arXiv subjects

Wenhui Xu

Publications and source records attributed to Wenhui Xu.

15 recordsLinked to original sources

arXiv2026

Protein Structure Prediction: From Evolutionary Constraints to Generative Modeling

Accurate protein structure prediction is fundamental to structural biology because protein structure underlies molecular function and provides a basis for mechanistic interpretation. Recent advances in deep learning have transformed the field from multiple sequence alignment (MSA)-driven monomer folding into broader frameworks capable of modeling protein complexes and increasingly heterogeneous molecular systems. Existing reviews have summarized this progress from the perspectives of representative models, application domains, and protein design. Building on these efforts, this review focuses on the methodological evolution of the field itself. It examines recent developments through three closely related dimensions: representations and data, architectures and learning strategies, and confidence and evaluation. Within this perspective, the field is organized into four methodological phases and three cross-cutting transitions: from explicit evolutionary coupling features and early contact prediction to learned sequence representations in AlphaFold2, RoseTTAFold, and ESMFold; from protein-only monomer folding to increasingly integrated modeling of heterogeneous molecular systems in AlphaFold-Multimer, RoseTTAFoldNA, and AlphaFold3; and, more recently, from prediction-oriented structure inference to design-oriented generative modeling in RFdiffusion and related frameworks. This framework provides a clearer understanding of how methodological shifts have shaped the capabilities, limitations, and practical roles of recent models.

cs.AI↗

arXiv2026

Color Fundus Photography Analysis: Co-evolution of Data, Preprocessing, and Modeling toward Multimodal AI

Color Fundus Photography (CFP) is a primary non-invasive imaging modality for large-scale screening of ophthalmic and systemic diseases. Existing surveys mainly summarize task-specific algorithms, datasets, or preprocessing techniques independently, lacking a unified perspective on their co-evolution with modern artificial intelligence. This review provides an integrated overview of CFP AI through the interplay of dataset evolution, preprocessing paradigms, and modeling frameworks. We show that CFP datasets have evolved from small single-center collections with task-specific labels to large multi-center resources featuring multimodal pairings and longitudinal clinical records. Preprocessing has progressed from conventional image enhancement to neural data-engineering pipelines, hardware-aware token optimization, and self-supervised imputation for incomplete electronic health records (EHRs). Meanwhile, modeling has advanced from convolutional neural networks (CNNs) to vision foundation models, state space models (SSMs), and multimodal expert architectures. At the multimodal frontier, CFP is increasingly integrated with EHRs and longitudinal patient information, enabling more comprehensive clinical reasoning beyond isolated image analysis. We conclude that future progress depends on the collaborative optimization of datasets, preprocessing, and multimodal modeling, providing a roadmap toward robust clinical deployment, improved cross-domain generalization, and resource-efficient edge intelligence.

cs.CV↗

arXiv2026

S1-Omni: A Unified Multimodal Reasoning Model for Scientific Understanding, Prediction, and Generation

We present S1-Omni, a unified multimodal reasoning model for scientific understanding, prediction, and generation. AI for Science (AI4S) has advanced significantly through domain-specific models, tool-augmented LLMs, and scientific language models. However, model capabilities remain highly fragmented, limiting the joint modeling of heterogeneous data, scientific laws, and expert knowledge. S1-Omni addresses this gap by consolidating these capabilities into a single, coherent scientific reasoning model. The architecture of S1-Omni is built upon three core components: unified representation of scientific data, natural-world knowledge alignment, and decoding for domain-specific tasks. First, S1-Omni maps natural-language instructions and scientific objects, including CIF, SMILES, protein sequences, spectra, and scientific images, into a shared representation space. Second, it incorporates scientific laws and expert knowledge into data construction and training, enabling the model to reason from scientific evidence. Third, it performs task-specific decoding to support a broad range of applications, including property prediction, spectrum-to-molecular generation, protein site and structure prediction, and scientific image generation and editing. S1-Omni is trained on S1-Omni-Corpus, which covers 200 scientific tasks and contains millions of reasoning samples, and is evaluated on over 60 scientific benchmarks. It outperforms GPT-5.5 and Gemini-3.1-Pro on most benchmarks and matches or surpasses domain-specific models on several benchmarks. Overall, S1-Omni provides a practical path toward unified scientific modeling.

cs.AI↗

arXiv2026

Monolithic integration of diverse crystalline thin films on diamond for near-junction thermal management

The pursuit of extreme miniaturization and high power in 6G RF front-ends has cast thermal dissipation as the central challenge. Here, we have demonstrated the monolithic integration of functionally distinct single-crystal thin films, including \b{eta}-Ga2O3, Si, GaN, and LiTaO3, onto a single diamond substrate using a multi-step transfer printing technique. Focusing on the critical \b{eta}-Ga2O3/diamond interface, we achieve an exceptional interfacial thermal conductance (ITC) of 149 MW m-2 K-1 through ultra-high vacuum (UHV) annealing, creating an atomically sharp interface featuring covalent bonding. Vibrational electron energy-loss spectroscopy (EELS) analysis combining with molecular dynamics (MD) simulations reveal that distinctive interfacial phonon modes at the \b{eta}-Ga2O3/diamond heterointerface dominate ultrahigh ITC. We experimentally demonstrate that by improving the ITC, the thermal resistance (Rth) of a diamond-based \b{eta}-Ga2O3 MOSFET is driven to a record-low value of 1.58 K mm W-1, underscoring the critical role of interface engineering in near-junction thermal management for diamond-integrated devices. This work demonstrates a scalable, diamond-based monolithic integration platform designed to solve the near-junction thermal challenges in high-power RF front-ends.

cond-mat.mtrl-sci↗

arXiv2025

Dipole condensates in synthetic rank-2 electric fields

Dipole condensates, formed from particle-hole pairs, represent a unique class of charge-neutral quantum fluids that evade conventional vector gauge fields, making their electrodynamic responses difficult to probe in natural materials. Here, we propose a tunable platform using strongly interacting two-component ultracold atoms to realize dipole condensates and probe their coupling to rank-2 electric fields. By applying spin-dependent forces and treating spin as a synthetic dimension, we engineer a synthetic rank-2 electric field that induces measurable electrodynamic responses. We identify the atomic analog of perfect Coulomb drag: increasing intercomponent interactions leads to equal and opposite displacements of the centers of mass of the two spin components. Furthermore, a rank-2 electric field imprints a phase twist in the dipole condensate and generates a supercurrent of dipoles that obeys the dipolar Josephson relation -- a smoking gun for dipole condensation. Our results establish a powerful platform for exploring dipolar superfluidity under tensor gauge fields.

cond-mat.quant-gas↗

arXiv2025

Excitonic skin effect

We show that strong interactions combined with band-dependent imaginary vector potentials give rise to boundary localization of particle-hole pairs, which we term the excitonic skin effect. In a bilayer system with layer-specific gain/loss and an in-plane magnetic field, excitons experience a net imaginary vector potential, resulting in directional amplification of particle-hole pairs. Including nearest-neighbor interactions leads to a non-Hermitian bosonic Kitaev model, where the pairing effects grow exponentially with the size of the system, revealing a unique form of critical skin effect in interacting systems. Our framework applies to both atomic and electronic platforms and is directly testable in current experiments. These results also provide a route to explore non-Hermitian analogs of tensor gauge fields.

cond-mat.quant-gas↗

arXiv2024

Robust Ptychographic Reconstruction with an Out-of-Focus Electron Probe

As a burgeoning technique, out-of-focus electron ptychography offers the potential for rapidly imaging atomic-scale large fields of view (FoV) using a single diffraction dataset. However, achieving robust out-of-focus ptychographic reconstruction poses a significant challenge due to the inherent scan instabilities of electron microscopes, compounded by the presence of unknown aberrations in the probe-forming lens. In this study, we substantially enhance the robustness of out-of-focus ptychographic reconstruction by extending our previous calibration method (the Fourier method), which was originally developed for the in-focus scenario. This extended Fourier method surpasses existing calibration techniques by providing more reliable and accurate initialization of scan positions and electron probes. Additionally, we comprehensively explore and recommend optimized experimental parameters for robust out-of-focus ptychography, includingaperture size and defocus, through extensive simulations. Lastly, we conduct a comprehensive comparison between ptychographic reconstructions obtained with focused and defocused electron probes, particularly in the context of low-dose and precise phase imaging, utilizing our calibration method as the basis for evaluation.

physics.optics↗

arXiv2024

A high-performance reconstruction method for partially coherent ptychography

Ptychography is now integrated as a tool in mainstream microscopy allowing quantitative and high-resolution imaging capabilities over a wide field of view. However, its ultimate performance is inevitably limited by the available coherent flux when implemented using electrons or laboratory X-ray sources. We present a universal reconstruction algorithm with high tolerance to low coherence for both far-field and near-field ptychography. The approach is practical for partial temporal and spatial coherence and requires no prior knowledge of the source properties. Our initial visible-light and electron data show that the method can dramatically improve the reconstruction quality and accelerate the convergence rate of the reconstruction. The approach also integrates well into existing ptychographic engines. It can also improve mixed-state and numerical monochromatisation methods, requiring a smaller number of coherent modes or lower dimensionality of Krylov subspace while providing more stable and faster convergence. We propose that this approach could have significant impact on ptychography of weakly scattering samples.

physics.comp-ph↗

arXiv2023

Multipolar condensates and multipolar Josephson effects

When single-particle dynamics are suppressed in certain strongly correlated systems, dipoles arise as elementary carriers of quantum kinetics. These dipoles can further condense, providing physicists with a rich realm to study fracton phases of matter. Whereas recent theoretical discoveries have shown that an unconventional lattice model may host a dipole condensate as the ground state, fundamental questions arise as to whether dipole condensation is a generic phenomenon rather than a specific one unique to a particular model and what new quantum macroscopic phenomena a dipole condensate may bring us with. Here, we show that dipole condensates prevail in bosonic systems. Because of a self-proximity effect, where single-particle kinetics inevitably induces a finite order parameter of dipoles, dipole condensation readily occurs in conventional normal phases of bosons. Our findings allow experimentalists to manipulate the phase of a dipole condensate and deliver dipolar Josephson effects, where supercurrents of dipoles arise in the absence of particle flows. The self-proximity effects can also be utilized to produce a generic multipolar condensate. The kinetics of the $n$-th order multipoles unavoidably creates a condensate of the $(n+1)$-th order multipoles, forming a hierarchy of multipolar condensates that will offer physicists a whole new class of macroscopic quantum phenomena.

cond-mat.quant-gas↗

arXiv2022

An Integrated Constrained Gradient Descent (iCGD) Protocol to Correct Scan-Positional Errors for Electron Ptychography with High Accuracy and Precision

Correcting scan-positional errors is critical in achieving electron ptychography with both high resolution and high precision. This is a demanding and challenging task due to the sheer number of parameters that need to be optimized. For atomic-resolution ptychographic reconstructions, we found classical refining methods for scan positions not satisfactory due to the inherent entanglement between the object and scan positions, which can produce systematic errors in the results. Here, we propose a new protocol consisting of a series of constrained gradient descent (CGD) methods to achieve better recovery of scan positions. The central idea of these CGD methods is to utilize a priori knowledge about the nature of STEM experiments and add necessary constraints to isolate different types of scan positional errors during the iterative reconstruction process. Each constraint will be introduced with the help of simulated 4D-STEM datasets with known positional errors. Then the integrated constrained gradient decent (iCGD) protocol will be demonstrated using an experimental 4D-STEM dataset of the 1H-MoS2 monolayer. We will show that the iCGD protocol can effectively address the errors of scan positions across the spectrum and help to achieve electron ptychography with high accuracy and precision.

cond-mat.other↗

arXiv2022

Effective Actor-centric Human-object Interaction Detection

While Human-Object Interaction(HOI) Detection has achieved tremendous advances in recent, it still remains challenging due to complex interactions with multiple humans and objects occurring in images, which would inevitably lead to ambiguities. Most existing methods either generate all human-object pair candidates and infer their relationships by cropped local features successively in a two-stage manner, or directly predict interaction points in a one-stage procedure. However, the lack of spatial configurations or reasoning steps of two- or one- stage methods respectively limits their performance in such complex scenes. To avoid this ambiguity, we propose a novel actor-centric framework. The main ideas are that when inferring interactions: 1) the non-local features of the entire image guided by actor position are obtained to model the relationship between the actor and context, and then 2) we use an object branch to generate pixel-wise interaction area prediction, where the interaction area denotes the object central area. Moreover, we also use an actor branch to get interaction prediction of the actor and propose a novel composition strategy based on center-point indexing to generate the final HOI prediction. Thanks to the usage of the non-local features and the partly-coupled property of the human-objects composition strategy, our proposed framework can detect HOI more accurately especially for complex images. Extensive experimental results show that our method achieves the state-of-the-art on the challenging V-COCO and HICO-DET benchmarks and is more robust especially in multiple persons and/or objects scenes.

cs.CV↗

arXiv2021

Learning Oriented Remote Sensing Object Detection via Naive Geometric Computing

Detecting oriented objects along with estimating their rotation information is one crucial step for analyzing remote sensing images. Despite that many methods proposed recently have achieved remarkable performance, most of them directly learn to predict object directions under the supervision of only one (e.g. the rotation angle) or a few (e.g. several coordinates) groundtruth values individually. Oriented object detection would be more accurate and robust if extra constraints, with respect to proposal and rotation information regression, are adopted for joint supervision during training. To this end, we innovatively propose a mechanism that simultaneously learns the regression of horizontal proposals, oriented proposals, and rotation angles of objects in a consistent manner, via naive geometric computing, as one additional steady constraint (see Figure 1). An oriented center prior guided label assignment strategy is proposed for further enhancing the quality of proposals, yielding better performance. Extensive experiments demonstrate the model equipped with our idea significantly outperforms the baseline by a large margin to achieve a new state-of-the-art result without any extra computational burden during inference. Our proposed idea is simple and intuitive that can be readily implemented. Source codes and trained models are involved in supplementary files.

cs.CV↗

arXiv2021

Thermal Visualization of Buried Interfaces by Transient and Steady-State Responses of Time-Domain Thermoreflectance

Thermal resistances from interfaces impede heat dissipation in micro/nanoscale electronics, especially for high-power electronics. Despite the growing importance of understanding interfacial thermal transport, advanced thermal characterization techniques which can visualize thermal conductance across buried interfaces, especially for nonmetal-nonmetal interfaces, are still under development. This work reports a dual-modulation-frequency TDTR mapping technique to visualize the thermal conduction across buried semiconductor interfaces for beta-Ga2O3-SiC samples. Both the beta-Ga2O3 thermal conductivity and the buried beta-Ga2O3-SiC thermal boundary conductance (TBC) are visualized for an area of 200 um x 200 um. Areas with low TBC values ( smaller than 20 MW/m2-K) are successfully identified on the TBC map, which correspond to weakly bonded interfaces caused by high-temperature annealing. The steady-state temperature rise (detector voltage), usually ignored in TDTR measurements, is found to be able to probe TBC variations of the buried interfaces without the limit of thermal penetration depth. This technique can be applied to detect defects/voids in deeply buried heterogeneous interfaces non-destructively, and also opens a door for the visualization of thermal conductance in nanoscale nonhomogeneous structures.

cond-mat.mtrl-sci↗

arXiv2020

Critical polynomials in the nonplanar and continuum percolation models

Exact or precise thresholds have been intensively studied since the introduction of the percolation model. Recently the critical polynomial $P_{\rm B}(p,L)$ was introduced for planar-lattice percolation models, where $p$ is the occupation probability and $L$ is the linear system size. The solution of $P_{\rm B} = 0$ can reproduce all known exact thresholds and leads to unprecedented estimates for thresholds of unsolved planar-lattice models. In two dimensions, assuming the universality of $P_{\rm B}$, we use it to study a nonplanar lattice model, i.e., the equivalent-neighbor lattice bond percolation, and the continuum percolation of identical penetrable disks, by Monte Carlo simulations and finite-size scaling analysis. It is found that, in comparison with other quantities, $P_{\rm B}$ suffers much less from finite-size corrections. As a result, we obtain a series of high-precision thresholds $p_c(z)$ as a function of coordination number $z$ for equivalent-neighbor percolation with $z$ up to O$(10^5)$, and clearly confirm the asymptotic behavior $zp_c-1 \sim 1/\sqrt{z}$ for $z \rightarrow \infty$. For the continuum percolation model, we surprisingly observe that the finite-size correction in $P_{\rm B}$ is unobservable within uncertainty O$(10^{-5})$ as long as $L \geq 3$. The estimated threshold number density of disks is $ρ_c = 1.436 325 05(10)$, slightly below the most recent result $ρ_c = 1.436 325 45(8)$ of Mertens and Moore obtained by other means. Our work suggests that the critical polynomial method can be a powerful tool for studying nonplanar and continuum systems in statistical mechanics.

cond-mat.stat-mech↗

arXiv2020

Wafer-scale Heterogeneous Integration of Monocrystalline \b{eta}-Ga2O3 Thin Films on SiC for Thermal Management by Ion-Cutting Technique

The ultra-wide bandgap, high breakdown electric field, and large-area affordable substrates make \b{eta}-Ga2O3 promising for applications of next-generation power electronics while its thermal conductivity is at least one order of magnitude lower than other wide/ultrawide bandgap semiconductors. To avoid the degradation of device performance and reliability induced by the localized Joule-heating, aggressive thermal management strategies are essential, especially for high-power high-frequency applications. This work reports a scalable thermal management strategy to heterogeneously integrate wafer-scale monocrystalline \b{eta}-Ga2O3 thin films on high thermal conductivity SiC substrates by ion-cutting technique. The thermal boundary conductance (TBC) of the \b{eta}-Ga2O3-SiC interfaces and thermal conductivity of the \b{eta}-Ga2O3 thin films were measured by Time-domain Thermoreflectance (TDTR) to evaluate the effects of interlayer thickness and thermal annealing. Materials characterizations were performed to understand the mechanisms of thermal transport in these structures. The results show that the \b{eta}-Ga2O3-SiC TBC values increase with decreasing interlayer thickness and the \b{eta}-Ga2O3 thermal conductivity increases more than twice after annealing at 800 oC due to the removal of implantation-induced strain in the films. A Callaway model is built to understand the measured thermal conductivity. Small spot-to-spot variations of both TBC and Ga2O3 thermal conductivity confirm the uniformity and high-quality of the bonding and exfoliation. Our work paves the way for thermal management of power electronics and \b{eta}-Ga2O3 related semiconductor devices.

physics.app-ph↗