arXiv subjects

Yunfei Xu

Publications and source records attributed to Yunfei Xu.

At least 19 recordsLinked to original sources

arXiv2026

VideoNorms: Benchmarking Cultural Awareness of Video Language Models

As Video Large Language Models (VideoLLMs) are deployed globally, it is important to assess their ability to reason across cultural contexts. To advance cultural norm awareness evaluation in VideoLLMs, we introduce VideoNorms, a dataset of cultural norm annotations from popular US and Chinese TV shows annotated with adherence or violation labels and (non-)verbal evidence. Through a human-AI collaboration framework, each item was first annotated by a large VideoLLM, and then reviewed by at least three trained monocultural annotators with significant lived experience in the target culture, resulting in a dataset of over 3,000 human judgments. Human verification showed disparity in US and Chinese norm extraction performance, cautioning against fully automatic approaches cultures under-represented in training data. Hierarchical linear modeling analysis of $7$ open-weight VideoLLMs' performance revealed that: 1) models perform worse in Chinese compared to US, particularly for norm adherence prediction; 2) models have more difficulty in providing non-verbal evidence compared to verbal evidence for norm adherence/violation predictions. Ablation studies confirm video modality is indeed necessary for accurate performance, and scaling model size does not yield classification score improvements. Our findings and data contribute to culturally grounded video model training and evaluation.

cs.CV↗

arXiv2026

PRISA: Proactive Infrastructure LiDAR Framework for Intersection Safety Assessment

Urban intersections are among the most hazardous locations in road networks, posing significant risks to vehicles and vulnerable road users (VRUs) such as pedestrians and cyclists. The complexity of multi-agent interactions demands continuous, real-time monitoring systems capable of anticipating conflicts before they escalate into crashes. We present PRISA, a modular infrastructure LiDAR framework leveraging privacy-preserving, low-light-robust roadside sensors for long-term traffic observation and real-time risk detection at the edge. The framework comprises two core components: a sensing and perception layer and a plug-and-play risk assessment module. The latter automatically curates site-specific training data from accumulated perception outputs to train a trajectory prediction model without manual annotation. It then deploys the trained model for continuous motion forecasting and dual surrogate safety evaluation, using Time-to-Collision (TTC) for longitudinal conflicts and Predicted Post-Encroachment Time (PPET) for crossing and VRU-involved interactions. PRISA is evaluated on the public R-LiViT dataset and deployed on an NVIDIA Jetson AGX Thor at a live signalized intersection in Chattanooga, Tennessee. PPET-based assessment operates at 194~ms end-to-end latency over a 2.4-second predictive horizon, with TTC-based detection and perception remaining within real-time constraints, demonstrating practical feasibility for proactive multi-agent intersection safety monitoring.

cs.LG↗

arXiv2026

hia-gat: A Heterogeneous Interaction-Aware Graph Attention Network For Frame-Level Traffic Conflict Risk Prediction On Freeways

This paper formulates frame-level freeway risk assessment as a multi-agent scene graph-level binary classification problem, where each video or trajectory frame is labeled risky if any TTC- or PET-based conflict violates a specified severity threshold. We construct a relation-aware graph per frame with vehicles as nodes and two interaction types as edges: same-lane (longitudinal) and adjacent-lane (lateral), augmented with physics-informed edge features aligned to rear-end and lane-change conflict mechanisms. Building on a structured benchmarking suite of non-graph models and graph baselines, we propose HIA-GAT, a dual-stream heterogeneous graph attention network that processes longitudinal and lateral interactions through dedicated attention pathways and fuses them via a conflict-type-aware gating mechanism with event-level gate supervision derived from SSM conflict attribution. Experiments on the NGSIM I-80 and US-101 freeway datasets across nine TTC and PET threshold configurations show that HIA-GAT achieves the best average risk-ranking performance (AUC 0.835 on I-80 and 0.867 on US-101), with the largest gains on PET-only (lane-change) settings where relational structure is essential. Beyond accuracy, the learned gate provides interpretable per-vehicle attribution of dominant conflict type, supporting actionable, real-time freeway safety monitoring. We show that graph structure is critical for modeling lateral conflict risk, while longitudinal risk can often be captured by non-relational aggregation.

cs.LG↗

arXiv2026

M-EPDet: Real-Time Real-Bogus Classification and Transient Candidate Judgement for the EP-WXT Pipeline via Multi-Modal Data

The Wide-field X-ray Telescope (WXT) onboard the Einstein Probe (EP) produces a large post-detection candidate stream in which genuine astrophysical sources coexist with instrumental artifacts and Cosmic Ray events. We present M-EPDet, a three-step post-detection framework for real-time candidate vetting in EP-WXT lobster-eye Micro-pore Optics (MPO) data. The framework combines a ResNet-based Arm filter, a dual-branch temporal-spectral Cosmic Ray filter, and a background-aware Bayesian Blocks module for single-exposure variability screening. Using on-orbit EP-WXT observations, we report decoupled metrics for the cascading system. M-EPDet achieves a Real-Bogus Recall of 98.31\% ($98.53\% \times 99.78\%$) for genuine astrophysical sources, together with rejection rates of 92.99\% for instrumental artifacts and 98.18\% for Cosmic Ray events. In the final step, the Bayesian Blocks module flags 0.75\% of the post-filtration observations, corresponding to a 99.25\% reduction in candidate volume. The system is deployed in the EP-WXT pipeline as a lightweight real-time service, reducing the manual-inspection burden in candidate vetting.

astro-ph.IM↗

arXiv2026

FedSPC: Shared Parameter Correction for Personalized Federated Learning

Personalized federated learning (PFL) is one of the important approaches in federated learning for addressing statistical heterogeneity while enabling client-specific adaptation. Many PFL methods split the model into shared and personalized parameters, which are jointly trained on each client. However, this creates an optimization issue: shared parameters are updated by clients optimizing different local objectives, which can lead to inconsistent shared updates and weaken the shared representation. To address this problem, we propose Federated Shared Parameter Correction (FedSPC), a modular correction method for PFL. FedSPC applies control-variate correction only to the shared parameters of a given PFL method, while leaving personalized parameters unchanged. It can be integrated into three common PFL settings: shared feature extractors, shared classifiers, and fully shared models with local regularization. Experiments on CIFAR-100 and Tiny-ImageNet with ViT, ResNet-34, and VGG-11 show that FedSPC improves performance across representative PFL methods, including FedPer, FedRep, FedBABU, LG-FedAvg, and Ditto.

cs.LG↗

arXiv2026

EP250827b/SN 2025wkm: An X-ray Flash-Supernova Powered by a Central Engine and Circumstellar Interaction

We present the discovery of EP250827b/SN 2025wkm, an X-ray Flash (XRF) discovered by the Einstein Probe (EP), accompanied by a broad-line Type Ic supernova (SN Ic-BL) at $z = 0.1194$. EP250827b possesses a prompt X-ray luminosity of $\sim 10^{45} \, \rm{erg \, s^{-1}}$, lasts over 1000 seconds, and has a peak energy $E_{\rm{p}} < 1.5$ keV at 90\% confidence. SN 2025wkm possesses a double-peaked optical light curve (LC), though its bolometric luminosity plateaus after its initial peak for $\sim 20$ days, consistent with a central engine injecting additional energy into the explosion. Its spectrum transitions from a blue to red continuum with clear blueshifted broad absorption features consistent with a SN Ic-BL classification. We do not detect any transient radio emission and rule out the existence of an on-axis, energetic jet $\gtrsim 10^{50}~$erg assuming a typical LGRB circumburst constant density ($n \approx 10^{-3}$--$10^{-1}~{\rm cm}^{-3}$) and microphysical parameters ($ε_{\rm e} = 0.1$ and $ε_{\rm B} = 0.01$). In the model we invoke, the collapse gives rise to a long-lived magnetar, potentially surrounded by an accretion disk. Magnetically--driven winds from the magnetar and the disk mix together and break out with a velocity $\sim 0.35c$ and interact with an extended circumstellar medium with radius $\sim 10^{13}$ cm, generating X-ray breakout emission through non-thermal free-free processes. The disk outflows and magnetar winds power blackbody photospheric emission as they cool adiabatically and thermalize, producing the first SN peak. The spin-down luminosity of the magnetar and radioactive decay of $^{56}$Ni powers the late-time emission. We end by discussing the landscape of XRF-SNe within the context of EP's recent discoveries.

astro-ph.HE↗

arXiv2026

Deep learning-based astronomical multimodal data fusion: A comprehensive review

With the rapid advancements in observational technologies and the widespread implementation of large-scale sky surveys, diverse electromagnetic wave data (e.g., optical and infrared) and non-electromagnetic wave data (e.g., gravitational waves) have become increasingly accessible. Astronomy has thus entered an unprecedented era of data abundance and complexity. Astronomers have long relied on unimodal data analysis to perceive the universe, but these efforts often provide only limited insights when confronted with the current massive and heterogeneous astronomical data. In this context, multimodal data fusion (MDF), as an emerging method, provides new opportunities to enhance the value of astronomical data and deepening the understanding of the universe by integrating information from different modalities. Recent progress in artificial intelligence (AI), particularly in deep learning (DL), has greatly accelerated the development of multimodal research in astronomy. Therefore, a timely review of this field is essential. This paper begins by discussing the motivation and necessity of astronomical MDF, followed by an overview of astronomical data sources and major data modalities. It then introduces representative DL models commonly used in astronomical multimodal studies, the general fusion process as well as various fusion strategies, emphasizing their characteristics, applicability, advantages, and limitations. Subsequently, the paper surveys existing astronomical multimodal studies and datasets. Finally, the discussion section synthesizes key findings, identifies potential challenges, and suggests promising directions for future research. By offering a structured overview and critical analysis, this review aims to inspire and guide researchers engaged in DL-based MDF in astronomy.

astro-ph.IM↗

arXiv2026

Vclip: Face-based Speaker Generation by Face-voice Association Learning

This paper discusses the task of face-based speech synthesis, a kind of personalized speech synthesis where the synthesized voices are constrained to perceptually match with a reference face image. Due to the lack of TTS-quality audio-visual corpora, previous approaches suffer from either low synthesis quality or domain mismatch induced by a knowledge transfer scheme. This paper proposes a new approach called Vclip that utilizes the facial-semantic knowledge of the CLIP encoder on noisy audio-visual data to learn the association between face and voice efficiently, achieving 89.63% cross-modal verification AUC score on Voxceleb testset. The proposed method then uses a retrieval-based strategy, combined with GMM-based speaker generation module for a downstream TTS system, to produce probable target speakers given reference images. Experimental results demonstrate that the proposed Vclip system in conjunction with the retrieval step can bridge the gap between face and voice features for face-based speech synthesis. And using the feedback information distilled from downstream TTS helps to synthesize voices that match closely with reference faces. Demos available at sos1sos2sixteen.github.io/vclip.

eess.AS↗

arXiv2025

Adversarial Attacks and Robust Defenses in Speaker Embedding based Zero-Shot Text-to-Speech System

Speaker embedding based zero-shot Text-to-Speech (TTS) systems enable high-quality speech synthesis for unseen speakers using minimal data. However, these systems are vulnerable to adversarial attacks, where an attacker introduces imperceptible perturbations to the original speaker's audio waveform, leading to synthesized speech sounds like another person. This vulnerability poses significant security risks, including speaker identity spoofing and unauthorized voice manipulation. This paper investigates two primary defense strategies to address these threats: adversarial training and adversarial purification. Adversarial training enhances the model's robustness by integrating adversarial examples during the training process, thereby improving resistance to such attacks. Adversarial purification, on the other hand, employs diffusion probabilistic models to revert adversarially perturbed audio to its clean form. Experimental results demonstrate that these defense mechanisms can significantly reduce the impact of adversarial perturbations, enhancing the security and reliability of speaker embedding based zero-shot TTS systems in adversarial environments.

eess.AS↗

arXiv2025

Photometric redshift estimation for emission line galaxies of DESI Legacy Imaging Surveys by CNN-MLP

Emission Line Galaxies (ELGs) are crucial for cosmological studies, particularly in understanding the large-scale structure of the Universe and the role of dark energy. ELGs form an essential component of the target catalogue for the Dark Energy Spectroscopic Instrument (DESI), a major astronomical survey. However, the accurate selection of ELGs for such surveys is challenging due to the inherent uncertainties in determining their redshifts with photometric data. In order to improve the accuracy of photometric redshift estimation for ELGs, we propose a novel approach CNN-MLP that combines Convolutional Neural Networks (CNNs) with Multilayer Perceptrons (MLPs). This approach integrates both images and photometric data derived from the DESI Legacy Imaging Surveys Data Release 10. By leveraging the complementary strengths of CNNs (for image data processing) and MLPs (for photometric feature integration), the CNN-MLP model achieves a $σ_{\mathrm{NMAD}}$ (normalised median absolute deviation) of 0.0140 and an outlier fraction of 2.57%. Compared to other models, CNN-MLP demonstrates a significant improvement in the accuracy of ELG photometric redshift estimation, which directly benefits the target selection process for DESI. In addition, we explore the photometric redshifts of different galaxy types (Starforming, Starburst, AGN, Broadline). Furthermore, this approach will contribute to more reliable photometric redshift estimation in ongoing and future large-scale sky surveys (e.g. LSST, CSST, Euclid), enhancing the overall efficiency of cosmological research and galaxy surveys.

astro-ph.IM↗

arXiv2025

FALCO: a Foundation model of Astronomical Light Curves for time dOmain astronomy

Time-domain surveys have advanced astronomical research by revealing diverse variable phenomena, from stellar flares to transient events. The scale and complexity of survey data, along with the demand for rapid classification, present significant challenges for analysis. While machine learning offers solutions, most existing models are tailored to single tasks, struggle to generalize, and depend heavily on large, accurately labeled datasets. We introduce FALCO, a foundation model for astronomical light curve analysis in time-domain astronomy. This work presents the initial version of FALCO trained via self-supervised learning on unlabeled Kepler light curves using a Transformer-based architecture. The model has been evaluated on three distinct tasks and demonstrates strong generalization: achieving 95 percent accuracy in stellar variability classification across eight classes, an overall RMSE of 0.1305 dex in surface gravity estimation (notably improved to below 0.08 dex when log g is less than 1, and approximately 0.02 dex near log g equals 3), and 87 percent precision in flare identification. These results highlight the model's versatility and ability to learn generalizable representations from light curves, enabling straightforward adaptation to diverse tasks. We further analyze the impact of model scaling and sequence length, finding performance improves with larger models and longer input sequences. We also apply FALCO to derive surface gravity (log g) measurements for 179,732 Kepler stars from their light curves.

astro-ph.IM↗

arXiv2024

SongGLM: Lyric-to-Melody Generation with 2D Alignment Encoding and Multi-Task Pre-Training

Lyric-to-melody generation aims to automatically create melodies based on given lyrics, requiring the capture of complex and subtle correlations between them. However, previous works usually suffer from two main challenges: 1) lyric-melody alignment modeling, which is often simplified to one-syllable/word-to-one-note alignment, while others have the problem of low alignment accuracy; 2) lyric-melody harmony modeling, which usually relies heavily on intermediates or strict rules, limiting model's capabilities and generative diversity. In this paper, we propose SongGLM, a lyric-to-melody generation system that leverages 2D alignment encoding and multi-task pre-training based on the General Language Model (GLM) to guarantee the alignment and harmony between lyrics and melodies. Specifically, 1) we introduce a unified symbolic song representation for lyrics and melodies with word-level and phrase-level (2D) alignment encoding to capture the lyric-melody alignment; 2) we design a multi-task pre-training framework with hierarchical blank infilling objectives (n-gram, phrase, and long span), and incorporate lyric-melody relationships into the extraction of harmonized n-grams to ensure the lyric-melody harmony. We also construct a large-scale lyric-melody paired dataset comprising over 200,000 English song pieces for pre-training and fine-tuning. The objective and subjective results indicate that SongGLM can generate melodies from lyrics with significant improvements in both alignment and harmony, outperforming all the previous baseline methods.

eess.AS↗

arXiv2024

LLM-DER:A Named Entity Recognition Method Based on Large Language Models for Chinese Coal Chemical Domain

Domain-specific Named Entity Recognition (NER), whose goal is to recognize domain-specific entities and their categories, provides an important support for constructing domain knowledge graphs. Currently, deep learning-based methods are widely used and effective in NER tasks, but due to the reliance on large-scale labeled data. As a result, the scarcity of labeled data in a specific domain will limit its application.Therefore, many researches started to introduce few-shot methods and achieved some results. However, the entity structures in specific domains are often complex, and the current few-shot methods are difficult to adapt to NER tasks with complex features.Taking the Chinese coal chemical industry domain as an example,there exists a complex structure of multiple entities sharing a single entity, as well as multiple relationships for the same pair of entities, which affects the NER task under the sample less condition.In this paper, we propose a Large Language Models (LLMs)-based entity recognition framework LLM-DER for the domain-specific entity recognition problem in Chinese, which enriches the entity information by generating a list of relationships containing entity types through LLMs, and designing a plausibility and consistency evaluation method to remove misrecognized entities, which can effectively solve the complex structural entity recognition problem in a specific domain.The experimental results of this paper on the Resume dataset and the self-constructed coal chemical dataset Coal show that LLM-DER performs outstandingly in domain-specific entity recognition, not only outperforming the existing GPT-3.5-turbo baseline, but also exceeding the fully-supervised baseline, verifying its effectiveness in entity recognition.

cs.CL↗

arXiv2024

MetaBGM: Dynamic Soundtrack Transformation For Continuous Multi-Scene Experiences With Ambient Awareness And Personalization

This paper introduces MetaBGM, a groundbreaking framework for generating background music that adapts to dynamic scenes and real-time user interactions. We define multi-scene as variations in environmental contexts, such as transitions in game settings or movie scenes. To tackle the challenge of converting backend data into music description texts for audio generation models, MetaBGM employs a novel two-stage generation approach that transforms continuous scene and user state data into these texts, which are then fed into an audio generation model for real-time soundtrack creation. Experimental results demonstrate that MetaBGM effectively generates contextually relevant and dynamic background music for interactive applications.

cs.SD↗

arXiv2024

REMAST: Real-time Emotion-based Music Arrangement with Soft Transition

Music as an emotional intervention medium has important applications in scenarios such as music therapy, games, and movies. However, music needs real-time arrangement according to changing emotions, bringing challenges to balance emotion real-time fit and soft emotion transition due to the fine-grained and mutable nature of the target emotion. Existing studies mainly focus on achieving emotion real-time fit, while the issue of smooth transition remains understudied, affecting the overall emotional coherence of the music. In this paper, we propose REMAST to address this trade-off. Specifically, we recognize the last timestep's music emotion and fuse it with the current timestep's input emotion. The fused emotion then guides REMAST to generate the music based on the input melody. To adjust music similarity and emotion real-time fit flexibly, we downsample the original melody and feed it into the generation model. Furthermore, we design four music theory features by domain knowledge to enhance emotion information and employ semi-supervised learning to mitigate the subjective bias introduced by manual dataset annotation. According to the evaluation results, REMAST surpasses the state-of-the-art methods in objective and subjective metrics. These results demonstrate that REMAST achieves real-time fit and smooth transition simultaneously, enhancing the coherence of the generated music.

cs.SD↗

arXiv2024

X-ray Sources Classification Using Machine Learning: A Study with EP-WXT Pathfinder LEIA

X-ray observations play a crucial role in time-domain astronomy. The Einstein Probe (EP), a recently launched X-ray astronomical satellite, emerges as a forefront player in the field of time-domain astronomy and high-energy astrophysics. With a focus on systematic surveys in the soft X-ray band, EP aims to discover high-energy transients and monitor variable sources in the universe. To achieve these objectives, a quick and reliable classification of observed sources is essential. In this study, we developed a machine learning classifier for autonomous source classification using data from the EP-WXT Pathfinder Lobster Eye Imager for Astronomy (LEIA) and EP-WXT simulations. The proposed Random Forest classifier, built on selected features derived from light curves, energy spectra, and location information, achieves an accuracy of approximately 95% on EP simulation data and 98% on LEIA observational data. The classifier is integrated into the LEIA data processing pipeline, serving as a tool for manual validation and rapid classification during observations. This paper presents an efficient method for the classification of X-ray sources based on single observations, along with implications of most effective features for the task. This work facilitates rapid source classification for the EP mission and also provides valuable insights into feature selection and classification techniques for enhancing the efficiency and accuracy of X-ray source classification that can be adapted to other X-ray telescope data.

astro-ph.IM↗

arXiv2024

Analytical Modeling of Acoustic Exponential Materials and Physical Mechanism of Broadband Anti-Reflection

Spatially exponential distributions of material properties are ubiquitous in many natural and engineered systems, from the vertical distribution of the atmosphere to acoustic horns and anti-reflective coatings. These media seamlessly interface different impedances, enhancing wave transmission and reducing internal reflections. This work advances traditional transfer matrix theory by integrating analytical solutions for acoustic exponential materials, which possess exponential density and/or bulk modulus, offering a more accurate predictive tool and revealing the physical mechanism of broadband anti-reflection for sound propagation in such non-uniform materials. Leveraging this method, we designed an acoustic dipole array that effectively mimics exponential mass distribution. Through experiments with precisely engineered micro-perforated plates, we demonstrate an ultra-low reflection rate of about 0.86% across a wide frequency range from 420 Hz to 10,000 Hz. Our modified transfer matrix approach underpins the design of exponential materials, and our layering strategy for stacking acoustic dipoles suggests a pathway to more functional gradient acoustic metamaterials.

physics.app-ph↗

arXiv2023

Reverberation Time Control by Acoustic Metamaterials in a Small Room

In recent years, metamaterials have gained considerable attention as a promising material technology due to their unique properties and customizable design, distinguishing them from traditional materials. This article delves into the value of acoustic metamaterials in room acoustics, particularly in small room acoustics that poses specific challenges due to their significant cavity resonant nature. Small rooms usually exhibit an inhomogeneous frequency response spectrum, requiring higher wall absorption with specific spectrum to achieve a uniform acoustic environment, i.e., a constant reverberation time over a wide audible frequency band. To tackle this issue, we developed a design that simultaneously incorporates numerous subwavelength acoustic resonators at different frequencies to achieve customized broadband absorption for the walls of a specific example room. The on-site experimental measurements agree well with the numerical predictions, attesting to the robustness of the design and method. The proposed method of reverse-engineering metamaterials by targeting specific acoustic requirements has broad applicability and unique advantages in small confined spaces with high acoustic requirements, such as recording studios, listening rooms, and car cabins.

physics.app-ph↗