arXiv subjects

Pei Zhang

Publications and source records attributed to Pei Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2025

Nonlocal Nonlinear Control of Photonic Spin Hall Effect in Strongly Interacting Rydberg Media

We present a theoretical study demonstrating enhanced tunability of the photonic spin Hall effect (PSHE) using a strongly interacting Rydberg atomic medium under electromagnetically induced transparency (EIT) conditions. In contrast to conventional approaches that rely on static refractiveindex profiles or metamaterials, here the PSHE is controlled via a nonlocal third-order nonlinear susceptibility arising from long range Rydberg-Rydberg interactions. We show that this nonlocal nonlinearity enables dynamic modulation of spin-dependent light trajectories, amplifying the normally weak PSHE into a readily observable and adjustable effect. These results pave the way for new capabilities in photonic information processing and sensing. In particular, an adjustable PSHE may enable beam steering based on photon spin, improve the sensitivity of precision measurements, and support photonic devices whose functionality can be reconfigured in real time.

quant-ph↗

arXiv2025

Sampling-Efficient Test-Time Scaling: Self-Estimating the Best-of-N Sampling in Early Decoding

Test-time scaling enhances large language model performance by allocating additional compute resources during inference. Best-of-N (BoN) sampling serves as a common sampling-based scaling technique, broadening the search space in parallel to find better solutions from the model distribution. However, its cost-performance trade-off is still underexplored. Two main challenges limit the efficiency of BoN sampling: (1) Generating N full samples consumes substantial GPU memory, reducing inference capacity under limited resources. (2) Reward models add extra memory and latency overhead, and training strong reward models introduces potential training data costs. Although some studies have explored efficiency improvements, none have addressed both challenges at once. To address this gap, we propose Self-Truncation Best-of-N (ST-BoN), a decoding method that avoids fully generating all N samples and eliminates the need for reward models. It leverages early sampling consistency in the model's internal states to identify the most promising path and truncate suboptimal ones. In terms of cost, ST-BoN reduces dynamic GPU memory usage by over 80% and inference latency by 50%. In terms of cost-performance trade-off, ST-BoN achieves the same performance as Full-BoN while saving computational cost by 70%-80%, and under the same cost, it can improve accuracy by 3-4 points.

cs.CL↗

arXiv2025

PolyMath: Evaluating Mathematical Reasoning in Multilingual Contexts

In this paper, we introduce PolyMath, a multilingual mathematical reasoning benchmark covering 18 languages and 4 easy-to-hard difficulty levels. Our benchmark ensures difficulty comprehensiveness, language diversity, and high-quality translation, making it a highly discriminative multilingual mathematical benchmark in the era of reasoning LLMs. We conduct a comprehensive evaluation for advanced LLMs and find that even Qwen-3-235B-A22B-Thinking and Gemini-2.5-pro, achieve only 54.6 and 52.2 benchmark scores, with about 40% accuracy under the highest level From a language perspective, our benchmark reveals several key challenges of LLMs in multilingual reasoning: (1) Reasoning performance varies widely across languages for current LLMs; (2) Input-output language consistency is low in reasoning LLMs and may be correlated with performance; (3) The thinking length differs significantly by language for current LLMs. Additionally, we demonstrate that controlling the output language in the instructions has the potential to affect reasoning performance, especially for some low-resource languages, suggesting a promising direction for improving multilingual capabilities in LLMs.

cs.CL↗

arXiv2025

Intelligent Sampling of Extreme-Scale Turbulence Datasets for Accurate and Efficient Spatiotemporal Model Training

With the end of Moore's law and Dennard scaling, efficient training increasingly requires rethinking data volume. Can we train better models with significantly less data via intelligent subsampling? To explore this, we develop SICKLE, a sparse intelligent curation framework for efficient learning, featuring a novel maximum entropy (MaxEnt) sampling approach, scalable training, and energy benchmarking. We compare MaxEnt with random and phase-space sampling on large direct numerical simulation (DNS) datasets of turbulence. Evaluating SICKLE at scale on Frontier, we show that subsampling as a preprocessing step can, in many cases, improve model accuracy and substantially lower energy consumption, with observed reductions of up to 38x.

cs.LG↗

arXiv2025

Qwen3Guard Technical Report

As large language models (LLMs) become more capable and widely used, ensuring the safety of their outputs is increasingly critical. Existing guardrail models, though useful in static evaluation settings, face two major limitations in real-world applications: (1) they typically output only binary "safe/unsafe" labels, which can be interpreted inconsistently across diverse safety policies, rendering them incapable of accommodating varying safety tolerances across domains; and (2) they require complete model outputs before performing safety checks, making them fundamentally incompatible with streaming LLM inference, thereby preventing timely intervention during generation and increasing exposure to harmful partial outputs. To address these challenges, we present Qwen3Guard, a series of multilingual safety guardrail models with two specialized variants: Generative Qwen3Guard, which casts safety classification as an instruction-following task to enable fine-grained tri-class judgments (safe, controversial, unsafe); and Stream Qwen3Guard, which introduces a token-level classification head for real-time safety monitoring during incremental text generation. Both variants are available in three sizes (0.6B, 4B, and 8B parameters) and support up to 119 languages and dialects, providing comprehensive, scalable, and low-latency safety moderation for global LLM deployments. Evaluated across English, Chinese, and multilingual benchmarks, Qwen3Guard achieves state-of-the-art performance in both prompt and response safety classification. All models are released under the Apache 2.0 license for public use.

cs.CL↗

arXiv2025

FLAMMABLE: A Multi-Model Federated Learning Framework with Multi-Model Engagement and Adaptive Batch Sizes

Multi-Model Federated Learning (MMFL) is an emerging direction in Federated Learning (FL) where multiple models are trained in parallel, generally on various datasets. Optimizing the models' accuracies and training times in the MMFL setting requires adapting to data and system heterogeneity across clients as in single-model FL; these challenges are amplified in the MMFL setting due to additional heterogeneity across models. Neither existing solutions nor naïve extensions of single-model FL frameworks efficiently address these challenges. To bridge this gap, we propose FLAMMABLE, a comprehensive MMFL training framework. FLAMMABLE optimizes model training by intelligently adapting client batch sizes while engaging them to train multiple carefully chosen models, depending on their system capabilities, in each training round. To evaluate FLAMMABLE, we develop the first benchmark platform for the MMFL setting, which may enable future reproducible MMFL research. Extensive evaluations on multiple datasets and models show that FLAMMABLE boosts the MMFL time-to-accuracy performance by 1.1$\sim$10.0$\times$ while improving the final model accuracy by 1.3$\sim$5.4\% compared to several known baselines.

cs.DC↗

arXiv2025

Vim-F: Visual State Space Model Benefiting from Learning in the Frequency Domain

In recent years, State Space Models (SSMs) with efficient hardware-aware designs, known as the Mamba deep learning models, have made significant progress in modeling long sequences such as language understanding. Therefore, building efficient and general-purpose visual backbones based on SSMs is a promising direction. Compared to traditional convolutional neural networks (CNNs) and Vision Transformers (ViTs), the performance of Vision Mamba (ViM) methods is not yet fully competitive. To enable SSMs to process image data, ViMs typically flatten 2D images into 1D sequences, inevitably ignoring some 2D local dependencies, thereby weakening the model's ability to interpret spatial relationships from a global perspective. We use Fast Fourier Transform (FFT) to obtain the spectrum of the feature map and add it to the original feature map, enabling ViM to model a unified visual representation in both frequency and spatial domains. The introduction of frequency domain information enables ViM to have a global receptive field during scanning. We propose a novel model called Vim-F, which employs pure Mamba encoders and scans in both the frequency and spatial domains. Moreover, we question the necessity of position embedding in ViM and remove it accordingly in Vim-F, which helps to fully utilize the efficient long-sequence modeling capability of ViM. Finally, we redesign a patch embedding for Vim-F, leveraging a convolutional stem to capture more local correlations, further improving the performance of Vim-F. Code is available at: https://github.com/yws-wxs/Vim-F.

cs.CV↗

arXiv2025

PART: Progressive Alignment Representation Training for Multilingual Speech-To-Text with LLMs

Large language models (LLMs) have expanded from text to speech, giving rise to Speech Large Models (SLMs) that support recognition, translation, and synthesis. A key challenge is aligning speech and text representations, which becomes harder in multilingual settings. Existing methods often freeze LLM parameters and train encoders on multilingual data, but this forces cross-language convergence and limits performance. We introduce Progressive Alignment Representation Training (PART), a multi-stage and multi-task framework that separates within-language from cross-language alignment. During cross-language training, LLM parameters are dynamically activated, and text-based tasks are later introduced to enhance multilingual understanding. Experiments on CommonVoice 15, Fleurs, Wenetspeech, and CoVoST2 show that PART surpasses conventional approaches, with analysis confirming its ability to balance language-specific distinctions and cross-language generalization. These results demonstrate PART's effectiveness and generality for multilingual speech modality alignment.

cs.CL↗

arXiv2025

Estimating the Heritability of Longitudinal Rate-of-Change: Genetic Insights into PSA Velocity in Prostate Cancer-Free Individuals

Serum prostate-specific antigen (PSA) is widely used for prostate cancer screening. While the genetics of PSA levels has been studied to enhance screening accuracy, the genetic basis of PSA velocity, the rate of PSA change over time, remains unclear. The Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial, a large, randomized study with longitudinal PSA data (15,260 cancer-free males, averaging 5.34 samples per subject) and genome-wide genotype data, provides a unique opportunity to estimate PSA velocity heritability. We developed a mixed model to jointly estimate heritability of PSA levels at age 54 and PSA velocity. To accommodate the large dataset, we implemented two efficient computational approaches: a partitioning and meta-analysis strategy using average information restricted maximum likelihood (AI-REML), and a fast restricted Haseman-Elston (REHE) regression method. Simulations showed that both methods yield unbiased estimates of both heritability metrics, with AI-REML providing smaller variability in the estimation of velocity heritability than REHE. Applying AI-REML to PLCO data, we estimated heritability at 0.32 (s.e. = 0.07) for baseline PSA and 0.45 (s.e. = 0.18) for PSA velocity. These findings reveal a substantial genetic contribution to PSA velocity, supporting future genome-wide studies to identify variants affecting PSA dynamics and improve PSA-based screening.

stat.AP↗

arXiv2025

Qwen3-Omni Technical Report

We present Qwen3-Omni, a single multimodal model that, for the first time, maintains state-of-the-art performance across text, image, audio, and video without any degradation relative to single-modal counterparts. Qwen3-Omni matches the performance of same-sized single-modal models within the Qwen series and excels particularly on audio tasks. Across 36 audio and audio-visual benchmarks, Qwen3-Omni achieves open-source SOTA on 32 benchmarks and overall SOTA on 22, outperforming strong closed-source models such as Gemini-2.5-Pro, Seed-ASR, and GPT-4o-Transcribe. Qwen3-Omni adopts a Thinker-Talker MoE architecture that unifies perception and generation across text, images, audio, and video, yielding fluent text and natural real-time speech. It supports text interaction in 119 languages, speech understanding in 19 languages, and speech generation in 10 languages. To reduce first-packet latency in streaming synthesis, Talker autoregressively predicts discrete speech codecs using a multi-codebook scheme. Leveraging the representational capacity of these codebooks, we replace computationally intensive block-wise diffusion with a lightweight causal ConvNet, enabling streaming from the first codec frame. In cold-start settings, Qwen3-Omni achieves a theoretical end-to-end first-packet latency of 234 ms. To further strengthen multimodal reasoning, we introduce a Thinking model that explicitly reasons over inputs from any modality. Since the research community currently lacks a general-purpose audio captioning model, we fine-tuned Qwen3-Omni-30B-A3B to obtain Qwen3-Omni-30B-A3B-Captioner, which produces detailed, low-hallucination captions for arbitrary audio inputs. Qwen3-Omni-30B-A3B, Qwen3-Omni-30B-A3B-Thinking, and Qwen3-Omni-30B-A3B-Captioner are publicly released under the Apache 2.0 license.

cs.CL↗

arXiv2025

Translationese-index: Using Likelihood Ratios for Graded and Generalizable Measurement of Translationese

Translationese refers to linguistic properties that usually occur in translated texts. Previous works study translationese by framing it as a binary classification between original texts and translated texts. In this paper, we argue that translationese should be graded instead of binary and propose the first measure for translationese -- the translationese-index (T-index), computed from the likelihood ratios of two contrastively fine-tuned language models (LMs). We use synthesized translations and translations in the wild to evaluate T-index's generalizability in cross-domain settings and its validity against human judgments. Our results show that T-index can generalize to unseen genres, authors, and language pairs. Moreover, T-index computed using two 0.5B LMs fine-tuned on only 1-5k pairs of synthetic data can effectively capture translationese, as demonstrated by alignment with human pointwise ratings and pairwise judgments. Additionally, the correlation between T-index and existing machine translation (MT) quality estimation (QE) metrics such as BLEU and COMET is low, suggesting that T-index is not covered by these metrics and can serve as a complementary metric in MT QE.

cs.CL↗

arXiv2025

Photonic Spin Hall Effect using bilayer Graphene in Nano Optomechanical Cavities

We propose a theoretical model to obtain the photonic spin Hall effect (SHE) in an optomechanical nanocavity using a graphene bilayer as the intracavity medium. In our model, the pump and probe fields coherently drive the first mirror, whereas the second mirror has mechanical oscillation due to the radiation pressure. We show that the right- and left-circular polarization components of the Gaussian probe field striking at an arbitrary incident angle become spatially separate along a direction orthogonal to the plane of incidence. Photonic SHE can be coherently controlled by adjusting the optomechanical interaction, cavity field and G-mode phonon coupling, as well as G-mode phonon and electronic state interaction. The findings of photonic SHE are equally valid for standard optomechanical systems in the absence of cavity field and G-mode phonon coupling and electronic state interaction. The cavity field and G-mode phonon coupling broadened the detuning range of the probe field to observe the dominant photonic SHE. Adding G-mode phonon and electronic state interaction generates enhanced photonic SHE at three different probe field detunings due to optomechanical-induced transparency being split into three windows. We show that asymmetric photonic SHE can be controlled through cavity field and G-mode phonon coupling and G-mode phonon and electronic state interaction when probe field detuning is non-zero. The photonic SHE in bilayer graphene integrated with an optomechanical cavity may enable further studies of spin-dependent photonic effects and quantum sensing applications.

quant-ph↗

arXiv2025

CultureSynth: A Hierarchical Taxonomy-Guided and Retrieval-Augmented Framework for Cultural Question-Answer Synthesis

Cultural competence, defined as the ability to understand and adapt to multicultural contexts, is increasingly vital for large language models (LLMs) in global environments. While several cultural benchmarks exist to assess LLMs' cultural competence, current evaluations suffer from fragmented taxonomies, domain specificity, and heavy reliance on manual data annotation. To address these limitations, we introduce CultureSynth, a novel framework comprising (1) a comprehensive hierarchical multilingual cultural taxonomy covering 12 primary and 130 secondary topics, and (2) a Retrieval-Augmented Generation (RAG)-based methodology leveraging factual knowledge to synthesize culturally relevant question-answer pairs. The CultureSynth-7 synthetic benchmark contains 19,360 entries and 4,149 manually verified entries across 7 languages. Evaluation of 14 prevalent LLMs of different sizes reveals clear performance stratification led by ChatGPT-4o-Latest and Qwen2.5-72B-Instruct. The results demonstrate that a 3B-parameter threshold is necessary for achieving basic cultural competence, models display varying architectural biases in knowledge processing, and significant geographic disparities exist across models. We believe that CultureSynth offers a scalable framework for developing culturally aware AI systems while reducing reliance on manual annotation\footnote{Benchmark is available at https://github.com/Eyr3/CultureSynth.}.

cs.CL↗

arXiv2025

Human Body Weight Estimation Through Music-Induced Bed Vibrations

Rapid and accurate body weight estimation is critical in emergency medical care, as it directly influences treatment decisions, such as drug dosing, defibrillation energy selection, and fluid resuscitation. Traditional methods such as stand-on scales, length-based tapes, or transfer-based weighing scales are often impractical for immobilized patients, inaccurate, or labor-intensive and time-consuming. This paper introduces MelodyBedScale, a non-intrusive and rapid on-bed weight estimation system that leverages bed vibration induced by music. The core insight is that body weight affects the vibration transfer function of the bed-body system, which is captured using vibration sensors placed on opposite sides of the bed. First, we identify weight-sensitive frequency bands and compose clinically acceptable soft, natural music with high signal energy in these frequency bands. This music is then played through a speaker mounted on the bed to induce bed vibrations. Additionally, to efficiently capture the complex weight-vibration relationship with limited data and enhance generalizability to unseen individuals and weights, we theoretically analyze the weight-vibration relationship and integrate the results into the activation functions of the neural network for physics-informed weight regression. We evaluated MelodyBedScale on both wooden and steel beds across 11 participants, achieving a mean absolute error of up to 1.55 kg.

eess.SP↗

arXiv2025

Topological robustness of classical and quantum optical skyrmions in atmospheric turbulence

The degradation of classical and quantum structured light induced by complex media constitutes a critical barrier to its practical implementation in a range of applications, from communication and energy transport to imaging and sensing. Atmospheric turbulence is an exemplary case due to its complex phase structure and dynamic variations, driving the need to find invariances in light. Here we construct classical and quantum optical skyrmions and pass them through experimentally simulated atmospheric turbulence, revealing the embedded topological resilience of their structure. In the quantum realm, we show that while skyrmions undergo diminished entanglement, their topological characteristics maintain stable. This is paralleled classically, where the vectorial structure is scrambled by the medium yet the skyrmion remains stable by virtue of its intrinsic topological protection mechanism. Our experimental results are supported by rigorous analytical and numerical modelling, validating that the quantum-classical equivalence of the topological behaviour is due to the non-separability of the states and one-sided nature of the channel. Our work blurs the classical-quantum divide in the context of topology and opens a new path to information resilience in noisy channels, such as terrestrial and satellite-to-ground communication networks.

physics.optics↗

arXiv2025

Goos-Hanchen Shift with a rotating atomic superfluid in a ring

We investigate the Goos Hanchen shift of the transmitted probe and show that optomechanical interference in a ring Bose Einstein condensate provides a sensitive, rotation tunable beam shift response at ultralow optical powers. Without quantized circulation, conventional optomechanically induced transparency produces a strictly positive and bounded shift whose magnitude is governed primarily by cooperativity; small detuning offsets can introduce a weak, transient sign change consistent with a Fano type asymmetry, but the overall response remains limited. With circulation, the Bragg scattered mechanical side modes split, yielding a double transparency dispersion with steep dispersive flanks that strongly amplify the phase derivative and bias its sign. In this regime, the peak shift grows monotonically with control field strength, reflecting enhanced linearized coupling and increased transmission across the angular scan. At fixed power, the detuning dependence is decisive: the shift is maximized at the red sideband condition and diminishes away from resonance, tracking how effectively the scan samples the rotation split dispersive flanks. Increasing the winding number broadens the central absorption and steepens accessible phase gradients, further boosting the attainable shift. The protocol remains minimally invasive under experimentally realistic conditions, and interatomic interactions have a negligible influence on the transmission features that set the phase slope. These results identify circulation, control power, and cavity detuning as practical knobs for in situ control of the Goos Hanchen shift, enabling interferometric beam steering and phase gradient metrology in hybrid atom optomechanical platforms.

cond-mat.quant-gas↗

arXiv2025

Normal Mode Splitting and Force Sensing in Cavity Magnomechanical System

In this study, we investigate the dynamics of system composed of a single cavity consisting of an optical parametric amplifier (OPA) and a YIG sphere influenced by a bias magnetic field. This bias field leads to magnetostrictive effects on magnon modes that induces phonons. We investigate the position fluctuation spectrum and the output field spectrum, finding that at G =0, the system displays a single peak, indicative of weak coupling between the optical and phononic modes. As G increases (e.g., G =0.1 kappa_a, 0.2 kappa_a, 0.4 kappa_a, we observe a transition to double peak, which reflects stronger coupling in the vicinity of cavity along with phonon modes that leads to normal mode splitting (NMS) in cavity magnomechanic system. Furthermore, we examine that the OPA amplifies the Y quadrature while squeezing the X quadrature of the output field spectrum. This sensitive behavior results in a more pronounced splitting in the Y quadrature spectra compared to the X quadrature. Our findings emphasize the essential role of the OPA in adjusting the interaction strength between the optical and phononic modes as well as underscore the importance of quadrature analysis in characterizing the system's response. NMS mechanism open avenues for advanced applications in quantum sensing and information processing, highlighting the potential for tunable devices in emerging quantum technologies.

physics.optics↗

arXiv2025

Locate-and-Focus: Enhancing Terminology Translation in Speech Language Models

Direct speech translation (ST) has garnered increasing attention nowadays, yet the accurate translation of terminology within utterances remains a great challenge. In this regard, current studies mainly concentrate on leveraging various translation knowledge into ST models. However, these methods often struggle with interference from irrelevant noise and can not fully utilize the translation knowledge. To address these issues, in this paper, we propose a novel Locate-and-Focus method for terminology translation. It first effectively locates the speech clips containing terminologies within the utterance to construct translation knowledge, minimizing irrelevant information for the ST model. Subsequently, it associates the translation knowledge with the utterance and hypothesis from both audio and textual modalities, allowing the ST model to better focus on translation knowledge during translation. Experimental results across various datasets demonstrate that our method effectively locates terminologies within utterances and enhances the success rate of terminology translation, while maintaining robust general translation performance.

cs.CL↗