Searcharxiv⌕ Search

arXiv subjects

Xing He

Publications and source records attributed to Xing He.

At least 19 recordsLinked to original sources
arXiv2026

Fracture Risk Prediction in Adults Over 50 Years Old Using DXA and EHR: Comparison of Traditional and Machine Learning Models in Two Large Cohorts

Accurate fracture risk prediction is important for osteoporosis management, but commonly used clinical tools may not fully use information available in electronic health records (EHRs) and dual-energy X-ray absorptiometry (DXA) reports. We developed and externally validated time-to-event fracture prediction models among adults aged 50 years or older with clinically obtained DXA reports in 2 US health care systems. The development cohort was derived from NewYork-Presbyterian/Weill Cornell Medical Center and the external validation cohort from the Indiana Network for Patient Care. Predictors included demographics, lifestyle factors, prior fracture, comorbidities, medication exposures, osteoporosis treatment history, and DXA-derived T-scores extracted from radiology reports. The outcome was time from index DXA to first incident fragility fracture identified from structured diagnosis codes. We evaluated penalized Cox regression, random survival forest, gradient-boosting survival, and XGBoost survival models using 2 prespecified predictor settings and compared discrimination with clinically reported FRAX major osteoporotic fracture probabilities. The development cohort included 11,510 adults, of whom 858 sustained incident fragility fractures; the external validation cohort included 1,932 adults, of whom 180 sustained fractures. In internal validation, the expanded Cox model achieved a mean Harrell C-index of 0.779, compared with 0.653 for FRAX. In external validation, the corresponding Cox model achieved a Harrell C-index of 0.714, compared with 0.590 for FRAX; gradient-boosting survival had the highest external discrimination (0.725). EHR- and DXA-enhanced models showed better discrimination than clinically reported FRAX scores in this DXA-tested population, but calibration assessment, prospective evaluation, and implementation workflow assessment are needed before clinical use.

stat.AP↗
arXiv2026

Exploring Post-Training Alignment of Small Language Models for Biomedical Data-to-Text Generation: A Case Study of Medication Leaflet

Translating complex biomedical data into patient-friendly narratives is central to modern biomedical informatics. This study presents a comparative analysis of training small language models (SLMs) in specialized biomedical datato-text generation tasks. We explore widely adopted post-training methods including supervised fine-tuning (SFT), direct preference optimization (DPO), odds ratio preference optimization (ORPO), and group relative policy optimization (GRPO) with Qwen-based SLMs on a medicine package leaflets dataset. To assess cross-dataset generalizability, we also curated drug label data from openFDA. We evaluate models using both standard lexical overlap metrics like ROUGE as well as semantic similarity measures. Across our experiments, the results show that (1) the aligned SLMs outperform proprietary models like GPT-5; (2) ORPO outperforms the SFTbaselines; (3) GRPO yields the most robust cross-dataset performance among the alignment methods tested as well as GPT-5.

cs.CL↗
arXiv2026

Vortex pinning of Ba$_{0.62}$K$_{0.38}$BiO$_3$ investigated by magneto-optical Kerr-effect and magnetization measurements

Vortex pinning plays a crucial role in determining properties of type-II superconductors. For example, it governs the irreversible magnetic response as well as dissipation caused by vortex motion. Here, we study vortex pinning in the three-dimensional oxide superconductor Ba1-xKxBiO3 using ultra-high-resolution magneto-optical Kerr effect (MOKE) and detailed magnetization measurements. We find that the zero-field MOKE signal in the superconducting state exhibits a pronounced magnetic-history dependence. This behavior closely resembles the remanent magnetization caused by trapped vortices. Furthermore, we demonstrate that the observed evolution of the MOKE signals is well described by Bean's critical-state model for trapped vortices. Our results establish MOKE as a viable optical and mesoscopic probe of vortex pinning in type-II superconductors, providing a new complementary approach to investigate mixed-state phenomena. We also find that the training-field dependence of the MOKE is linear near zero training field, without any anomalies indicative of spontaneous time-reversal-symmetry breaking in an unconventional superconducting state. Our study defines a clear protocol to distinguish vortex-induced MOKE responses from those associated with a time-reversal-symmetry broken superconducting order parameter.

cond-mat.supr-con↗
arXiv2026

Customization under Fire: Plugin Poisoning in Text-to-Image Ecosystem

The prosperity of text-to-image (T2I) models has fostered a vibrant share-and-play ecosystem centered on Low-Rank Adaptation (LoRA) plugins, which allow users to customize and share model capabilities with ease. This democratization, however, comes with a hidden but severe security risk. Malicious users could share and distribute seemingly benign LoRA plugins that contain hidden functionalities to poison the model-sharing market, like Civitai or Liblib, severely undermining the user trust that underpins this collaborative ecosystem and threatening the safety of countless downstream applications. Despite these risks, plugin poisoning in the real-world T2I ecosystem remains underexplored. This paper introduces PoisonLoRA, the first systematic study of LoRA plugin supply-chain risks that exploits the trust and characteristics within the T2I ecosystem. We identify two primary attack instances: (1) Concept Hijacking, where a hijacked LoRA could generate images to influence public opinion and spread propaganda, and (2) Task Injection, where a LoRA is injected to produce harmful content (e.g., NSFW images) only activated by a secret key. Critically, the malicious payload persists with virus-like propagation. Such propagations weaponize the very act of creative collaboration (e.g., LoRA merging) to spread its contagion, turning every remix into a new carrier. Extensive experiments validate that PoisonLoRA is both effective and stealthy. Specifically, we achieve approximately 100% attack success rates (ASR) on both Civitai and Liblib on 6 datasets across 4 scenarios, without being detected by the platforms. The poisoned LoRA demonstrates extreme robustness, with nearly 100% ASR even transferred to different base models and remixed more than 5 times.

cs.CR↗
arXiv2026

Optical Signature of Moir\'e Superlattices Formed by Twisted SrTiO$_3$ Membranes

Moir\'e superlattices formed at the interfaces of mismatched lattices have attracted significant interest over the past decade due to their large tunability of band parameters and interactions among electrons, spins, and lattices. Superlattices made from twisted perovskite oxides may have strong structure and potential modulation, but evidence of such modulation over macroscopic areas, particularly at large twisting angles, has not been clearly demonstrated so far. Here, we fabricated millimeter-scale twisted oxide bilayers at $36^\circ$ angle, close to the simple coincidence site lattice condition $\Sigma5$, from freestanding SrTiO$_3$ membranes. We discovered new low-frequency vibrational modes whose Raman activity, according to molecular dynamics simulations, is greatly enhanced by an asymmetric, twisted interface between the SrO and TiO$_2$ layers. Such an interface is energetically favorable from first-principles calculations and is corroborated by the observation of strong second harmonic generation from the interface comparable to that from the SrTiO$_3$ surface throughout the bilayer region. The results are consistent with interlayer coupling enhanced by high-temperature annealing and confirmed by cross-sectional scanning transmission electron microscopy imaging. Our work sheds light on the structural behavior of twisted oxides and provides directions for tuning their phononic and nonlinear optical properties in future studies.

cond-mat.mtrl-sci↗
arXiv2026

Carrier-coupled ultrafast structural dynamics and interlayer energy transport of supported transition metal dichalcogenide heterostructures

Understanding the electronic coupling and energy flow across layered two-dimensional heterostructures (HSs) is crucial to the exploitation of carrier and phonon transports as well as thermal management in next-generation optoelectronic devices. By using reflection ultrafast electron diffraction, we directly examine photoinduced out-of-plane structural dynamics of supported MoS2/WS2 bilayer HSs and their individual monolayers. Experimental evidence reveals the launch of ultrafast carrier-coupled intralayer atomic motions due to interlayer charge transfer across the van der Waals (vdW) heterojunctions that is absent for individual monolayers. Such a notable carrier-lattice correlation is in addition to the electronic coupling manifested in the enhanced optical absorption for HSs. Also, different pathways of energy flow as a result of carrier-phonon coupling and phonon scattering are reported with the corresponding characteristic times. On longer timescales, relaxation of thermalized atomic motions can be sufficiently described by a thermal transport model. A higher thermal boundary conductance (TBC) across MoS2/WS2 HSs is obtained compared to those at the monolayer-substrate interfaces; however, the similar TBC values suggest comparable couplings of phonons across vdW contacts. These results further shed light on the optical, phonon, and interfacial thermal properties of vertically-stacked vdW HSs.

cond-mat.mes-hall↗
arXiv2026

Tuning charge-transport properties and magnetic order in metallic EuTiO$_{3-\delta}$

The stoichiometric antiferromagnetic insulator EuTiO$_3$ is proximate to a ferroelectric phase. Whereas cation substitution has been used as a tuning parameter to introduce charge carriers and manipulate the magnetism, the effects of oxygen-vacancy doping have been less explored. Here we report a detailed study of the charge transport and magnetic properties of metallic, oxygen-vacancy-doped EuTiO$_{3-\delta}$. Using CaH$_2$ as an oxygen getter to achieve a higher carrier concentration than previously reported, we find that the phase diagram of the oxygen-vacancy-doped system is distinct from that obtained via cation doping. In particular, we uncover a change from antiferromagnetic to ferromagnetic order in the metallic state, with a maximum Curie temperature of TC $\approx$ 11 K at the highest carrier concentration of n $\approx$ 10$^{21}$ cm$^{-3}$. These findings are supported by density functional theory calculations, which indicate a significant change in the nearest-neighbor magnetic exchange constant with increasing electron doping. We also present x-ray diffuse scattering and complementary first-principles results that reveal that, similar to the prominent incipient ferroelectric perovskite SrTiO$_3$, the data for EuTiO$_3$ are consistent with thermal diffuse scattering and with the absence of quasi-elastic contributions. Finally, we report specific heat measurements that confirm the magnetic transition temperatures deduced from magnetization measurements and corroborate the lattice dynamics picture inferred from the diffuse scattering data.

cond-mat.mtrl-sci↗
arXiv2026

FreqCycle: A Multi-Scale Time-Frequency Analysis Method for Time Series Forecasting

Mining time-frequency features is critical for time series forecasting. Existing research has predominantly focused on modeling low-frequency patterns, where most time series energy is concentrated. The overlooking of mid to high frequency continues to limit further performance gains in deep learning models. We propose FreqCycle, a novel framework integrating: (i) a Filter-Enhanced Cycle Forecasting (FECF) module to extract low-frequency features by explicitly learning shared periodic patterns in the time domain, and (ii) a Segmented Frequency-domain Pattern Learning (SFPL) module to enhance mid to high frequency energy proportion via learnable filters and adaptive weighting. Furthermore, time series data often exhibit coupled multi-periodicity, such as intertwined weekly and daily cycles. To address coupled multi-periodicity as well as long lookback window challenges, we extend FreqCycle hierarchically into MFreqCycle, which decouples nested periodic features through cross-scale interactions. Extensive experiments on seven diverse domain benchmarks demonstrate that FreqCycle achieves state-of-the-art accuracy while maintaining faster inference speeds, striking an optimal balance between performance and efficiency.

cs.LG↗
arXiv2026

Fedcompass: Federated Clustered and Periodic Aggregation Framework for Hybrid Classical-Quantum Models

Federated learning enables collaborative model training across decentralized clients under privacy constraints. Quantum computing offers potential for alleviating computational and communication burdens in federated learning, yet hybrid classical-quantum federated learning remains susceptible to performance degradation under non-IID data. To address this,we propose FEDCOMPASS, a layered aggregation framework for hybrid classical-quantum federated learning. FEDCOMPASS employs spectral clustering to group clients by class distribution similarity and performs cluster-wise aggregation for classical feature extractors. For quantum parameters, it uses circular mean aggregation combined with adaptive optimization to ensure stable global updates. Experiments on three benchmark datasets show that FEDCOMPASS improves test accuracy by up to 10.22% and enhances convergence stability under non-IID settings, outperforming six strong federated learning baselines.

cs.LG↗
arXiv2025

Direct Observation of Energy Transport Dynamics and High Thermal Conductance across Single Solid-Molecule Junctions

Interfaces play a crucial role in energy transport at the nanoscale. However, direct experimental observations of interfacial thermal conductance across molecular junctions have remained challenging due to the high spatiotemporal resolution required for probing. Here, we report dynamic energy transport processes across multi-component molecular junctions observed at the atomic level by employing reflection ultrafast electron diffraction. A clear temporal sequence of energy transfer is revealed at early times following photoexcitation of Au(111) surfaces chemically bonded with self-assembled monolayers (SAMs) of alkanethiols: from the gold surface layer (SL) to the head groups of a SAM and then to the CH2-CH2 methylene lattice. Remarkably, the structural dynamics of the gold SL differ significantly from those of clean gold. Furthermore, the methylene lattice dynamics exhibit chain-length insensitivity but with a length-dependent retention time to reach full thermalization. Quantitatively, we find an agreement in the increased out-of-plane atomic motions between the surface gold atoms and the SAM methylene units, signifying the nature of motion-based coupling for interfacial energy transport. High interfacial thermal conductance (~300 MW/m^2/K) and high thermal conductivity for the methylene lattice (~60 W/m/K) are obtained under the condition of impulsive heating, which provide strong support for previous room-temperature theoretical predictions and also the ballistic nature of intrachain heat transfer. This time- and spatially-resolved experimental approach will enable future quantitative assessment of interfacial energy transport across solid-molecule junctions.

physics.chem-ph↗
arXiv2025

PointSFDA: Source-free Domain Adaptation for Point Cloud Completion

Conventional methods for point cloud completion, typically trained on synthetic datasets, face significant challenges when applied to out-of-distribution real-world scans. In this paper, we propose an effective yet simple source-free domain adaptation framework for point cloud completion, termed \textbf{PointSFDA}. Unlike unsupervised domain adaptation that reduces the domain gap by directly leveraging labeled source data, PointSFDA uses only a pretrained source model and unlabeled target data for adaptation, avoiding the need for inaccessible source data in practical scenarios. Being the first source-free domain adaptation architecture for point cloud completion, our method offers two core contributions. First, we introduce a coarse-to-fine distillation solution to explicitly transfer the global geometry knowledge learned from the source dataset. Second, as noise may be introduced due to domain gaps, we propose a self-supervised partial-mask consistency training strategy to learn local geometry information in the target domain. Extensive experiments have validated that our method significantly improves the performance of state-of-the-art networks in cross-domain shape completion. Our code is available at \emph{\textcolor{magenta}{https://github.com/Starak-x/PointSFDA}}.

cs.CV↗
arXiv2025

PointSea: Point Cloud Completion via Self-structure Augmentation

Point cloud completion is a fundamental yet not well-solved problem in 3D vision. Current approaches often rely on 3D coordinate information and/or additional data (e.g., images and scanning viewpoints) to fill in missing parts. Unlike these methods, we explore self-structure augmentation and propose PointSea for global-to-local point cloud completion. In the global stage, consider how we inspect a defective region of a physical object, we may observe it from various perspectives for a better understanding. Inspired by this, PointSea augments data representation by leveraging self-projected depth images from multiple views. To reconstruct a compact global shape from the cross-modal input, we incorporate a feature fusion module to fuse features at both intra-view and inter-view levels. In the local stage, to reveal highly detailed structures, we introduce a point generator called the self-structure dual-generator. This generator integrates both learned shape priors and geometric self-similarities for shape refinement. Unlike existing efforts that apply a unified strategy for all points, our dual-path design adapts refinement strategies conditioned on the structural type of each point, addressing the specific incompleteness of each point. Comprehensive experiments on widely-used benchmarks demonstrate that PointSea effectively understands global shapes and generates local details from incomplete input, showing clear improvements over existing methods.

cs.CV↗
arXiv2024

CAMH: Advancing Model Hijacking Attack in Machine Learning

In the burgeoning domain of machine learning, the reliance on third-party services for model training and the adoption of pre-trained models have surged. However, this reliance introduces vulnerabilities to model hijacking attacks, where adversaries manipulate models to perform unintended tasks, leading to significant security and ethical concerns, like turning an ordinary image classifier into a tool for detecting faces in pornographic content, all without the model owner's knowledge. This paper introduces Category-Agnostic Model Hijacking (CAMH), a novel model hijacking attack method capable of addressing the challenges of class number mismatch, data distribution divergence, and performance balance between the original and hijacking tasks. CAMH incorporates synchronized training layers, random noise optimization, and a dual-loop optimization approach to ensure minimal impact on the original task's performance while effectively executing the hijacking task. We evaluate CAMH across multiple benchmark datasets and network architectures, demonstrating its potent attack effectiveness while ensuring minimal degradation in the performance of the original task.

cs.CR↗
arXiv2024

Redefinition of Digital Twin and its Situation Awareness Framework Designing Towards Fourth Paradigm for Energy Internet of Things

Traditional knowledge-based situation awareness (SA) modes struggle to adapt to the escalating complexity of today's Energy Internet of Things (EIoT), necessitating a pivotal paradigm shift. In response, this work introduces a pioneering data-driven SA framework, termed digital twin-based situation awareness (DT-SA), aiming to bridge existing gaps between data and demands, and further to enhance SA capabilities within the complex EIoT landscape. First, we redefine the concept of digital twin (DT) within the EIoT context, aligning it with data-intensive scientific discovery paradigm (the Fourth Paradigm) so as to waken EIoT's sleeping data; this contextual redefinition lays the cornerstone of our DT-SA framework for EIoT. Then, the framework is comprehensively explored through its four fundamental steps: digitalization, simulation, informatization, and intellectualization. These steps initiate a virtual ecosystem conducive to a continuously self-adaptive, self-learning, and self-evolving big model (BM), further contributing to the evolution and effectiveness of DT-SA in engineering. Our framework is characterized by the incorporation of system theory and Fourth Paradigm as guiding ideologies, DT as data engine, and BM as intelligence engine. This unique combination forms the backbone of our approach. This work extends beyond engineering, stepping into the domain of data science -- DT-SA not only enhances management practices for EIoT users/operators, but also propels advancements in pattern analysis and machine intelligence (PAMI) within the intricate fabric of a complex system. Numerous real-world cases validate our DT-SA framework.

cs.NI↗
arXiv2024

Structural Properties of Plastically Deformed SrTiO3 and KTaO3

Dislocation engineering has the potential to open new avenues toward the exploration and modification of the properties of quantum materials. Strontium titanate (SrTiO3, STO) and potassium tantalate (KTaO3, KTO) are incipient ferroelectrics that show metallization and superconductivity at extremely low charge carrier concentrations, and have been the subject of resurgent interest. These materials also exhibit remarkable ambient-temperature ductility, and thus represent exceptional platforms for studies of the effects of deformation-induced dislocation structures on electronic properties. Recent work on plastically deformed STO revealed an enhancement of the superconducting transition temperature and the emergence of local ferroelectricity and magnetism near self-organized dislocation walls. Here we present a comprehensive structural analysis of plastically deformed STO and KTO, employing specially designed strain cells, diffuse neutron and x-ray scattering, Raman scattering, and nuclear magnetic resonance (NMR). Diffuse scattering and NMR provide insight into the dislocation configurations and densities and their dependence on strain. As in the prior work on STO, Raman scattering reveals evidence for local ferroelectric order near dislocation walls in plastically deformed KTO. Our findings provide valuable information about the self-organized defect structures in both materials, and they position KTO as a second model system with which to explore the associated emergent physics.

cond-mat.mtrl-sci↗
arXiv2024

Impacts of Hot Electron Diffusion, Electron-Phonon Coupling, and Surface Atoms on Metal Surface Dynamics Revealed by Reflection Ultrafast Electron Diffraction

Metals exhibit nonequilibrium electron and lattice subsystems at transient times following femtosecond laser excitation. In the past four decades, various optical spectroscopy and time-resolved diffraction methods have been used to study electron-phonon coupling and the effects of underlying dynamical processes. Here, we take advantage of the surface specificity of reflection ultrafast electron diffraction (UED) to examine the structural dynamics of photoexcited metal surfaces, which are apparently slower in recovery than predicted by thermal diffusion from the profile of absorbed energy. Fast diffusion of hot electrons is found to critically reduce surface excitation and affect the temporal dependence of the increased atomic motions on not only the ultrashort but sub-nanosecond times. Whereas the two-temperature model with the accepted physical constants of platinum can reproduce the observed surface lattice dynamics, gold is found to exhibit appreciably larger-than-expected dynamic vibrational amplitudes of surface atoms while keeping the commonly used electron-phonon coupling constant. Such surface behavioral difference at transient times can be understood in the context of the different strengths of binding to surface atoms for the two metals. In addition, with the quantitative agreements between diffraction and theoretical results, we provide convincing evidence that surface structural dynamics can be reliably obtained by reflection UED even in the presence of laser-induced transient electric fields.

physics.chem-ph↗
arXiv2023

Exploration of Superposition Theorem in Spectrum Space for Composite Event Analysis in an ADN

This study presents a formulation of the Superposition Theorem (ST) in the spectrum space, tailored for the analysis of composite events in an active distribution network (ADN). Our formulated ST enables a quantitative analysis on a composite event, uncovering the property of additivity among independent atom events in the spectrum space. This contribution is a significant addition to the existing literature and has profound implications in various application scenarios. To accomplish this, we leverage random matrix theory (RMT), specifically the asymptotic empirical spectral distribution, Stieltjes transform, and R transform. These mathematical tools establish a nonlinear, model-free, and unsupervised addition operation in the spectrum space. Comprehensive details, including a related roadmap,theorems, deductions, and proofs, are provided in this work. Case studies, utilizing field data, validate our newly derived ST formulation by demonstrating a remarkable performance. Our ST formulation is model-free, non-linear, non-supervised, theory-guided, and uncertainty-insensitive, making it a valuable asset in the realm of composite event analysis in ADN.

stat.AP↗
arXiv2023

SVDFormer: Complementing Point Cloud via Self-view Augmentation and Self-structure Dual-generator

In this paper, we propose a novel network, SVDFormer, to tackle two specific challenges in point cloud completion: understanding faithful global shapes from incomplete point clouds and generating high-accuracy local structures. Current methods either perceive shape patterns using only 3D coordinates or import extra images with well-calibrated intrinsic parameters to guide the geometry estimation of the missing parts. However, these approaches do not always fully leverage the cross-modal self-structures available for accurate and high-quality point cloud completion. To this end, we first design a Self-view Fusion Network that leverages multiple-view depth image information to observe incomplete self-shape and generate a compact global shape. To reveal highly detailed structures, we then introduce a refinement module, called Self-structure Dual-generator, in which we incorporate learned shape priors and geometric self-similarities for producing new points. By perceiving the incompleteness of each point, the dual-path design disentangles refinement strategies conditioned on the structural type of each point. SVDFormer absorbs the wisdom of self-structures, avoiding any additional paired information such as color images with precisely calibrated camera intrinsic parameters. Comprehensive experiments indicate that our method achieves state-of-the-art performance on widely-used benchmarks. Code will be available at https://github.com/czvvd/SVDFormer.

cs.CV↗