Searcharxiv⌕ Search

arXiv subjects

Hongyu Wang

Publications and source records attributed to Hongyu Wang.

At least 37 records · Page 2Linked to original sources
arXiv2026

Visual Para-Thinker: Divide-and-Conquer Reasoning for Visual Comprehension

Existing LLM test-time scaling laws emphasize the emergence of self-reflective behaviors through extended reasoning length. Nevertheless, this vertical scaling strategy often encounters plateaus in exploration as the model becomes locked into specific thinking pattern. By shifting from depth to parallelism, parallel thinking mitigates the narrowing of exploration. However, the extension of this paradigm to visual domain remains an open research question. In this paper, we first examine the role of visual partitioning in parallelized reasoning and subsequently propose two distinct strategies. Based on the above, we introduce Visual Para-Thinker, representing the inaugural parallel reasoning framework for MLLMs. To maintain path independence and promote diversity in reasoning, our approach integrates Pa-Attention alongside LPRoPE. Leveraging the vLLM framework, we have developed a native multimodal implementation that facilitates high-efficiency parallel processing. Empirical results on benchmark datasets such as V*, CountBench, RefCOCO, and HallusionBench confirm that Visual Para-Thinker successfully extends the benefits of parallel reasoning to the visual domain.

cs.CV↗
arXiv2026

Truth or Tribe: How In-group Favoritism Prioritize Facts in Persona Agents

In-group favoritism refers to the phenomena of favoring members of one's in-group over out-group members and is widely observed in numerous social cooperative behaviors. Recently, in-group favoritism biases have also been identified in generative language models. However, whether the in-group favoritism exists when persona agents are faced with contradicting information (e.g., misinformation), and how to mitigate the adverse effects of in-group favoritism biases in persona agents have been understudied. To address these problems, we propose a Truth or Tribe simulation framework to study the agent cooperation within the spread of contradicting information through a triadic interaction paradigm, and conduct controlled trials to evaluate the primary moderating factors. Extensive results showcase that persona agents display strong in-group favoritism, accepting incorrect answers from identity-similar peers at much higher rates than from dissimilar peers. In-group favoritism continues to emerge in defeasible reasoning contexts where no absolute truth exists, and it intensifies as cognitive complexity increases. Furthermore, three intervention strategies--Identity-Blind Instruction, Structured Counterfactual Reasoning, and Heterogeneous Perspective Ensemble--are proposed to mitigate the in-group favoritism.

cs.AI↗
arXiv2026

High-Integration multimode waveguide grating based CWDM4 MUX/DEMUX with Flat Wide Passband and Ultra-Low Crosstalk for 2xFR4 Module Applications

This work presents a compact, low-crosstalk CWDM4 MUX/DEMUX utilizing cascaded multimode waveguide grating filters. The individual filters are designed with finite Gaussian apodization and positive dispersion, enabling strong unilateral sidelobe suppression while maintaining a minimum feature size compatible with UV lithography. By cascading these filters, we demonstrate a DEMUX that achieves channel crosstalk below -25 dB, insertion loss under 1 dB, and a flat-top bandwidth of approximately 18 nm. The entire device occupies a compact footprint of only 1.6 mm x 40 um, with a channel spacing compatible with commercial TIA and driver chips. Furthermore, a series-parallel hybrid cascade configuration can further suppress the crosstalk to -40 dB.

physics.optics↗
arXiv2026

A Digital Optical Switch Based on a Thermally Tuned Multimode Waveguide Grating Filter

All-optical switching technology is a key solution to the future energy crisis in AI computing, where the performance of optical switches plays a critical role. Conventional integrated optical switches typically suffer from poor robustness to voltage fluctuations, fabrication variations, and temperature drifts. These limitations necessitate complex high-precision real-time calibration and control circuits, which greatly restrict their practical use. This paper presents a digital optical switch based on a thermally tuned multimode waveguide grating (MWG) filter. The switch maintains its on- and off-states across two voltage ranges: 0-0.7 V and 1.1-1.7 V, with a wide operating voltage margin of 0.6 V. It also exhibits excellent robustness to fabrication variations and temperature drifts. By introducing an innovative combination of positive dispersion and parabolic apodization design, the power consumption is reduced by two-thirds, reaching a maximum of only 6 mW. Owing to its low power consumption and wide voltage range, the device can be directly driven by digital signals, allowing for a simplified driver circuitry and a significant reduction in both energy use and overall cost. In addition, the switch offers low insertion loss (<0.5 dB), high extinction ratio (>20 dB), and fast switching (300 μs), demonstrating excellent overall performance and promising application prospects.

physics.optics↗
arXiv2026

SparseBalance: Load-Balanced Long Context Training with Dynamic Sparse Attention

While sparse attention mitigates the computational bottleneck of long-context LLM training, its distributed training process exhibits extreme heterogeneity in both \textit{1)} sequence length and \textit{2)} sparsity sensitivity, leading to a severe imbalance problem and sub-optimal model accuracy. Existing algorithms and training frameworks typically focus on single issue, failing to systematically co-optimize these two problems. Therefore, we propose SparseBalance, a novel algorithm-system co-design framework, which exploits the sparsity and sequence heterogeneity to optimize model accuracy and system efficiency jointly. First, we propose workload-aware dynamic sparsity tuning, which employs a bidirectional sparsity adjustment to eliminate stragglers and exploit inherent bubbles for free accuracy. Second, we propose a sparsity-aware batching strategy to achieve coarse-grained balance, which complements dynamic sparsity tuning. Experimental results demonstrate that SparseBalance achieves up to a 1.33$\times$ end-to-end speedup while still improving the long-context capability by 0.46\% on the LongBench benchmark.

cs.LG↗
arXiv2026

Discrete Token Modeling for Multi-Stem Music Source Separation with Language Models

We propose a generative framework for multi-track music source separation (MSS) that reformulates the task as conditional discrete token generation. Unlike conventional approaches that directly estimate continuous signals in the time or frequency domain, our method combines a Conformer-based conditional encoder, a dual-path neural audio codec (HCodec), and a decoder-only language model to autoregressively generate audio tokens for four target tracks. The generated tokens are decoded back to waveforms through the codec decoder. Evaluation on the MUSDB18-HQ benchmark shows that our generative approach achieves perceptual quality approaching state-of-the-art discriminative methods, while attaining the highest NISQA score on the vocals track. Ablation studies confirm the effectiveness of the learnable Conformer encoder and the benefit of sequential cross-track generation.

eess.AS↗
arXiv2026

Breaking the Training Barrier of Billion-Parameter Universal Machine Learning Interatomic Potentials

Universal Machine Learning Interatomic Potentials (uMLIPs), pre-trained on massively diverse datasets encompassing inorganic materials and organic molecules across the entire periodic table, serve as foundational models for quantum-accurate physical simulations. However, uMLIP training requires second-order derivatives, which lack corresponding parallel training frameworks; moreover, scaling to the billion-parameter regime causes explosive growth in computation and communication overhead, making its training a tremendous challenge. We introduce MatRIS-MoE, a billion-parameter Mixture-of-Experts model built upon invariant architecture, and {Janus}, a pioneering high-dimensional distributed training framework for uMLIPs with hardware-aware optimizations. Deployed across two Exascale supercomputers, our code attains a peak performance of 1.2/1.0 EFLOPS (24\%/{35.5\%} of theoretical peak) in single precision at over 90\% parallel efficiency, compressing the training of billion-parameter uMLIPs from weeks to hours. This work establishes a new high-water mark for AI-for-Science (AI4S) foundation models at Exascale and provides essential infrastructure for rapid scientific discovery.

cs.DC↗
arXiv2026

A Regime Shift in Atlantic Surface Currents Reveals a Step-like Decline of the Meridional Overturning Circulation

The Atlantic surface currents associated with the Atlantic Meridional Overturning Circulation (AMOC) play a central role in regulating Earth's climate, yet their large scale dynamical response to climate variability remains poorly understood. Here we identify a previously unrecognized basin scale phase of Atlantic surface circulation, termed the Atlantic Convergence Divergence Mode (ACDM), characterized by a convergence divergence pattern in the North Atlantic and coherent meridional flows in the South Atlantic. We show that the ACDM experienced a pronounced regime shift in 2009, marked by weakened vertical water exchange and reduced meridional transport. This transition closely coincides with direct RAPID MOCHA AMOC observations and is driven by AMOC modulated multicale forcing: a low frequency oceanic thermal reorganization that preconditions the system, and episodic atmospheric shocks that trigger the shift. By identifying the ACDM variability as a sensitive and physically grounded proxy for interannual AMOC fluctuations, we reveal that the observed 2009 shift signifies a nonlinear, step like weakening of AMOC that triggered a fundamental basin scale reorganization of Atlantic surface currents. Our results offer a dynamical explanation for the AMOC's recent decline and demonstrate its inherently nonlinear nature, highlighting the need to account for step like transitions in assessing its stability and future evolution.

physics.ao-ph↗
arXiv2026

Microscopic evidence of spin-driven multiferroicity and topological spin textures in monolayer NiI2

In type II multiferroics, noncollinear spin textures are expected to induce electric polarization directly, leading to strong magnetoelectric coupling. Realizing such spin driven multiferroicity in two-dimensional systems, and elucidating the interplay between local spins and electric polarization, are of both fundamental and technological importance. Here, using vectorial spin polarized scanning tunneling microscopy, we investigated the spin-driven multiferroicity in monolayer NiI2 at atomic scale. We identify a canted spin-spiral state with fully determined spin rotation plane, accompanied by a 2Q charge modulation. At spin spiral domain walls, we discover topological spin textures that composed of meron/antimeron pairs. These textures are associated with distinct charge pattern and notable band shifts, indicating local bound charges induced by variations of ferroelectricity at domain wall. Our observations are well captured by a realistic spin model incorporating Kitaev interactions and generalized spin-current model of type II multiferroicity. The findings provide microscopic evidence of spin-driven multiferroicity in an extreme 2D system and establish a platform for low-dissipation, electric-field control of topological spin textures.

cond-mat.mes-hall↗
arXiv2026

Decoupling Numerical and Structural Parameters: An Empirical Study on Adaptive Genetic Algorithms via Deep Reinforcement Learning for the Large-Scale TSP

Proper parameter configuration is a prerequisite for the success of Evolutionary Algorithms (EAs). While various adaptive strategies have been proposed, it remains an open question whether all control dimensions contribute equally to algorithmic scalability. To investigate this, we categorize control variables into numerical parameters (e.g., crossover and mutation rates) and structural parameters (e.g., population size and operator switching), hypothesizing that they play distinct roles. This paper presents an empirical study utilizing a dual-level Deep Reinforcement Learning (DRL) framework to decouple and analyze the impact of these two dimensions on the Traveling Salesman Problem (TSP). We employ a Recurrent PPO agent to dynamically regulate these parameters, treating the DRL model as a probe to reveal evolutionary dynamics. Experimental results confirm the effectiveness of this approach: the learned policies outperform static baselines, reducing the optimality gap by approximately 45% on the largest tested instance (rl5915). Building on this validated framework, our ablation analysis reveals a fundamental insight: while numerical tuning offers local refinement, structural plasticity is the decisive factor in preventing stagnation and facilitating escape from local optima. These findings suggest that future automated algorithm design should prioritize dynamic structural reconfiguration over fine-grained probability adjustment. To facilitate reproducibility, the source code is available at https://github.com/StarDream1314/DRLGA-TSP

cs.NE↗
arXiv2026

FNODE: Flow-Matching for data-driven simulation of constrained multibody systems

Data-driven modeling of constrained multibody dynamics remains challenged by (i) the training cost of Neural ODEs, which typically require backpropagation through an ODE solver, and (ii) error accumulation in rollout predictions. We introduce a Flow-Matching Neural ODE (FNODE) framework that learns the acceleration mapping directly from trajectory data by supervising accelerations rather than integrated states, turning training into a supervised regression problem and eliminating the ODE-adjoint/solver backpropagation bottleneck. Acceleration targets are obtained efficiently via numerical differentiation using a hybrid fast Fourier transform (FFT) and finite-difference (FD) scheme. Kinematic constraints are enforced through coordinate partitioning: FNODE learns accelerations only for the independent generalized coordinates, while the dependent coordinates are recovered by solving the position-level constraint equations. We evaluate FNODE on single and triple mass-spring-damper systems, a double pendulum, a slider crank with and without friction, a vehicle model, and a cart-pole, and compare against MBD-NODE, LSTM, and fully connected baselines. Across these benchmarks, FNODE achieves improved prediction accuracy and training/runtime efficiency, while maintaining constraint satisfaction through the partitioning procedure. Our code and scripts are released as open source to support reproducibility and follow-on research.

cs.LG↗
arXiv2026

On the $\mathcal{D}^+_J$ operator on higher-dimensional almost Kähler manifolds

In this paper, we introduce $\mathcal{D}^+_J$, a generalization of $\partial\bar{\partial}$ operator on higher dimensional almost Kähler manifolds. Using the $\mathcal{D}^+_J$ operator, we investigate the $\bar{\partial}$-problem in almost Kähler geometry and explore the generalized Monge-Ampère equation on almost Kähler manifolds. We establish a uniqueness up to the addition of a constant and local existence theorem for this equation. At last, we find an elliptical system for $\mathcal{D}^+_J$ operator. As an application, we reorganize the result of Tosatti-Weinkove-Yau in \cite{TWY}.

math.DG↗
arXiv2026

MatRIS: Toward Reliable and Efficient Pretrained Machine Learning Interatomic Potentials

Foundation MLIPs demonstrate broad applicability across diverse material systems and have emerged as a powerful and transformative paradigm in chemical and computational materials science. Equivariant MLIPs achieve state-of-the-art accuracy in a wide range of benchmarks by incorporating equivariant inductive bias. However, the reliance on tensor products and high-degree representations makes them computationally costly. This raises a fundamental question: as quantum mechanical-based datasets continue to expand, can we develop a more compact model to thoroughly exploit high-dimensional atomic interactions? In this work, we present MatRIS (\textbf{Mat}erials \textbf{R}epresentation and \textbf{I}nteraction \textbf{S}imulation), an invariant MLIP that introduces attention-based modeling of three-body interactions. MatRIS leverages a novel separable attention mechanism with linear complexity $O(N)$, enabling both scalability and expressiveness. MatRIS delivers accuracy comparable to that of leading equivariant models on a wide range of popular benchmarks (Matbench-Discovery, MatPES, MDR phonon, Molecular dataset, etc). Taking Matbench-Discovery as an example, MatRIS achieves an F1 score of up to 0.847 and attains comparable accuracy at a lower training cost. The work indicates that our carefully designed invariant models can match or exceed the accuracy of equivariant models at a fraction of the cost, shedding light on the development of accurate and efficient MLIPs.

cs.LG↗
arXiv2026

BitVLA: 1-bit Vision-Language-Action Models for Robotics Manipulation

Deploying powerful Vision-Language-Action (VLA) models on edge devices is limited by their massive size. In this paper, we take a deployment-oriented view of VLA training: we target efficiency through model design and optimization, rather than relying solely on post-hoc compression. Thus, we propose BitVLA, a fully native 1-bit VLA model for robotic manipulation, where every parameters is ternary, i.e., {-1,0,1}. BitVLA is built on the publicly available 1-bit LLM BitNet b1.58 2B4T, and is trained as a vision-language-action policy that inherits the compactness of 1-bit pretraining while retaining strong task performance. To further reduce the memory footprint of the vision backbone, we introduce Quantize-then-Distill, a post-training quantization-aware training strategy that compresses a full-precision vision encoder to 1.58-bit weights, while a full-precision teacher guides representation alignment during training. Across simulation benchmarks and real-world tasks, BitVLA matches the performance of the full-precision OpenVLA-OFT baseline, while reducing model memory by 11.0x and end-to-end latency by 4.4x. These results suggest a practical path toward training-time efficiency-accuracy co-design for embodied policies, enabling competitive manipulation capability on memory-constrained edge robotic platforms. We release the code in https://github.com/ustcwhy/BitVLA, model weights in https://huggingface.co/lxsy/bitvla-bf16.

cs.RO↗
arXiv2026

Efficient Generative AI Boosts Probabilistic Forecasting of Sudden Stratospheric Warmings

Sudden Stratospheric Warmings (SSWs) are key sources of subseasonal predictability and major drivers of extreme weather in winter. Accurate and efficient probabilistic forecasting of these events remains a persistent challenge for Numerical Weather Prediction (NWP) systems due to computational bottlenecks and limitations in physical representation. While data-driven forecasting is rapidly evolving, its application to the complex, three-dimensional dynamics of SSWs remains underexplored. Here, we bridge this gap by developing a Flow Matching-based generative AI model (FM-Cast) for efficient and skillful probabilistic forecasting of the spatiotemporal evolution of stratospheric circulation in winter. Evaluated across 18 major SSW events (1998-2024), FM-Cast successfully forecasts the onset, intensity, and 3D morphology of the polar vortex up to 15 days in advance for most cases. Notably, it achieves long-range probabilistic forecast skill comparable to or exceeding leading operational NWP systems (ECMWF and CMA) while generating a 30-day forecast with 50-member ensemble, in just two minutes on a consumer GPU. Furthermore, using idealized "perfect troposphere" experiments, we uncover distinct predictability regimes: events driven by continuous wave forcing versus those governed by an initial trigger and subsequent stratospheric dynamical memory. This work establishes a computationally efficient paradigm for probabilistic stratospheric forecasting that simultaneously deepens our physical understanding of atmosphere-climate dynamics.

cs.LG↗
arXiv2026

Polarization-resolved measurement of forward volume spin waves by micro-focused Brillouin light scattering

We show how the micro-focused BLS signal of forward volume spin waves is formed and why it remains observable despite symmetry-based "suppression" expectations. A reciprocity-theorem based model with vectorial diffraction-limited focusing identifies the nonnegligible longitudinal focal-field component as the key element responsible for BLS sensitivity in the forward volume geometry. We further demonstrate that full polarization analysis, implemented through polarizer-analyzer maps of coherently excited spin waves, provides information beyond the conventional crossed polarizer-analyzer readout. In a BiYIG thin film, the measured maps exhibit Stokes/anti-Stokes polarization asymmetries and nontrivial patterns that stem from quadratic magneto-optical coupling terms. Fitting the data with a model including Voigt and Cotton-Mouton contributions yields an effective Cotton-Mouton constant and shows that the quadratic response is comparable to the linear Voigt contribution.

cond-mat.mes-hall↗
arXiv2026

ICL-EVADER: Zero-Query Black-Box Evasion Attacks on In-Context Learning and Their Defenses

In-context learning (ICL) has become a powerful, data-efficient paradigm for text classification using large language models. However, its robustness against realistic adversarial threats remains largely unexplored. We introduce ICL-Evader, a novel black-box evasion attack framework that operates under a highly practical zero-query threat model, requiring no access to model parameters, gradients, or query-based feedback during attack generation. We design three novel attacks, Fake Claim, Template, and Needle-in-a-Haystack, that exploit inherent limitations of LLMs in processing in-context prompts. Evaluated across sentiment analysis, toxicity, and illicit promotion tasks, our attacks significantly degrade classifier performance (e.g., achieving up to 95.3% attack success rate), drastically outperforming traditional NLP attacks which prove ineffective under the same constraints. To counter these vulnerabilities, we systematically investigate defense strategies and identify a joint defense recipe that effectively mitigates all attacks with minimal utility loss (<5% accuracy degradation). Finally, we translate our defensive insights into an automated tool that proactively fortifies standard ICL prompts against adversarial evasion. This work provides a comprehensive security assessment of ICL, revealing critical vulnerabilities and offering practical solutions for building more robust systems. Our source code and evaluation datasets are publicly available at: https://github.com/ChaseSecurity/ICL-Evader .

cs.CR↗
arXiv2026

SimBench: A Framework for Evaluating and Diagnosing LLM-Based Digital-Twin Generation for Multi-Physics Simulation

We introduce SimBench, a benchmark designed to evaluate the proficiency of simulator-oriented LLMs (S-LLMs) in generating digital twins (DTs) that can be used in simulators for virtual testing. Given a collection of S-LLMs, this benchmark ranks them according to their ability to produce high-quality DTs. We demonstrate this by comparing over 33 open- and closed-source S-LLMs. Using multi-turn interactions, SimBench employs an LLM-as-a-judge (J-LLM) that leverages both predefined rules and human-in-the-loop guidance to assign scores for the DTs generated by the S-LLM, thus providing a consistent and expert-inspired evaluation protocol. The J-LLM is specific to a simulator, and herein the proposed benchmarking approach is demonstrated in conjunction with the open-sourceChrono multi-physics simulator. Chrono provided the backdrop used to assess an S-LLM in relation to the latter's ability to create digital twins for multibody dynamics, finite element analysis, vehicle dynamics, robotic dynamics, and sensor simulations. The proposed benchmarking principle is broadly applicable and enables the assessment of an S-LLM's ability to generate digital twins for other simulation packages, e.g., ANSYS, ABAQUS, OpenFOAM, StarCCM+, IsaacSim, and pyBullet.

cs.AI↗