arXiv subjects

Wenbing Li

Publications and source records attributed to Wenbing Li.

14 recordsLinked to original sources

arXiv2026

SubdivAR: Autoregressive Next-Scale Prediction for Neural Mesh Subdivision

Mesh subdivision is a fundamental operation for converting coarse, editable meshes into high-resolution surfaces, with broad applications in digital asset creation. Classical rule-based schemes rely on fixed local refinement rules and often produce over-smoothed surfaces. Recent neural subdivision methods improve detail synthesis, but remain constrained by local modeling and exhibit limited generalizability. We present SubdivAR, a neural mesh subdivision framework based on our proposed Mesh Autoregressive Representation (MAR). MAR arranges meshes at different subdivision levels into an ordered scale sequence, reformulating subdivision as autoregressive next-scale prediction. To support this formulation, we introduce a Hybrid Topology-Aware Transformer that combines global semantic attention with topology-constrained local feature aggregation. SubdivAR adopts a next-scale coordinate prediction paradigm, regressing vertex offsets at each refinement stage to preserve subdivision topology while recovering fine-grained geometric details. To enable reliable learning, we construct FII-40K, a curated dataset of nearly 40,000 high-quality meshes with multi-level subdivision supervision. Experiments show that SubdivAR outperforms state-of-the-art baselines, reducing Hausdorff Distance and Chamfer Distance by 18.8% and 14.2%, respectively, and demonstrates strong robustness on complex open-surface geometries.

cs.CV↗

arXiv2026

Large Language Model as Token Compressor and Decompressor

In this paper, we study whether an off-the-shelf LLM can be adapted into a discrete, variable-length token compressor and decompressor for long-context processing. To this end, we design a self-expressive autoencoding framework that fine-tunes a pretrained LLM with lightweight LoRA adapters to map long texts into compact sequences of learned latent codes, termed Z-tokens, and to decode them back into natural language or task outputs. The resulting representation is content-adaptive: less predictable or information-dense segments can receive more Z-tokens, while redundant regions can be represented more compactly through a budget-aware length regularizer. Our method is evaluated on long-context datasets such as Wikipedia, CNN/DailyMail, HotpotQA, and QuALITY, showing that it preserves reconstruction quality and downstream performance while reducing effective context length, generation-stage memory usage, and end-to-end latency. This simple design supports both direct decoding from compressed contexts and autoregressive generation in the Z-token space, providing a practical interface for efficient long-context inference.

cs.CL↗

arXiv2026

LATO: 3D Mesh Flow Matching with Structured TOpology Preserving LAtents

In this paper, we introduce LATO, a novel topology-preserving latent representation that enables scalable, flow matching-based synthesis of explicit 3D meshes. LATO represents a mesh as a Vertex Displacement Field (VDF) anchored on surface, incorporating a sparse voxel Variational Autoencoder (VAE) to compress this explicit signal into a structured, topology-aware voxel latent. To decapsulate the mesh, the VAE decoder progressively subdivides and prunes latent voxels to instantiate precise vertex locations. In the end, a dedicated connection head queries the voxel latent to predict edge connectivity between vertex pairs directly, allowing mesh topology to be recovered without isosurface extraction or heuristic meshing. For generative modeling, LATO adopts a two-stage flow matching process, first synthesizing the structure voxels and subsequently refining the voxel-wise topology features. Compared to prior isosurface/triangle-based diffusion models and autoregressive generation approaches, LATO generates meshes with complex geometry, well-formed topology while being highly efficient in inference.

cs.CV↗

arXiv2025

Semantic-Aware Logical Reasoning via a Semiotic Framework

Logical reasoning is a fundamental capability of large language models. However, existing studies often overlook the interaction between logical complexity and semantic complexity, leading to systems that struggle with abstract propositions, ambiguous contexts, and conflicting stances that are central to human reasoning. We propose LogicAgent, a semiotic-square-guided framework that jointly addresses these two axes of difficulty. The semiotic square provides a principled structure for multi-perspective semantic analysis, and LogicAgent integrates automated deduction with reflective verification to manage logical complexity across deeper reasoning chains. To support evaluation under these conditions, we introduce RepublicQA, a benchmark that couples semantic complexity with logical depth. RepublicQA reaches college-level semantic difficulty (FKGL 11.94), contains philosophically grounded abstract propositions with systematically constructed contrary and contradictory forms, and offers a semantically rich setting for assessing logical reasoning in large language models. Experiments show that LogicAgent achieves state-of-the-art performance on RepublicQA with a 6.25 percent average improvement over strong baselines, and generalizes effectively to mainstream logical reasoning benchmarks including ProntoQA, ProofWriter, FOLIO, and ProverQA, achieving an additional 7.05 percent average gain. These results demonstrate the effectiveness of semiotic-grounded multi-perspective reasoning in enhancing logical performance. Code is available at https://github.com/AI4SS/Logic-Agent.

cs.AI↗

arXiv2025

LoRA-Mixer: Coordinate Modular LoRA Experts Through Serial Attention Routing

Recent attempts to combine low-rank adaptation (LoRA) with mixture-of-experts (MoE) for multi-task adaptation of Large Language Models (LLMs) often replace whole attention/FFN layers with switch experts or append parallel expert branches, undermining parameter efficiency and limiting task specialization. We introduce LoRA-Mixer, a modular MoE framework that routes task-specific LoRA experts into the core projection matrices of the attention module, namely input and output linear layers, rather than primarily targeting FFN blocks. The design delivers fine-grained token-level specialization by fully exploiting the attention mechanism, while remaining drop-in compatible with Transformers and state-space models (SSMs), since linear projection layers are ubiquitous. To train robust routers from limited data while promoting stable, selective decisions and high expert reuse, LoRA-Mixer employs an adaptive Routing Specialization Loss (RSL) that jointly enforces global load balance and input-aware specialization via an entropy-shaping objective. The framework supports two regimes: (i) joint optimization of adapters and router with a differentiable hard-soft top-k routing scheme, and (ii) plug-and-play routing over frozen, pre-trained LoRA modules sourced from public repositories. Across 15 benchmarks, including MedQA, GSM8K, HumanEval, and GLUE, RSL-optimized LoRA-Mixer outperforms state-of-the-art routing and LoRA-MoE baselines while using 48 percent of their trainable parameters, with gains of 3.79, 2.90, and 3.95 percentage points on GSM8K, CoLA, and ARC-C, respectively. Cross-model transfer and adapter reuse experiments further demonstrate the approach's versatility and data efficiency. Our code is available at https://github.com/hustcselwb/LoRA-Mixer.

cs.LG↗

arXiv2025

Gate-tunable hot electron extraction in a two-dimensional semiconductor heterojunction

Hot carrier solar cells (HCSCs), harvesting excess energy of the hot carriers generated by above-band-gap photoexcitation, is crucial for pushing the solar cell efficiency beyond the Shockley Queisser limit, which is challenging to realize mainly due to fast hot-carrier cooling. By performing transient reflectance spectroscopy in a MoSe2/hBN/WS2 junction, we demonstrate the gate-tunable harvest of hot electrons from MoSe2 to WS2. By spectrally distinguishing hot-electron extraction from lattice temperature increase, we find that electrostatically doped electrons in MoSe2 can boost hot-electron extraction density (n_ET) by factor up to several tens. Such enhancement arises from interaction between hot excitons and doped electrons, which converts the excess energy of hot excitons to excitations of the Fermi sea and hence generates hot electrons. Moreover, n_ET can be further enhanced by reducing the conduction band offset with external electric field. Our results provide in-depth insights into design of HCSCs with electrostatic strategies.

cond-mat.mes-hall↗

arXiv2024

Coupled Mamba: Enhanced Multi-modal Fusion with Coupled State Space Model

The essence of multi-modal fusion lies in exploiting the complementary information inherent in diverse modalities. However, prevalent fusion methods rely on traditional neural architectures and are inadequately equipped to capture the dynamics of interactions across modalities, particularly in presence of complex intra- and inter-modality correlations. Recent advancements in State Space Models (SSMs), notably exemplified by the Mamba model, have emerged as promising contenders. Particularly, its state evolving process implies stronger modality fusion paradigm, making multi-modal fusion on SSMs an appealing direction. However, fusing multiple modalities is challenging for SSMs due to its hardware-aware parallelism designs. To this end, this paper proposes the Coupled SSM model, for coupling state chains of multiple modalities while maintaining independence of intra-modality state processes. Specifically, in our coupled scheme, we devise an inter-modal hidden states transition scheme, in which the current state is dependent on the states of its own chain and that of the neighbouring chains at the previous time-step. To fully comply with the hardware-aware parallelism, we devise an expedite coupled state transition scheme and derive its corresponding global convolution kernel for parallelism. Extensive experiments on CMU-MOSEI, CH-SIMS, CH-SIMSV2 through multi-domain input verify the effectiveness of our model compared to current state-of-the-art methods, improved F1-Score by 0.4\%, 0.9\%, and 2.3\% on the three datasets respectively, 49\% faster inference and 83.7\% GPU memory save. The results demonstrate that Coupled Mamba model is capable of enhanced multi-modal fusion.

cs.AI↗

arXiv2022

Fluorescence calorimetry of an ion crystal

Motivated by the challenge of identifying intruder ions in a cold ion crystal, we investigate calorimetry from emitted fluorescence light. Under continuous Doppler cooling, the ion crystal reaches a temperature equilibrium with a fixed level of fluorescence intensity and any change in the motional energy of the crystal results in a modification of this intensity. We theoretically determine the fluorescence rate of an ion crystal as a function of the temperature, assuming that laser light is scattered along a two-level electronic transition, which couples to the crystal's vibrations via the mechanical effects of light. We analyze how the heat dissipated by collisions of an incoming intruder ion alters the scattering rate. We argue that an energy change by an incoming $^{229}$Th$^{10+}$ ion can be unambiguously detected within 100 $μ$s via illuminating a fraction of a 10$^{3}$ ion crystal. This method enables applications including capture and spectroscopy of charged states of thorium isotopes and investigation of highly charged ions.

physics.atom-ph↗

arXiv2022

Robust Polarization Gradient Cooling of Trapped Ions

We implement three-dimensional polarization gradient cooling of trapped ions. Counter-propagating laser beams near $393\,$nm impinge in lin$\,\perp\,$lin configuration, at a frequency below the S$_{1/2}$ to P$_{3/2}$ resonance in $^{40}$Ca$^+$. We demonstrate mean phonon numbers of $5.4(4)$ at a trap frequency of $2π\times 285\,$kHz and $3.3(4)$ at $2π\times480\,$kHz, in the axial and radial directions, respectively. Our measurements demonstrate that cooling with laser beams detuned to lower frequencies from the resonance is robust against an elevated phonon occupation number, and thus works well for an initial ion motion far out of the Lamb-Dicke regime, for up to four ions, and for a micromotion modulation index $β\leq 0.1$. Still, we find that the spectral impurity of the laser field influences both, cooling rates and cooling limits. Thus, a Fabry-Pérot cavity filter is employed for efficiently suppressing amplified spontaneous emission of the diode laser.

physics.atom-ph↗

arXiv2021

Development of a Cesium Fountain Clock at HUST: Preliminary Results

A cesium atomic fountain clock is under development at Huazhong University of Science and Technology (HUST) in China. In this paper, we describe the construction of the entire fountain clock system and report the preliminary results. A frequency stability of $2.5\times 10^{-13} τ^{-1/2}$ has been achieved by inter-comparison with a hydrogen maser, and the factors limiting the frequency stability are also discussed.

physics.atom-ph↗

arXiv2021

A low phase noise microwave source for high performance CPT Rb atomic clock

Phase noise of the frequency synthesizer is one of the main limitations to the short-term stability of microwave atomic clocks. In this work, we demonstrated a low-noise, simple-architecture microwave frequency synthesizer for a coherent population trapping (CPT) clock. The synthesizer is mainly composed of a 100 MHz oven controlled crystal oscillator (OCXO), a microwave comb generator and a direct digital synthesizer (DDS). The absolute phase noises of 3.417 GHz signal are measured to be -55 dBc/Hz, -81 dBc/Hz, -111 dBc/Hz and -134 dBc/Hz, respectively, for 1 Hz, 10 Hz, 100 Hz and 1 kHz offset frequencies, which shows only 1 dB deterioration at the second harmonic of the modulation frequency of the atomic clock. The estimated frequency stability of intermodulation effect is 4.7*10^{-14} at 1s averaging time, which is about half order of magnitude lower than that of the state-of-the-art CPT Rb clock. Our work offers an alternative microwave synthesizer for high-performance CPT Rb atomic clock.

physics.ins-det↗

arXiv2020

Development of a recoil ion source providing slow Th ions including $^{229(m)}$Th in a broad charge state distribution

Ions of the isomer $^{229m}$Th are a topic of high interest for the construction of a "nuclear clock" and in the field of fundamental physics for testing symmetries of nature. They can be efficiently captured in Paul traps which are ideal for performing high precision quantum logic spectroscopy. Trapping and identification of long-lived $^{232}$Th$^{+}$ ions from a laser ablation source was already demonstrated by the TACTICa collaboration on Trapping And Cooling of Thorium Ions with Calcium. The $^{229m}$Th is most easily accessible as $α$-decay daughter of the decay of $^{233}$U. We report on the development of a source for slow Th ions, including $^{229(m)}$Th for the TACTICa experiment. The $^{229(m)}$Th source is currently under construction and comprises a $^{233}$U monolayer, from which $^{229(m)}$Th ions recoil. These are decelerated in an electric field. Conservation of the full initial charge state distribution of the $^{229(m)}$Th recoil ions is one of the unique features of this source. We present ion-flight simulations for our adopted layout and give a final design. This source will provide Th ions in their original charge state at energies suitable for capture in a linear Paul trap for spectroscopy investigations.

physics.atom-ph↗

arXiv2019

A simple scheme of low phase noise microwave synthesizers based on the sub-sampling phase lock loop

In this paper, we demonstrate a simple scheme of 6.835 GHz microwave frequency synthesizer based on the sub-sampling phase lock loop (PLL) technique. The application of the sub-sampling PLL is the key to simplify the architecture of the synthesizer in this scheme. A 100 MHz oven controlled crystal oscillator (OCXO) with ultra-low phase noise is used as the initial signal source. Then a dielectric resonant oscillator (DRO) of 6.8 GHz is directly phase locked to the 100 MHz OCXO utilizing the sub-sampling PLL. Benefiting from the sub-sampling PLL, the processes of microwave frequency multiplication and phase lock which are necessary in the development of microwave synthesizer are greatly simplified. Therefore, the architecture of the synthesizer is very simple. Correspondingly, the power consumption and cost of the synthesizer are low. The absolute phase noises of the 6.835 GHz output signal are measured to be -47 dBc/Hz, -77 dBc/Hz, -104 dBc/Hz and -121 dBc/Hz at 1 Hz, 10 Hz, 100 Hz and 1 kHz offset frequencies, respectively. he synthesizer can be used as the local oscillator of the Rb atomic clocks. For the Rb atomic clocks operated in the continuous or pulsed optically pumped (POP) mode, Tthe short-term frequency stability limited by the absolute phase noises of the synthesizer through the intermodulation or the Dick effect is theoretically calculated to be better than 5.0E10^-14/t^1/2. This low phase noise microwave frequency synthesizer can be used in other experiments of fundamental physics measurement.

physics.atom-ph↗

arXiv2018

A low phase noise microwave frequency synthesizer based on parameters optimized NLTL for Cs fountain clock

We report on the development and phase noise performance of a 9.1926 GHz microwave frequency synthesizer to be used as the local oscillator for a Cs fountain clock. It is based on frequency multiplication and synthesis from an ultralow phase noise 5 MHz Oven Controlled Crystal Oscillator (OCXO) and 100 MHz Voltage Controlled Crystal Oscillator (VCXO).The key component of the frequency multiplication is a non-linear transmission-line (NLTL) used as a frequency comb generator. The phase noise of the synthesizer is improved by carefully optimizing the input power, the input and output impedances of the NLTL. The absolute phase noises of the 9.1926 GHz output signal are measured to be -64 dBc/Hz, -83 dBc/Hz, -92 dBc/Hz, -117 dBc/Hz and -119 dBc/Hz at 1 Hz, 10Hz, 100Hz, 1 kHz and 10 kHz offset frequencies, respectively. The residual phase noise of the synthesizer is measured to be -82 dBc/Hz at 1 Hz offset frequency. The measurement result shows that the absolute phase noise at the frequency range of 1 - 100 Hz is mainly limited by the phase noise of the OCXO. The contribution of the absolute phase noise to the fountain clock short-term frequency stability is calculated to be 7.0x10^(-14). The residual frequency stability of the synthesizer is measured to be1.5x10^(-14), which is consistent with the calculated frequency stability due to the residual phase noise of the synthesizer. Meanwhile we designed and realized an interferometric microwave switch in the synthesizer to eliminate the frequency shifts induced by the microwave leakage. The extinction ratio of the switch is measured to be more than 50 dB. In the scheme, we use only commercially available components to build the microwave frequency synthesizer with excellent phase noise performance for high-performance Cs fountain clocks.

physics.atom-ph↗