arXiv subjects

Fan Wang

Publications and source records attributed to Fan Wang.

At least 37 records · Page 2Linked to original sources

arXiv2026

AsmEvo: Agentic Assembly-Level Optimization of AMD GPU Kernels with Functional Equivalence Verification

High-performance ML systems increasingly rely on GPU kernels whose editable source is unavailable, generated, or too distant from final machine code to expose remaining optimizations. Existing LLM kernel optimizers and autotuners mainly operate on CUDA, Triton, HIP, or tensor-program source and validate against reference implementations. We study a stricter setting: optimizing an already compiled AMDGPU code object, where the deployed binary is the only behavioral oracle. We present AsmEvo, an agentic assembly-level optimizer for AMD GPU kernels. Given an AMDGPU code object K0, AsmEvo reconstructs a reassemblable representation, proposes low-level edits with a long-horizon agent, rebuilds an ABI-preserving optimized object, and accepts candidates only after differential verification against K0 under identical launches. AsmEvo combines code-object recovery, metadata-aware rebuilding, profiling-guided hot-window editing, correctness-gated timing, and conservative in-place patch fallback. We conduct extensive experiments with AsmEvo on various AMD GPU kernels. On MI308X, AsmEvo improves 29 of 30 selected KernelBench kernels, reaching 1.35x geometric-mean and 3.88x maximum speedup. On MI300X production workloads, it improves all evaluated AITer binaries and vLLM/SGLang Triton assembly kernels, reaching 1.09x/1.31x and 1.18x/1.34x geometric-mean/maximum speedups, respectively, while preserving functional equivalence.

cs.CL↗

arXiv2026

Improving Generalization Robustness of Multimodal RLVR

Reinforcement Learning with Verifiable Rewards (RLVR) makes Multimodal Large Language Models more accurate, but the gains are brittle: simply paraphrasing a question or changing the prompt template can degrade them, which challenges reliable deployment in high-stakes scenarios like medical VQA. We trace this to two issues of the standard RL objective. First, the binary verifier conflates format with content, so the reward signal cannot tell a wrong answer apart from a misformatted one. Second, the training distribution covers only a thin slice of the real-world prompts that the model might meet at deployment, so policies that perform well on the training distribution can behave differently under unseen prompts during test. Both failures call for a robust post-training method that helps the policy cover a broader distribution of semantically equivalent prompts, and we identify two measures that help achieve this objective: separating format from semantics in the reward, and applying policy invariance across perturbed prompts with equivalent semantics. We therefore propose Prompt-Invariant RLVR (PIRL), consisting of a dynamic trinary reward and a consistency regularizer based on an embedding-space adversary. Under stress testing, PIRL's average accuracy on benchmarks drops by only $\le 1\%$, where GRPO drops ~3%. On dynamic evaluation, PIRL also achieves the smallest performance drop.

cs.AI↗

arXiv2026

SynBoost: A Synergistic Framework for Fast Sampling of Diffusion Models

Diffusion probabilistic models (DPMs) have demonstrated remarkable success in visual generation. However, their iterative sampling mechanism results in slow inference speeds. While reducing sampling steps offers an intuitive acceleration strategy, it introduces significant discretization error. Existing fast samplers have made substantial progress in mitigating this error through high-order solvers, yet further optimization appears constrained. This limitation prompts a critical question: can sampling efficiency be advanced beyond current paradigms? In this work, we re-examine the composition of sampling errors and identify two distinct components: the well-studied discretization error and the under-explored approximation error. By implementing a dual-error disentanglement strategy, we elucidate the dynamic interplay between these error types across sampling steps. Building on this empirical analysis, we propose SynBoost, a unified and training-free acceleration framework that simultaneously addresses both error sources to minimize total sampling error. Concretely, it mitigates approximation error by blending the current noise estimation with a more accurate prediction from a larger timestep. SynBoost seamlessly integrates with existing samplers, substantially enhancing their speed and output quality, particularly in regimes with extremely few steps. We validate our framework through extensive experiments across unconditional and conditional generation tasks, encompassing both pixel-space and latent-space DPMs.

cs.CV↗

arXiv2026

BAMU: Bitstream-Aware Marginal-Utility Allocation for Frozen Pretrained Neural Speech Codecs

Pretrained neural speech codecs typically use a fixed residual vector quantization (RVQ) depth for all frames, ignoring temporal variation in quantization difficulty. We propose BAMU, a bitstream-aware dynamic RVQ allocation framework for frozen pretrained codecs. A lightweight, rate-independent predictor estimates frame- and layer-wise marginal latent-distortion reductions, while a constrained allocator selects prefix-valid depths under an exact serialized-size budget. Experiments on EnCodec and DAC over LibriSpeech, together with VCTK evaluation, show consistent EnCodec gains and DAC improvements mainly at medium and high rates. A 30-listener study confirms a MOS improvement from 3.449 to 3.780 over matched fixed-depth coding.

eess.AS↗

arXiv2026

Convex Neural Energy Elements: Monolithic Finite-Element Assembly of Geometry-Parameterized Neural Operators with Stability and Error Guarantees

Extending the neural-operator element method from individually trained, fixed-geometry neural elements to a library of reusable, geometry-parameterized element types fails structurally: a field-predicting operator trained by value regression induces an energy whose assembled Hessian is indefinite, and Newton converges to spurious minima (247% error) even with 1%-accurate field predictions. We introduce convex neural energy elements: each element exports a scalar energy E(g,U), architecturally convex in its boundary degrees of freedom U and smoothly parameterized by its geometry g, realized as a hypernetwork-generated positive-semidefinite quadratic form (an input-convex correction is reserved for non-quadratic physics). A regularization-nullspace principle--the regularizer's nullspace must contain the physics nullspace--removes an otherwise irreducible bias, and assembled elements inherit the classical guarantee that singular element stiffnesses yield a positive-definite global system. We prove conditional error bounds (energy-to-solution accuracy, element-count scaling, geometry generalization) and verify each experimentally. On heat conduction with elliptic holes, one trained element assembles into 2x2 to 8x8 grids and an L-shaped layout of unseen geometries at 0.6-1.0% relative L2 error, with 175x faster per-geometry setup for boundary-quantity workloads. A second trained element type mixes freely with the first in one monolithic assembly, and a three-dimensional instantiation reaches 0.23% on eight-element assemblies--the guarantees are type- and dimension-agnostic. A plane-strain elasticity element, whose physics nullspace is three-dimensional, lands on the analytically predicted regularization floors. Making the energy the learned object turns neural operators from single-use surrogates into reusable elements that inherit the assembly guarantees of the method they extend.

cs.LG↗

arXiv2026

Ground state and multiple solutions for modified autonomous fourth-order elliptic equations with Berestycki-Lions type conditions

This article establishes the existence of a ground state and infinitely many solutions for the modified fourth-order elliptic equation: \[ \begin{aligned} \left\{ \begin{array}{ll} Δ^2 u - Δu + u - \frac{1}{2}uΔ(u^2) = f(u), & \text{in } \mathbb{R}^N, u \in H^2(\mathbb{R}^N), \end{array} \right. \end{aligned} \] where $4 < N \leq 6$ and$f:\mathbb{R}\rightarrow\mathbb{R}$ is a nonlinearity of Berestycki-Lions type. For the ground state solution, we develop a novel approach that combines Jeanjean's technique with a Pohozaev-Palais-Smale sequence construction. When $f$ is odd, we prove infinite multiplicity of radially symmetric solutions via minimax methods on a topologically constrained comparison functional. This work resolves the lack of results for this autonomous problem under almost the weakest nonlinearity conditions.

math.AP↗

arXiv2026

Sub-40 nm resolution deep tissue imaging by image scanning emission saturation nanoscopy

The development of deep-tissue super-resolution imaging serves as an essential bridge toward non-invasive in vivo optical observation. However, there remain challenges to balance spatial resolution, imaging depth and phototoxicity. Here, we present a nanoscopy namely Image Scanning Emission Saturation (ISES) nanoscopy, achieving a lateral resolution of 37 nm, 1/25th of the excitation wavelength, at an imaging depth of 200 μm. Using a 976-nm doughnut-shaped excitation beam within an imaging-scanning microscopy configuration, we apply saturation-based point spread function (PSF) engineering and pixel-level confocal-pinhole enhancement to improve spatial resolution. As the high- and low-frequency components of the image OTF are concurrently acquired in a single scan via different camera pixels, Fourier-domain fusion can be employed with a single scanning dataset to further improve image quality. Compared with the traditional doughnut excitation beam-based adaptive pixel reassignment method, our strategy preserves the original frequency distributions and mitigates reconstruction artifacts in complex biological sample imaging. This strategy is generalizable and compatible with a variety of probes displaying saturation behavior. Beyond enabling a versatile and practical approach for deep tissue super-resolution imaging, it also informs the development of next-generation nanoprobes for imaging.

physics.optics↗

arXiv2026

BladeYOLO: Wind Turbine Blade Defect Detection with Limited Annotations and Weak-Saliency Awareness

Wind turbine blade defect detection remains highly challenging in real-world inspection scenarios due to limited on-site data and the subtle visual characteristics of defects. In practice, blade defects are often small-scale, low-contrast, and difficult to distinguish from complex backgrounds, which significantly limits the robustness of existing detectors. To address these challenges, we propose BladeYOLO, a defect detection framework for wind turbine blades. Specifically, we integrate a Vision Transformer (ViT) backbone initialized with DINOv3 self-supervised pre-trained weights into YOLOv12-L, enabling the transfer of large-scale generic visual priors to blade defect detection and improving feature representation under limited training annotations. To enhance the perception of subtle defects, we further develop a Mamba-guided Weak-Defect Enhancement module, which consists of a Detail-Enhanced Multi-scale Branch for preserving high-frequency structural cues and a Cross-Mamba module for progressively propagating high-level semantic guidance to shallow features. In addition, we introduce a lightweight Style-Injector module that captures environment-related style information via Fourier decomposition and injects it into selected ViT self-attention layers, thereby improving robustness against environment-induced appearance variations. Extensive experiments demonstrate that BladeYOLO achieves superior performance on the WTBlade-Defect dataset, with additional annotation-budget experiments showing its favorable performance under reduced training annotations. Evaluation on the public Wind Surface Defect dataset further provides supportive evidence for the cross-dataset robustness of BladeYOLO. In particular, on this public dataset, BladeYOLO outperforms the best competing method by 3.5\% in mAP$_{50}$ and 2.5\% in mAP$_{50-95}$.

cs.CV↗

arXiv2026

GLI-AL: A Multi-Modal Glioma MRI Label Resource with Unified Anatomy-Lesion Labels

Existing BraTS-GLI datasets provide a widely used benchmark for adult glioma MRI segmentation, but their task definition focuses on tumor subregions and does not systematically represent coexisting white matter hyperintensities (WMH). In joint segmentation settings, such unlabeled abnormalities introduce task-specific label noise by treating pathological regions as normal tissue. To address this limitation, we introduce BraTS-GLI Anatomy-Lesion, a controlled-access, labels-only derived resource built from the BraTS 2023-GLI training cohort. The resource provides 1,251 unified eight-class anatomy-lesion label sets aligned with the original four-modal MRI cases, including image-repair labels for 116 cases requiring repaired imaging inputs. The cohort is organized into a 394-case purified subset and an 857-case extended subset, with case-level metadata covering label source, image-repair requirements, quality-control status, access conditions, checksums, and release boundaries. Compared with the original BraTS-GLI annotations, the resource substantially expands foreground supervision by incorporating healthy brain tissues and previously unlabeled coexisting abnormalities within a unified label space. A validation study using MedNeXt and T1/FLAIR inputs suggests that WMH-aware supervision preserves healthy-tissue segmentation performance across both in-domain GLI and external WMH datasets, while improving sensitivity to coexisting lesions relative to noisy-control training. The resource is intended for scientific research and supports joint anatomy-lesion supervision, label-noise analysis, and reproducible evaluation. Data are available at https://www.synapse.org/Synapse:syn75210889/wiki/, and code is available at https://github.com/xyx200/brats-gli-anatomy-lesion-code. The data resource DOI is https://doi.org/10.7303/SYN75210889.

cs.CV↗

arXiv2026

On the Schrödinger--Bopp--Podolsky system with indefinite potential: ground states, multiplicity and exponential decay

In this paper, we study the Schrödinger--Bopp--Podolsky system \begin{equation*} \begin{cases} -Δu + V(x)u + ϕu = f(x,u), & \text{in } \mathbb{R}^3, -Δϕ+ a^2 Δ^2 ϕ= 4πu^2, & \text{in } \mathbb{R}^3. \end{cases} \end{equation*} We consider the case where the potential $V$ is indefinite so that the Schrödinger operator $-Δ+ V$ has a finite-dimensional negative space. Under suitable assumptions on the potential $V$ and nonlinearity $f(x,u)$, we prove the existence of nontrivial solutions via a local linking argument and Morse theory. Moreover, these solutions are shown to decay exponentially at infinity. Additionally, a ground state solution is obtained by minimization techniques. Finally, if $f(x,u)$ is odd with respect to $u$, we obtain an unbounded sequence of solutions using the symmetric mountain pass theorem.

math.AP↗

arXiv2026

Astra: a generalizable report generation foundation model for 3D computed tomography

Interpreting computed tomography (CT) requires review of hundreds of volumetric slices and remains time-intensive and expertise-dependent. Automated CT report generation offers a promising route to improving clinical efficiency, yet the field still lacks a generalizable CT report generation foundation model that supports multi-region reporting and remains robust across external real-world cohorts. Intrinsic inconsistencies in reporting style and diagnostic terminology across cohorts make naive joint training difficult. Here we present Astra, a generalizable CT report generation foundation model developed on 90,678 thoracoabdominal CT-report pairs collected from five sites worldwide (CTRgDB), comprising 353,671 abnormalities spanning eight organ systems. By harmonizing report style and further refining diagnostic consistency via reinforcement learning, Astra achieves style-consistent and diagnostically accurate report generation across diverse anatomical regions and institutions. Evaluated on CTRgDB and six external cohorts, Astra achieves state-of-the-art performance with a 38.4% average improvement in fine-grained diagnostic metrics (P<0.001). Deployed at external clinical sites without any site-specific fine-tuning, Astra accelerated chest report drafting by 29.6% and improved abdominal report completeness by 11.3% among junior and mid-level radiologists (P<0.001). Furthermore, Astra demonstrates broad utility as a foundation for CT AI development, improving downstream diagnostic performance and scaling vision-language pretrain through high-quality report synthesis. Overall, Astra serves as a broadly accessible clinical assistant and a pivotal infrastructure for the next generation of AI-powered healthcare. The code for Astra is publicly available at https://github.com/zh-Wang-Med/Astra.

cs.CV↗

arXiv2026

Program-Synthesis-Driven Autodesign of Universal Unitary Operators

We demonstrate that AI-driven program synthesis can autonomously discover fundamental strategies for decomposing unitary matrices in photonic networks. By extending DreamCoder to complex-valued linear algebra, the system generates decomposition programs achieving the minimal $N(N-1)/2$ Mach-Zehnder interferometers, distinct from both Reck and Clements architectures. Learned programs encode dimension-agnostic invariants: strategies discovered for $5 \times 5$ matrices generalize to higher dimensions such as $64 \times 64$. The discovered programs encode interpretable, dimension-agnostic construction rules. These rules generalize across matrix sizes without retraining, demonstrating that autonomous program synthesis can serve as a scalable paradigm for algorithm discovery and the automated design of universal unitary operators. Beyond universal decompositions, the system automatically exploits matrix structure to reduce the interferometer count below the universal theoretical bound. For instance, for Householder matrices, it discovers a dimension-independent rule that requires only $2N-3$ MZIs. This achieves linear, rather than quadratic, scaling and generalizes to arbitrary $N$ without retraining. For matrices obtained from the singular value decomposition of sparse matrices, reductions generally increase with sparsity, reaching up to 38% fewer MZIs than the universal theoretical bound $N(N-1)/2$ at 95% sparsity. These MZI reductions translate directly into practical hardware benefits for scalable photonic implementations. Taken together, the system functions as a single unified engine that discovers both universal decomposition rules and matrix-specific optimizations, without being provided with the structural or analytical properties of the input matrices.

physics.optics↗

arXiv2026

StateLinFormer: Stateful Training Enhancing Long-term Memory in Navigation

Effective navigation intelligence relies on long-term memory to support both immediate generalization and sustained adaptation. However, existing approaches face a dilemma: modular systems rely on explicit mapping but lack flexibility, while Transformer-based end-to-end models are constrained by fixed context windows, limiting persistent memory across extended interactions. We introduce StateLinFormer, a linear-attention navigation model trained with a stateful memory mechanism that preserves recurrent memory states across consecutive training segments instead of reinitializing them at each batch boundary. This training paradigm effectively approximates learning on infinitely long sequences, enabling the model to achieve long-horizon memory retention. Experiments across both MAZE and ProcTHOR environments demonstrate that StateLinFormer significantly outperforms its stateless linear-attention counterpart and standard Transformer baselines with fixed context windows. Notably, as interaction length increases, persistent stateful training substantially improves context-dependent adaptation, suggesting an enhancement in the model's In-Context Learning (ICL) capabilities for navigation tasks.

cs.LG↗

arXiv2026

UnderOneFacade: Worldwide Facade Semantic Segmentation Benchmark Dataset

Globally consistent semantic digital twins require centimeter-accurate and geographically transferable 3D facade segmentation. However, progress in facade parsing is limited by the lack of large-scale, standardized benchmarks for evaluating cross-domain generalization. Existing datasets are geographically narrow, semantically inconsistent, or insufficiently precise. We introduce UnderOneFacade, the largest cross-country and cross-continent 3D facade benchmark to date, comprising centimeter-accurate point clouds with hierarchical, harmonized, and architecturally grounded semantic labels totaling 2.7 billion annotated points. Through a systematic evaluation of representative point-, graph- and transformer-based architectures, we show that current methods struggle to recognize fine-grained architectural elements and degrade significantly across geographic domains, with the best models achieving only up to 33 IoU on the fine-grained LoFG3 benchmark. By combining geometric precision with standardized semantics at unprecedented scale, UnderOneFacade establishes a rigorous benchmark for developing robust and transferable 3D segmentation models. The dataset, evaluation scripts, and pretrained models will be released upon publication.

cs.CV↗

arXiv2026

Lumos-Nexus: Efficient Frequency Bridging with Homogeneous Latent Space for Video Unified Models

Connector-based video unified models have demonstrated strong capability in instruction-grounded video synthesis, but integrating a large high-fidelity generator into the unified training loop is computationally prohibitive, limiting achievable visual quality. We therefore propose Lumos-Nexus, a training-efficient unified video generation framework that facilitates the development of strong reasoning-driven generation capabilities while significantly enhancing visual fidelity. Lumos-Nexus adopts a two-stage design: 1) During training, only a lightweight generator is aligned with the understanding block to learn to take in reasoning-driven semantic control. 2) During inference, we introduce Unified Progressive Frequency Bridging (UPFB) to progressively hand off generation to a high-capacity pretrained generator in the shared latent space, enabling coarse-to-fine refinement and producing high-fidelity videos without compromising reasoning quality. To fill the gap in reasoning-driven video generation benchmarks, we introduce VR-Bench, which assesses a model's capability to translate inferred intent into coherent and semantically aligned video content. Extensive experiments demonstrate that Lumos-Nexus achieves substantial gains in visual realism and temporal coherence on VBench, while exhibiting strong reasoning-based generative performance on VR-Bench. Code and models are available at https://jiazheng-xing.github.io/nexus-lumos-home/.

cs.CV↗

arXiv2026

Hessian sparsity-constrained self-supervised network for near-infrared single-photon single-pixel imaging

Near-infrared (NIR) imaging has emerged as an important technology for night vision, remote sensing, and biological imaging, yet conventional array-detector-based systems are often limited by insufficient sensitivity, high cost, and substantial dark noise. Single-pixel imaging (SPI) offers an attractive alternative, enabling single-photon-level NIR imaging by using a cost-effective single-element detector. Nevertheless, SPI remains restricted by photon noise, leading to degraded imaging quality and limited frame rate under extremely low photon flux conditions. Here, we present a Hessian sparsity-constrained self-supervised network (HS3N) for single-photon NIR SPI, which can suppress noise and enable high-fidelity and real-time imaging under ultra-low illumination conditions. The HS3N integrates the physical forward model of SPI with an untrained neural network regularized by both sparsity priors and Hessian-based structural constraints, enabling effective noise suppression while preserving structural fidelity and continuity. Both simulated and experimental results demonstrate that HS3N enables high-fidelity reconstructions under ultra-low NIR photon levels down to ~0.01 photons per pixel. Furthermore, we demonstrate its dynamic capability by monitoring the dynamic evolution and detachment of infrared-absorbing droplets, at a frame rate of ~20 Hz under ~0.19 photons per pixel, highlighting its potential for high-sensitivity infrared inspection. The proposed reconstruction framework paves the way for practical NIR imaging in extreme low light conditions, which can be extended to visible, mid-infrared or terahertz imaging, offering broad potential for photon-efficient sensing across a wide spectral range.

physics.optics↗

arXiv2026

Ultrasensitive infrared-to-visible artificial vision via self-evolving projection guided by single-pixel detection

Infrared detection and visualization are essential for augmenting human perception across diverse fields, ranging from night vision to industrial inspection and bio-imaging. Conventional infrared cameras are often hindered by high cost, bulky architecture, and complex fabrication requirements. Upconversion sensing systems offer a pixel-free and cost-effective alternative solution by upconverting infrared photons into visible-light signals. However, existing upconversion systems suffer from limitations such as high operating voltages, low quantum efficiency, which prevent their applications in photon-starved environments. Here, we report self-evolving infrared-to-visible upconversion with single-pixel detection (SIVIS) that enables real-time upconverted visualization under photon-starved conditions by integrating self-evolving projection with single-pixel sensing. SIVIS iteratively optimizes illumination patterns with a digital micromirror device based on real-time feedback from a single-pixel infrared detector. This self-evolving process enables the autonomous reconstruction of the target's geometric profile. Simultaneously, it projects a co-modulated visible beam onto the object itself or an adjacent screen, rendering the infrared target directly perceptible to the naked eye in real-time. SIVIS achieves sensing and projection without latency under an ultra-low infrared detection limit of 0.11 photons per pixel per frame (sub-pW -cm2 level) benefited from the high sensitivity. Furthermore, we also validate SIVIS to decrypt infrared-encoded anti-counterfeiting features and visualize vascular-like structures embedded within biological tissues. This photon-feedback-driven artificial vision framework offers a scalable and adaptive solution for ultrasensitive infrared vision, opening promising avenues for night vision, biomedical imaging, and sensing under extreme low-light conditions.

physics.optics↗

arXiv2026

UniGP: Taming Diffusion Transformer for Prior-Preserved Unified Generation and Perception

Recent advances in diffusion models have shown impressive performance in controllable image generation and dense prediction tasks. However, existing approaches typically treat diffusion-based controllable generation and dense prediction as separate tasks, overlooking the potential benefits of jointly modeling the heterogeneous distributions. In this work, we introduce UniGP, a framework built upon MMDiT, which unifies controllable generation and dense prediction through simple joint training, without the need for complex task-specific designs or losses, while preserving the backbone's versatile priors. By learning controllable generation and prediction under different conditions, our model effectively captures the joint distribution of image-geometry pairs. UniGP is capable of versatile controllable generation, dense prediction, and joint generation. Specifically, the proposed UniGP consists of DUGP and a unified dataset training strategy. The former, following the principle of Occam's razor, uses only a copied image branch of MMDiT to model dense distributions beyond RGB, while the latter integrates heterogeneous datasets into a unified training framework to jointly model generation and perception tasks. Extensive experiments demonstrate that our unified model surpasses prior unified approaches and performs on par with specialized methods. Furthermore, we demonstrate that multi-task joint training provides complementary benefits: generative priors enrich perceptual details, while perceptual learning improves structural alignment in generation.

cs.CV↗