arXiv subjects

Bonan Zhang

Publications and source records attributed to Bonan Zhang.

10 recordsLinked to original sources

arXiv2026

MoE-ViE: Mixture of Experts Vision Encoder for Efficient Image and Video Understanding

Vision encoders are a critical component of vision-language models, and scaling their capacity effectively improves performance. However, dense scaling increases compute cost and inference latency. Mixture-of-Experts (MoE) architectures offer a compelling alternative, having enabled efficient scaling in LLMs, yet the MoE design space for CLIP-style vision encoders remains underexplored at State-of-the-Art (SOTA) levels. In this work, we systematically study MoE designs for vision encoder scaling and find that fine-grained MoE topologies yield substantial gains over both dense and standard MoE counterparts. We further propose an auxiliary-loss-free balancing variant for better expert utilization, and design a specialized MoE kernel to mitigate inference latency overhead. To enhance video capabilities while preserving image knowledge, we introduce frame-level distillation paired with a novel freezing mechanism. We pretrain a series of Mixture-of-Experts Vision Encoders (MoE-ViE) across a range of sizes, all consistently outperforming their dense counterparts. Our largest model matches the zero-shot performance of a SOTA encoder 1.7x its size at 76% of its latency. When aligned with an LLM, MoE-ViE surpasses all compared encoders on image and video benchmarks, including those with up to 5x more activated parameters. Code is available at https://github.com/facebookresearch/moe_vie.

cs.CV↗

arXiv2026

Spectral-Domain Deep Learning of Intrinsic Scattering Operators for Arbitrarily Shaped Compact 3D Particles

Rapid prediction of optical scattering from arbitrarily shaped three-dimensional particles is important for particle optics and photonic characterization, but remains challenging because of the large variability of complex morphologies and the strong angular dependence of their scattering responses. To address both issues, a dual spectral-domain neural scattering model is introduced in which morphology and scattering are represented in physically ordered bases: particle geometry is compressed into only 256 spherical-harmonic coefficients, and the optical response is encoded by the complex T-matrix in a spherical-vector-wave basis. The morphology spectrum replaces high-dimensional Euclidean geometry representations, such as voxel grids, point clouds, or meshes, with a compact ordered descriptor, while the T-matrix represents a geometry-determined scattering operator that can be queried for different incidence directions, polarizations, and observation angles. A spectral-token Transformer trained on 50{,}000 irregular particles at 1064~nm maps the morphology spectrum directly to the T-matrix. The predicted operators recover modal structure and reproduce full-angle differential scattering maps and incidence-angle scans. Generalization to out-of-distribution synthetic shapes and natural sand-particle morphologies shows that the dual spectral architecture learns an intrinsic relation from the geometry spectrum to multipolar scattering. This establishes spectral-domain operator learning as a compact route for reusable, angle- and polarization-resolved optical scattering prediction of complex 3D particles.

cond-mat.mes-hall↗

arXiv2026

Topological observables and domain wall tension from finite temperature chiral perturbation theory

Within the framework of SU(2) chiral perturbation theory, we derive the general solution of the QCD $\theta$-vacuum for an arbitrary vacuum phase, explicitly incorporating isospin-breaking effects from the light quark mass difference, and compute the temperature dependence of the topological susceptibility, higher-order cumulants, and the domain wall tension up to next-to-leading order. We find that the topological susceptibility agrees with lattice data at low temperatures but deviates at higher temperatures as expected from the breakdown of the chiral expansion; moreover, we demonstrate that the normalized fourth-order cumulant and the domain wall tension decrease monotonically with increasing temperature, while the normalized sixth-order cumulant exhibits the opposite behavior. These results extend earlier analyses by showing how isospin breaking reshapes the full hierarchy of topological charge cumulants and the dynamics of $\theta$-vacuum domain walls, thereby offering new theoretical input on the $\theta$-vacuum properties, which are relevant for axion-related effective theories in hot QCD matter.

hep-ph↗

arXiv2025

Safe Path Planning and Observation Quality Enhancement Strategy for Unmanned Aerial Vehicles in Water Quality Monitoring Tasks

Unmanned Aerial Vehicle (UAV) spectral remote sensing technology is widely used in water quality monitoring. However, in dynamic environments, varying illumination conditions, such as shadows and specular reflection (sun glint), can cause severe spectral distortion, thereby reducing data availability. To maximize the acquisition of high-quality data while ensuring flight safety, this paper proposes an active path planning method for dynamic light and shadow disturbance avoidance. First, a dynamic prediction model is constructed to transform the time-varying light and shadow disturbance areas into three-dimensional virtual obstacles. Second, an improved Interfered Fluid Dynamical System (IFDS) algorithm is introduced, which generates a smooth initial obstacle avoidance path by building a repulsive force field. Subsequently, a Model Predictive Control (MPC) framework is employed for rolling-horizon path optimization to handle flight dynamics constraints and achieve real-time trajectory tracking. Furthermore, a Dynamic Flight Altitude Adjustment (DFAA) mechanism is designed to actively reduce the flight altitude when the observable area is narrow, thereby enhancing spatial resolution. Simulation results show that, compared with traditional PID and single obstacle avoidance algorithms, the proposed method achieves an obstacle avoidance success rate of 98% in densely disturbed scenarios, significantly improves path smoothness, and increases the volume of effective observation data by approximately 27%. This research provides an effective engineering solution for precise UAV water quality monitoring in complex illumination environments.

cs.RO↗

arXiv2025

ConfClip: Confidence-Weighted and Clipped Reward for Reinforcement Learning in LLMs

Reinforcement learning (RL) has become a standard paradigm for refining large language models (LLMs) beyond pre-training and instruction tuning. A prominent line of work is RL with verifiable rewards (RLVR), which leverages automatically verifiable outcomes (e.g., correctness or executability) to generate reward signals. While efficient, this framework faces two key limitations: First, its binary feedback is too sparse to capture the quality of the reasoning process. Second, its coarse-grained rewards potentially lead to vanishing gradients. Inspired by observations from human learning, we introduce a RL technique that integrates verifiable outcomes with the model's own confidence estimates. This joint design enriches the reward signal, providing finer-grained feedback and implicitly supervising the reasoning process. Experimental results demonstrate that our proposed method enhances RL performance across multiple datasets and reduces token consumption during inference, while incurring negligible additional training cost. Moreover, it can be used as a plug-in module to enhance other state-of-the-art RL methods.

cs.LG↗

arXiv2025

UAV-VL-R1: Generalizing Vision-Language Models via Supervised Fine-Tuning and Multi-Stage GRPO for UAV Visual Reasoning

Recent advances in vision-language models (VLMs) have demonstrated strong generalization in natural image tasks. However, their performance often degrades on unmanned aerial vehicle (UAV)-based aerial imagery, which features high resolution, complex spatial semantics, and strict real-time constraints. These challenges limit the applicability of general-purpose VLMs to structured aerial reasoning tasks. To address these challenges, we propose UAV-VL-R1, a lightweight VLM explicitly designed for aerial visual reasoning. It is trained using a hybrid method that combines supervised fine-tuning (SFT) and multi-stage reinforcement learning (RL). We leverage the group relative policy optimization (GRPO) algorithm to promote structured and interpretable reasoning through rule-guided rewards and intra-group policy alignment. To support model training and evaluation, we introduce a high-resolution visual question answering dataset named HRVQA-VL, which consists of 50,019 annotated samples covering eight UAV-relevant reasoning tasks, including object counting, transportation recognition, and spatial scene inference. Experimental results show that UAV-VL-R1 achieves a 48.17% higher zero-shot accuracy than the Qwen2-VL-2B-Instruct baseline and even outperforms its 72B-scale variant, which is 36x larger, on multiple tasks. Ablation studies reveal that while SFT improves semantic alignment, it may reduce reasoning diversity in mathematical tasks. GRPO-based RL compensates for this limitation by enhancing logical flexibility and the robustness of inference. Additionally, UAV-VL-R1 requires only 3.9GB of memory under FP16 inference and can be quantized to 2.5GB with INT8, supporting real-time deployment on resource-constrained UAV platforms.

cs.CV↗

arXiv2023

Exploring the axion potential and axion walls in dense quark matter

We study the potential of the Quantum Chromodynamics axion in hot and/or dense quark matter, within a Nambu-Jona-Lasinio-like model that includes the coupling of the axion to quarks. Differently from previous studies, we implement local electrical neutrality and $β-$equilibrium, which are relevant for the description of the quark matter in the core of compact stellar objects. Firstly we compute the effects of the chiral crossover on the axion mass and self-coupling. We find that the low energy properties of axion are very sensitive to the phase transition of Quantum Chromodynamics, in particular, when the bulk quark matter is close to criticality. Then, for the first time in the literature we compute the axion potential at finite quark chemical potential and study the axion domain walls in bulk quark matter. We find that the energy barrier between two adjacent vacuum states decrease in the chirally restored phase: this results in a lower surface tension of the walls. Finally, we comment on the possibility of production of walls in dense quark matter.

hep-ph↗

arXiv2022

Influence of particle geometry on dispersion force

Dispersion forces (van der Waals force and Casimir force) originating from quantum fluctuations are crucial in the cohesion of microscale and nanoscale particles. In reality, these particles have a variety of irregular shapes that differ considerably from any idealized geometry. Previous experiments have demonstrated that dispersion forces strongly depend on the geometry. Because of the nonadditivity of these forces, commonly used numerical additive methods, such as the Hamaker and Derjaguin approximations, are not suitable for calculations with complex geometries. Moreover, experimental studies are difficult to identify the contributions of the dispersion force from the many forces that constitute the cohesion. Therefore, no general law about the influence of particle geometry on dispersion forces has been established. Thus, in this paper, the fluctuating surface current (FSC) technique, an exact scattering theory-based nonadditive algorithm, was used to study this influence. To characterize complex geometries, a data-adaptive spatial filtering method was introduced to perform scale decomposition, and descriptors at three observation levels (global, local, and surface) were used. Based on the advanced geometric analyses and accurate numerical calculations, the influence of multiscale surface fluctuations on dispersion forces was determined. Furthermore, a convenient formula for predicting the dispersion forces between particles with complex shapes from the exact Lifshitz solution was established via multistage corrections.

cond-mat.mes-hall↗

arXiv2020

Finite Size Effects on the Chiral Phase Transition of Quantum Chromodynamics

We study the effect of periodic boundary conditions on chiral symmetry breaking and its restoration in Quantum Chromodynamics. As an effective model of the effective potential for the quark condensate, we use the quark-meson model, while the theory is quantized in a cubic box of size $L$. After specifying a renormalization prescription for the vacuum quark loop, we study the condensate at finite temperature, $T$, and quark chemical potential, $μ$. We find that lowering $L$ leads to a catalysis of chiral symmetry breaking. The excitation of the zero mode leads to a jump in the condensate at low temperature and high density, that we suggest to interpret as a gas-liquid phase transition that takes place between the chiral symmetry broken phase (hadron gas) and chiral symmetry restored phase (quark matter). We characterize this intermediate phase in terms of the increase of the baryon density, and of the correlation length of the fluctuations of the order parameter: for small enough $L$ the correlation domains occupy a substantial portion of the volume of the system, and the fluctuations are comparable to those in the critical region. For these reasons, we dub this phase as the {\it subcritical liquid}. The qualitative picture that we draw is in agreement with previous studies based on similar effective models. We also clarify the discrepancy on the behavior of the critical temperature versus $L$ found in different models.

hep-ph↗

arXiv2019

Thermodynamic Geometry of the Quark-Meson Model

We study the thermodynamic geometry of the Quark-Meson model, focusing on the curvature, $R$, around the chiral crossover at finite temperature and baryon chemical potential. We find a peculiar behavior of $R$ in the crossover region, in which the sign changes and a local maximum develops; in particular, the height of the peak of $R$ in the crossover region becomes large in proximity of the critical endpoint and diverges at the critical endpoint. The appearance of a pronounced peak of $R$ close to the critical endpoint supports the idea that $R$ grows with the correlation volume around the phase transition. We also analyze the mixed fluctuations of energy and baryon number, $\langleΔUΔN\rangle$, which grow up substantially in proximity of the critical endpoint: in the language of thermodynamic geometry these fluctuations are responsible for the vanishing of the determinant of the metric, which results in thermodynamic instability and are thus related to the appearance of the second order phase transition at the critical endpoint.

hep-ph↗