, we provide an overview of SIM technology, including its underlying hardware, benefits, and exciting applications in wireless communications. Specifically, we examine the utilization of SIMs in realizing transmit beamforming and semantic encoding in the wave domain. Additionally, channel estimation in SIM-aided communication systems is discussed. Finally, we highlight potential research opportunities and identify key challenges for deploying SIMs in wireless networks to motivate future research.

cs.IT↗

arXiv2025

A pilot survey on globular clusters with the Wide Field Survey Telescope (WFST)

We carry out an imaging survey of six globular clusters (GCs) with a limit magnitude to 22 mag at the 5 sigma level, down to the main sequence stars of the respective cluster, as one of the pilot observing program of the Wide Field Survey Telescope (WFST). This paper present the early results of this survey, where we investigate the tidal characters at the periphery of the clusters NGC 4147, NGC 5024, NGC 5053, NGC 5272, NGC 5904 and NGC 6341. We present the estimated number density of cluster candidates and their spatial distribution. We confirm the presence of tidal arms in NGC 4147 and NGC 5904 and identify several intriguing potential tidal structures in NGC 4147, NGC 5024, NGC 5272, corroborated the elliptical morphology of the periphery of NGC 6341. WFST shows its ability to detect faint main-sequence stars of clusters beyond 15 kpc in helio-centric distance. Our findings underscore the WFST's capability for probing faint structural features in GCs, paving the way for future in-depth studies, especially for the search of the large scale tidal streams associated with the clusters with the future wide field survey.

astro-ph.GA↗

arXiv2025

Automated Parking Trajectory Generation Using Deep Reinforcement Learning

Autonomous parking is a key technology in modern autonomous driving systems, requiring high precision, strong adaptability, and efficiency in complex environments. This paper proposes a Deep Reinforcement Learning (DRL) framework based on the Soft Actor-Critic (SAC) algorithm to optimize autonomous parking tasks. SAC, an off-policy method with entropy regularization, is particularly well-suited for continuous action spaces, enabling fine-grained vehicle control. We model the parking task as a Markov Decision Process (MDP) and train an agent to maximize cumulative rewards while balancing exploration and exploitation through entropy maximization. The proposed system integrates multiple sensor inputs into a high-dimensional state space and leverages SAC's dual critic networks and policy network to achieve stable learning. Simulation results show that the SAC-based approach delivers high parking success rates, reduced maneuver times, and robust handling of dynamic obstacles, outperforming traditional rule-based methods and other DRL algorithms. This study demonstrates SAC's potential in autonomous parking and lays the foundation for real-world applications.

cs.RO↗

arXiv2025

GOFA: A Generative One-For-All Model for Joint Graph Language Modeling

Foundation models, such as Large Language Models (LLMs) or Large Vision Models (LVMs), have emerged as one of the most powerful tools in the respective fields. However, unlike text and image data, graph data do not have a definitive structure, posing great challenges to developing a Graph Foundation Model (GFM). For example, current attempts at designing general graph models either transform graph data into a language format for LLM-based prediction or still train a GNN model with LLM as an assistant. The former can handle unlimited tasks, while the latter captures graph structure much better -- yet, no existing work can achieve both simultaneously. In this paper, we identify three key desirable properties of a GFM: self-supervised pretraining, fluidity in tasks, and graph awareness. To account for these properties, we extend the conventional language modeling to the graph domain and propose a novel generative graph language model GOFA to solve the problem. The model interleaves randomly initialized GNN layers into a frozen pre-trained LLM so that the semantic and structural modeling abilities are organically combined. GOFA is pre-trained on newly proposed graph-level next-word prediction, question-answering, and structural tasks to obtain the above GFM properties. The pre-trained model is further fine-tuned on downstream tasks to obtain task-solving ability. The fine-tuned model is evaluated on various downstream tasks, demonstrating a strong ability to solve structural and contextual problems in zero-shot scenarios. The code is available at https://github.com/JiaruiFeng/GOFA.

cs.LG↗

arXiv2025

Robust Transmission Design for Reconfigurable Intelligent Surface and Movable Antenna Enabled Symbiotic Radio Communications

This paper explores the application of movable antenna (MA), a cutting-edge technology with the capability of altering antenna positions, in a symbiotic radio (SR) system enabled by reconfigurable intelligent surface (RIS). The goal is to fully exploit the capabilities of both MA and RIS, constructing a better transmission environment for the co-existing primary and secondary transmission systems. For both parasitic SR (PSR) and commensal SR (CSR) scenarios with the channel uncertainties experienced by all transmission links, we design a robust transmission scheme with the goal of maximizing the primary rate while ensuring the secondary transmission quality. To address the maximization problem with thorny non-convex characteristics, we propose an alternating optimization framework that utilizes the General S-Procedure, General Sign-Definiteness Principle, successive convex approximation (SCA), and simulated annealing (SA) improved particle swarm optimization (SA-PSO) algorithms. Numerical results validate that the CSR scenario significantly outperforms the PSR scenario in terms of primary rate, and also show that compared to the fixed-position antenna scheme, the proposed MA scheme can increase the primary rate by 1.62 bps/Hz and 2.37 bps/Hz for the PSR and CSR scenarios, respectively.

eess.SP↗

arXiv2025

Small Alfvén Number Limit for the Global-in-time Solutions of Incompressible MHD Equations with General Initial Data

The small Alfvén number (denoted by $\varepsilon$) limit (one type of large parameter limits, i.e. singular limits) in magnetohydrodynamic (abbr. MHD) equations was first proposed by Klainerman--Majda in (Comm. Pure Appl. Math. 34: 481--524, 1981). Recently Ju--Wang--Xu mathematically verified that the \emph{local-in-time} solutions of three-dimensional (abbr. 3D) ideal (i.e. the absence of the dissipative terms) incompressible MHD equations with general initial data in $\mathbb{T}^3$ (i.e. a spatially periodic domain) tend to a solution of 2D ideal MHD equations in the distribution sense as $\varepsilon\to 0$ by Schochet's fast averaging method in (J. Differential Equations, 114: 476--512, 1994). In this paper, we revisit the small Alfvén number limit in $\mathbb{R}^n$ with $n=2$, $3$, and develop another approach, motivated by Cai--Lei's energy method in (Arch. Ration. Mech. Anal. 228: 969--993, 2018), to establish a new conclusion that the \emph{global-in-time} solutions of incompressible MHD equations (including the viscous resistive case) with general initial data converge to zero as $\varepsilon\to 0$ for any given time-space variable $(x,t)$ with $t>0$. In addition, we find that the large perturbation solutions and vanishing phenomenon of the nonlinear interactions also exist in the \emph{viscous resistive} MHD equations for small Alfvén numbers, and thus extend Bardos et al.'s results of the \emph{ideal} MHD equations in (Trans Am Math Soc 305: 175--191, 1988).

math.AP↗

arXiv2025

Mapping ultrafast timing jitter in dispersion-managed 89 GHz frequency microcombs via self-heterodyne linear interferometry

Laser frequency microcombs provide equidistant coherent frequency markers over a broad spectrum, enabling new frontiers in chip-scale frequency metrology, laser spectroscopy, dense optical communications, precision distance metrology and astronomy. Here we demonstrate thermally stabilized frequency microcomb formation in dispersion-managed microresonators at the different mode-locking states featured with the negligible center frequency shift and broad frequency bandwidth. Subsequently, femtosecond timing jitter in the microcombs are characterized, supported by precision metrology on the timing phase, relative intensity noise and instantaneous linewidth. We contrast the fundamental noise for a range of 89 GHz microcomb states, from soliton crystals to multiple solitons and single-soliton regimes, determined by pump-resonance detuning. For the single-soliton state, we report a close-to-shot-noise-limited relative intensity noise of -153.2 dB/Hz and a quantum-noise-limited timing jitter power spectral density of 0.4 as2/Hz, at 100 kHz offset frequency. This is enabled by a self-heterodyne linear interferometer with 94.2 zs/Hz1/2 timing resolution, 50.6 mHz/Hz1/2 RF frequency resolution, and 6.7 uV/Hz frequency discrimination sensitivity. We achieve an integrated timing jitter at 1.7 fs, integrated from 10 kHz to 1 MHz. Measuring and understanding the fundamental noise parameters in these high-clock-rate frequency microcombs are essential to advance soliton physics and precision microwave-optical clockwork.

physics.optics↗

arXiv2025

ScoreFusion: Fusing Score-based Generative Models via Kullback-Leibler Barycenters

We introduce ScoreFusion, a theoretically grounded method for fusing multiple pre-trained diffusion models that are assumed to generate from auxiliary populations. ScoreFusion is particularly useful for enhancing the generative modeling of a target population with limited observed data. Our starting point considers the family of KL barycenters of the auxiliary populations, which is proven to be an optimal parametric class in the KL sense, but difficult to learn. Nevertheless, by recasting the learning problem as score matching in denoising diffusion, we obtain a tractable way of computing the optimal KL barycenter weights. We prove a dimension-free sample complexity bound in total variation distance, provided that the auxiliary models are well-fitted for their own task and the auxiliary tasks combined capture the target well. The sample efficiency of ScoreFusion is demonstrated by learning handwritten digits. We also provide a simple adaptation of a Stable Diffusion denoising pipeline that enables sampling from the KL barycenter of two auxiliary checkpoints; on a portrait generation task, our method produces faces that enhance population heterogeneity relative to the auxiliary distributions.

stat.ML↗

arXiv2025

Hexagonal and k-hexagonal graph's normalized Laplacian spectrum and applications

Substituting each edge of a simple connected graph $G$ by a path of length 1 and $k$ paths of length 5 generates the $k$-hexagonal graph $H^k(G)$. Iterative graph $H^k_n(G)$ is produced when the preceding constructions are repeated $n$ times. According to the graph structure, we obtain a set of linear equations, and derive the entirely normalized Laplacian spectrum of $H^k_n(G)$ when $k = 1$ and $k \geqslant 2$ respectively by analyzing the structure of the solutions of these linear equations. We find significant formulas to calculate the Kemeny's constant, multiplicative degree-Kirchhoff index and number of spanning trees of $H^k_n(G)$ as applications.

math.CO↗

arXiv2025

TP-RAG: Benchmarking Retrieval-Augmented Large Language Model Agents for Spatiotemporal-Aware Travel Planning

Large language models (LLMs) have shown promise in automating travel planning, yet they often fall short in addressing nuanced spatiotemporal rationality. While existing benchmarks focus on basic plan validity, they neglect critical aspects such as route efficiency, POI appeal, and real-time adaptability. This paper introduces TP-RAG, the first benchmark tailored for retrieval-augmented, spatiotemporal-aware travel planning. Our dataset includes 2,348 real-world travel queries, 85,575 fine-grain annotated POIs, and 18,784 high-quality travel trajectory references sourced from online tourist documents, enabling dynamic and context-aware planning. Through extensive experiments, we reveal that integrating reference trajectories significantly improves spatial efficiency and POI rationality of the travel plan, while challenges persist in universality and robustness due to conflicting references and noisy data. To address these issues, we propose EvoRAG, an evolutionary framework that potently synergizes diverse retrieved trajectories with LLMs' intrinsic reasoning. EvoRAG achieves state-of-the-art performance, improving spatiotemporal compliance and reducing commonsense violation compared to ground-up and retrieval-augmented baselines. Our work underscores the potential of hybridizing Web knowledge with LLM-driven optimization, paving the way for more reliable and adaptive travel planning agents.

cs.CL↗