arXiv subjects

Pengfei Wang

Publications and source records attributed to Pengfei Wang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Regularized Offline Policy Optimization with Posterior Hybrid Bayesian Belief

Offline reinforcement learning (RL) aims to optimize policies from pre-collected datasets. A bottleneck of this paradigm is managing epistemic uncertainty, which arises from limited data coverage (sample-level) and the ambiguity in identifying transition dynamics from finite data (model-level). To provide a unified quantification of these uncertainties, Bayesian RL has been proposed by treating the dynamics model as a random variable and maintaining a corresponding belief. Despite its theoretical appeal, policy optimization in Bayesian RL remains computationally challenging as it requires solving composite objectives with expectations. Prior methods either employ search-based techniques with poor computational scalability or impose restrictive posterior assumptions that sacrifice the adaptability of Bayesian RL. To address these limitations, we propose Posterior Hybrid Bayesian Belief (PhyB), which reformulates the expectation as a convex combination over a subset of dynamics models. Theoretical analysis demonstrates that the objective discrepancy induced by this approximation remains bounded. Based on PhyB, we develop an iterative regularized policy optimization algorithm that provides metric-agnostic guarantees for monotonic improvement until convergence. Empirical results demonstrate that PhyB achieves state-of-the-art performance on various benchmarks.

cs.AI↗

arXiv2026

A re-entrant chip-free-space photonic interface for telecom-to-Rubidium spectroscopy

Photonic integrated circuits (PICs) generate, route, and process light with high efficiency, scalability, and functional density on a single chip. Yet the tightly confined on-chip modes can not easily access or effectively interact with atomic vapors, fluids, gain media, and biological samples. Existing approaches require bringing the medium onto the chip or into a weak, tightly confined evanescent field, which restricts the interaction volume and the range of accessible media. Here, we demonstrate a re-entrant chip-free-space interface in which a thin-film lithium niobate circuit frequency-doubles telecom light, emits the 780~nm field through a Rubidium vapor cell, and recollects the reflected probe on the same chip. This emit-interact-recollect loop resolves the saturated absorption spectrum and stabilizes the telecom laser to within $\pm 280$~kHz over 2 hours. Our study paves an route to embed external media into PICs through the re-entrant photonic interface.

physics.optics↗

arXiv2026

Spatial-Temporal Expert Learning for Video-based Person Re-identification

Video-based person re-identification (Re-ID) aims to retrieve the same identity in the query video clips from the gallery video clips. To solve this problem, exploiting fine-grained features is of great importance, especially when discriminating identities that are similar in appearance. In this paper, we propose to enhance the ability to explore fine-grained information with a novel input-aware extendable expert module. Instead of updating the network parameters with every sample in the dataset, we aim to train the experts within specific subsets that only contain similar samples and promote their ability to exploit fine-grained information within these similar samples. To achieve this goal, we incorporate two mechanisms in this module: input-aware expert selection mechanism and spatial-temporal selection mechanism. The first mechanism dynamically activates a set of experts on subsets of similar samples, pushing the experts to exploit subtle differences between these similar samples, while the second one further increases their sensitivity to the fine-grained differences in spatial and temporal aspects and allows the experts to dynamically utilize them for different input samples. In addition, to facilitate the expert module, we design an extendable scheme that allows the module to flexibly add new experts when necessary. As a result, our method achieves outstanding performance on two large-scale datasets.

cs.CV↗

arXiv2026

Empowering a Single-Frequency GNSS Receiver to Achieve High-Precision Positioning with Relative Observations

Global Navigation Satellite System (GNSS) navigation is widely used to provide absolute, outdoor positioning in field robotics. Advances in Real-Time Kinematic (RTK) technology can achieve centimeter-level accuracy, facilitating autonomous navigation tasks. However, the cost and extra infrastructure used for RTK still hinder the application and more cost-effective solutions are desired. In this letter, we present a novel tightly-coupled state estimation framework that achieves high-precision localization by using low-cost, mass-market single-frequency GNSS receivers with any relative motion sensors (e.g., wheel encoder, camera, LiDAR). We propose a sliding-window factor graph that integrates generic relative motion with global epoch-to-anchor constraints derived from continuous carrier phase tracking. To eliminate the reliance on physical base stations, we introduce a virtual anchor mechanism: upon the initial observation of a satellite, its state is locked as a virtual reference to establish global epoch-to-anchor constraints. By substituting multi-frequency hardware redundancy with single-frequency multi-modal kinematic priors and a robust cycle-slip recovery technique, our approach ensures carrier-phase integrity on cheap receivers. Extensive real-world experiments on heterogeneous low-cost sensor suites validate that our method improves the accuracy of a single-frequency receiver from several meters to decimeter-level precision across diverse environments, providing an accurate, cost-effective and reliable alternative for autonomous navigation.

cs.RO↗

arXiv2026

MLCR: Multi-Level Cue Refinement for Long-Term Multimodal Action Quality Assessment

Long-term multimodal action quality assessment (AQA) evaluates action execution in several-minute audiovisual sequences by mining discriminative quality cues for score prediction. Existing multimodal methods usually model entire sequences with a single temporal encoder and fuse modality features by direct alignment or concatenation, causing key cues to be obscured by global trends, weakened by modal redundancy, and distorted during one-shot score mapping. To address this issue, we reformulate long-term multimodal AQA as a quality cue organization problem and propose MLCR, a multi-level cue refinement framework. MLCR organizes quality evidence at three levels: intra-modal representation, cross-modal interaction, and stage-wise aggregation. Specifically, the intra-modal decoupling encoder (IMDE) preserves modality identity while refining global temporal context and local frequency details. The cross-modal dynamic complementarity-aware retrieval (CMDCR) module retrieves incremental evidence conditioned on the evolving fused state and suppresses redundant responses. The stage-wise multimodal integration (SMI) block progressively accumulates intra-modal and cross-modal cues to refine the fused representation. Experiments on the Rhythmic Gymnastics and Fis-V datasets show that MLCR achieves the best or second-best performance in both Spearman correlation and prediction error, demonstrating its effectiveness and robustness.

cs.CV↗

arXiv2026

scLLM-DSC: LLM-Knowledge Enhanced Cross-Modal Deep Structural Clustering for Single-Cell RNA Sequencing

Clustering is fundamental to scRNA-seq analysis, serving as a cornerstone for identifying cell populations and resolving tissue heterogeneity. However, existing methods focus on mining numerical statistical patterns, suffering from semantic agnosticism by neglecting the intrinsic biological functions encoded by genes. While Large Language Models (LLMs) offer promising semantic capabilities, their direct adaptation to cell clustering is hindered by the structural mismatch between generative pre-training objectives and discriminative downstream tasks. To bridge this gap, we propose scLLM-DSC, a novel LLM-Knowledge Enhanced Cross-Modal Deep Structural Clustering framework. Diverging from data-driven paradigms, scLLM-DSC establishes a semantically-grounded representation by synergizing two views: a Knowledge-Driven Semantic View derived from NCBI gene priors and contextualized Cell2Sentence embeddings, and a Structure-Aware Topological View extracted via a graph-guided encoder. Crucially, we introduce a cross-modal contrastive alignment mechanism to enforce consistency between biological semantics and transcriptomic features within a unified latent space. Extensive benchmarks demonstrate that scLLM-DSC significantly outperforms eleven state-of-the-art baselines in clustering accuracy.

cs.LG↗

arXiv2026

Robust Regularized Policy Iteration under Transition Uncertainty

Offline reinforcement learning (RL) enables data-efficient and safe policy learning without online exploration, but its performance often degrades under distribution shift. The learned policy may visit out-of-distribution state-action pairs where value estimates and learned dynamics are unreliable. To address policy-induced extrapolation and transition uncertainty in a unified framework, we formulate offline RL as robust policy optimization, treating the transition kernel as a decision variable within an uncertainty set and optimizing the policy against the worst-case dynamics. We propose Robust Regularized Policy Iteration (RRPI), which replaces the intractable max-min bilevel objective with a tractable KL-regularized surrogate and derives an efficient policy iteration procedure based on a robust regularized Bellman operator. We provide theoretical guarantees by showing that the proposed operator is a $γ$-contraction and that iteratively updating the surrogate yields monotonic improvement of the original robust objective with convergence. Experiments on D4RL benchmarks demonstrate that RRPI achieves strong average performance, outperforming recent baselines including percentile-based methods on the majority of environments while remaining competitive on the rest. Moreover, RRPI exhibits robust performance by aligning lower $Q$-values with high epistemic uncertainty, which prevents the policy from executing unreliable out-of-distribution actions.

cs.AI↗

arXiv2026

DepthMaster: Unified Monocular Depth Estimation for Perspective and Panoramic Images

While monocular depth estimation has achieved significant progress, achieving generalized metric depth estimation for both narrow field-of-view (FoV) perspectives and $360^\circ$ panoramas remains an unsolved challenge. Existing methods are often tailored to specific camera types and struggle to produce accurate metric depth that generalizes across diverse settings. This limitation stems from two key challenges: the inherent geometric discrepancy between perspective and panoramic cameras, and the scarcity of panoramic training data with metric annotations. In this work, we introduce DepthMaster, a unified metric depth estimation framework. Rather than employing specialized networks to learn spherical distortions, we reformulate the problem by decomposing panoramic images into overlapping perspective patches. Crucially, distinct from prior projection-based methods that rely on ad-hoc architectural modifications to handle boundaries, we introduce a novel Correspondence Consistency Loss (CCL) and inject virtual projection cameras as geometric priors, allowing us to seamlessly stitch the patches while avoiding specialized operators and keeping the backbone largely compatible with standard Transformer designs. This strategy also resolves the geometric differences by unifying all inputs into a canonical perspective representation, and effectively circumvents data scarcity by directly unlocking powerful metric priors from vast perspective datasets. Trained on a mixed dataset that contains only one panorama dataset, DepthMaster achieves state-of-the-art zero-shot performance on 13 diverse datasets, outperforming not only universal methods but also leading specialist models in both perspective and panoramic domains.

cs.CV↗

arXiv2026

OffsetCrust: Variable-Radius Offset Approximation with Power Diagrams

Offset surfaces, defined as the Minkowski sum of a base surface and a rolling ball, play a crucial role in geometry processing, with applications ranging from coverage motion planning to brush modeling. While considerable progress has been made in computing constant-radius offset surfaces, computing variable-radius offset surfaces remains a challenging problem. In this paper, we present OffsetCrust, a novel framework that efficiently addresses the variable-radius offsetting problem by computing a power diagram. Let $R$ denote the radius function defined on the base surface $S$. The power diagram is constructed from contributing sites, consisting of carefully sampled base points on $S$ and their corresponding off-surface points, displaced along $R$-dependent directions. In the constant-radius case only, these displacement directions align exactly with the surface normals of $S$. Moreover, our method mitigates the misalignment issues commonly seen in crust-based approaches through a lightweight fine-tuning procedure. We validate the accuracy and efficiency of OffsetCrust through extensive experiments, and demonstrate its practical utility in applications such as reconstructing original boundary surfaces from medial axis transform (MAT) representations.

cs.GR↗

arXiv2026

MOSS-Video-Preview: Toward Real-Time Video Understanding via Cross-Attention

Video understanding is shifting from the offline paradigm -- taking a fully recorded video as input and producing a single answer after it ends -- toward real-time interaction, in which the model perceives new frames while still replying, revises its answer as new evidence appears, and remains silent when there is nothing to say. We present MOSS-Video-Preview to validate this paradigm. Our central claim is that perception must not be blocked by generation; its natural realization is a two-channel architecture. We argue that a cross-attention backbone is better suited to real-time vision-language fusion than the prevailing decoder-only design: visual features enter through a side channel rather than joining the autoregressive sequence, so perception and generation run on separate, non-blocking pathways -- reducing the frequency of visual processing and exposing a clean channel-wise interface for independent compression. We complement this with a data synthesis pipeline that converts dense captions into real-time understanding QA whose answers are revised to match what the model has perceived so far, and we specialize an offline model on these data to elicit real-time behavior. Our model trails the strong Qwen2.5-VL-7B baseline overall -- a gap we attribute primarily to data and scale rather than the architecture -- yet attains competitive offline video and multimodal understanding, remains robust on the spatial and fine-grained temporal reasoning central to real-time use, and acquires behaviors that offline models lack: continuous perception, answer revision, and timely silence. On a single H200 with 256 frames per video, it achieves about a 5x speedup in time to first token and 2.7x higher decoding throughput, with negligible degradation in offline ability. Our study of paradigm, architecture, and data outlines a viable path toward real-time video understanding.

cs.CV↗

arXiv2026

Every Step Counts: Step-Level Credit Assignment for Tool-Integrated Text-to-SQL

Tool-integrated Text-to-SQL parsing has emerged as a promising paradigm, framing SQL generation as a sequential decision-making process interleaved with tool execution. However, existing reinforcement learning approaches mainly rely on coarse-grained outcome supervision, resulting in a fundamental credit assignment problem: models receive the same reward for any trajectory that yields the correct answer, even when intermediate steps are redundant, inefficient, or erroneous. Consequently, models are encouraged to explore suboptimal reasoning spaces, limiting both efficiency and generalization. To address this problem, we propose FineStep, a novel framework for step-level credit assignment in tool-augmented Text-to-SQL. First, we introduce a reward design with independent process rewards to alleviate the signal sparsity of outcome supervision. Next, we present a step-level credit assignment mechanism to precisely quantify the value of each reasoning step. Finally, we develop a policy optimization method based on step-level advantages for efficient updates. Extensive experiments on BIRD benchmarks show that FineStep achieves state-of-the-art performance and reduces redundant tool interactions, with a 3.25% average EX gain over GRPO at the 4B scale.

cs.CL↗

arXiv2026

Frozen LVLMs for Micro-Video Recommendation: A Systematic Study of Feature Extraction and Fusion

Frozen Large Video Language Models (LVLMs) are increasingly employed in micro-video recommendation due to their strong multimodal understanding. However, their integration lacks systematic empirical evaluation: practitioners typically deploy LVLMs as fixed black-box feature extractors without systematically comparing alternative representation strategies. To address this gap, we present the first systematic empirical study along two key design dimensions: (i) integration strategies with ID embeddings, specifically replacement versus fusion, and (ii) feature extraction paradigms, comparing LVLM-generated captions with intermediate decoder hidden states. Extensive experiments on representative LVLMs reveal three key principles: (1) intermediate hidden states consistently outperform caption-based representations, as natural-language summarization inevitably discards fine-grained visual semantics crucial for recommendation; (2) ID embeddings capture irreplaceable collaborative signals, rendering fusion strictly superior to replacement; and (3) the effectiveness of intermediate decoder features varies significantly across layers. Guided by these insights, we propose the Dual Feature Fusion (DFF) Framework, a lightweight and plug-and-play approach that adaptively fuses multi-layer representations from frozen LVLMs with item ID embeddings. DFF achieves state-of-the-art performance on two real-world micro-video recommendation benchmarks, consistently outperforming strong baselines and providing a principled approach to integrating off-the-shelf large vision-language models into micro-video recommender systems.

cs.IR↗

arXiv2026

Manifold k-NN: Accelerated k-NN Queries for Manifold Point Clouds

k-nearest neighbor (k-NN) search is a fundamental primitive in geometry processing and computer graphics. While spatial partitioning structures such as kd-trees are standard, they are often manifold-blind, failing to exploit the intrinsic low-dimensional structure of points sampled from 2-manifolds. Recent advances in dynamic programming-based nearest neighbor search (DP-NNS) leverage incrementally constructed Voronoi diagrams to accelerate queries, where each site p maintains a list of successors that progressively refine its Voronoi cell. However, DP-NNS is restricted to single nearest neighbor (k=1) searches, precluding their adoption in applications that require local neighborhood statistics. In this paper, we generalize the DP-NNS framework to support arbitrary k-NN queries for manifold-aligned data. Our approach is founded on the geometric observation that if p_i is the nearest neighbor of a query q in P, then the second nearest neighbor of q must reside either within the prefix set P_{1:i-1} = {p_1, \dots, p_{i-1}} or within p_i's successor list. By recursively extending this principle, we introduce Manifold k-NN, a recursive algorithmic scheme that significantly outperforms conventional kd-trees for manifold-aligned data. Our method achieves a 1\times--10\times speedup in volume-to-surface query scenarios and inherently supports dynamic prefix queries -- enabling k-NN searches within any subset P_{1:m} (m \leq n) with zero overhead. Furthermore, we extend the framework to support point deletion via local Delaunay updates, providing a complete suite of dynamic operations for point set modification. Comprehensive experiments on diverse geometric datasets demonstrate the efficiency and broad applicability of our approach for modern graphics pipelines. Source code is available at https://github.com/sssomeone/manifold-knn.

cs.CG↗

arXiv2026

Structural MAT: Clean and Scalable Medial Axis Simplification via Explicit Surface Correspondence

The Medial Axis Transform (MAT) is a complete shape descriptor capable of reconstructing the geometry of the original domain. A high-quality MAT should not only facilitate high-fidelity reconstruction but also capture structural features -- for instance, by aligning the MAT boundary with the locus of rolling ball centers within fillet regions. However, computing such an ideal MAT remains a significant challenge, particularly when the input is a discrete triangle mesh. In this paper, we follow the established technical pipeline of initializing the MAT via a 3D Voronoi diagram of surface samples and subsequently simplifying the Voronoi structure through a QEM-like scheme. Our key insight is to explicitly track the correspondence between MAT vertices and surface regions throughout the progressive simplification process, ensuring that the resulting MAT triangles accurately reflect the intrinsic symmetries between surface patches. We translate these geometric requirements into a suite of priority control strategies that govern the sequencing of edge collapses. Through extensive evaluation against state-of-the-art MAT algorithms, we validate the strong performance of our approach regarding runtime efficiency, structural alignment, boundary regularity, triangle quality, and robustness to noise. Our resulting MATs remain highly expressive for both articulated shapes and CAD models, even under extreme simplification -- effectively capturing the global structure of complex geometries with only a few hundred vertices.

cs.GR↗

arXiv2026

P2M++: Enhanced Solver for Point-to-Mesh Distance Queries

Point-to-mesh distance queries are fundamental in computer graphics and geometric modeling. While the state-of-the-art P2M method achieves high-speed queries via Voronoi-based localization, it suffers from prohibitive precomputation costs. Its iterative Voronoi sweep for interference detection leads to redundant predicate evaluations and scales poorly on rotationally symmetric structures (e.g., spheres, cones or cylinders), where candidate counts grow quadratically. We propose P2M++ to address these limitations through three key contributions. First, we adaptively augment the set of mesh vertices with auxiliary sites in regions of high Voronoi vertex density to localize complex interference within minimal spatial regions. Second, we reformulate interference detection as a series of sphere-triangle collision tests centered at Voronoi cell corners, which are efficiently resolved using the base mesh's BVH. Finally, we enhance runtime performance by replacing the standard kd-tree search with a faster recursive dynamic programming implementation. Experimental results demonstrate that P2M++ is 3x-10x faster than the original P2M during preprocessing and 1.5x faster in queries, with even more pronounced gains on rotationally symmetric geometries.

cs.GR↗

arXiv2026

On the Difference Between Pulsar Radio Emission Beams from the Two Poles

The long-standing assumption of symmetric radio emission beams from the two magnetic poles of pulsars is challenged by observational evidence of asymmetry and underfill. Direct testing of this symmetry remains difficult for most pulsars. As an indirect test, we collected polarization profiles of 11 interpulse pulsars observed with the Five-hundred-meter Aperture Spherical radio Telescope, MeerKAT, and Parkes. We developed a rotating vector model incorporating aberration and retardation effects to fit the position angle swings of selected pulsars, thereby determining the intrinsic emission region corresponding to the observed pulse windows. Based on both the conal and fan beam models, we compared three key parameters-beam radius, magnetic azimuth width, and emission intensity-between the intrinsic emission regions of the main pulse and interpulse. Among the eight pulsars with a confirmed double-pole geometry, none exhibits similarity in the azimuth width. Only two show potentially similar beam radii, while six demonstrate comparable emission intensities within specific parameter spaces. These results indicate that the emission beams from the two magnetic poles of a pulsar may be generally dissimilar in size, suggesting that the physical conditions governing pair production and particle acceleration differ between the two poles. The random distribution of active emission regions further implies inhomogeneity within the polar cap, which may originate from the differences in local magnetic field structure or surface properties.

astro-ph.HE↗

arXiv2026

To understand the radiative processes of pulsars and fast radio bursts with the FAST

The radiative mechanism of coherent radio emission has remained an enigma since the discovery of pulsars, even the emergence of fast radio bursts (FRBs), which exhibit similarities to the single-pulse behavior of pulsars and have opened a new view for deciphering the long-standing mystery. Besides tremendous efforts in modelling, advanced facilities matter for solving the problem. The authors review the observational breakthroughs from the Five-hundred-meter Aperture Spherical radio Telescope (FAST), which are providing pivotal insights to unravel the underlying physics of pulsars and FRBs. This study offers a novel perspective in the era when pulsars meet FRBs, and further investigations are encouraged to utilize the highly sensitive telescope, the FAST.

astro-ph.HE↗

arXiv2026

A Comprehensive Survey on Vector Database: Storage and Retrieval Technique, Challenge

As high-dimensional vector data increasingly surpasses the processing capabilities of traditional database management systems, Vector Databases (VDBs) have emerged and become tightly integrated with large language models, being widely applied in modern artificial intelligence systems. However, existing research has primarily focused on underlying technologies such as approximate nearest neighbor search, with relatively few studies providing a systematic architectural-level review of VDBs or analyzing how these core technologies collectively support the overall capacity of VDBs. This survey aims to offer a comprehensive overview of the core designs and algorithms of VDBs, establishing a holistic understanding of this rapidly evolving field. First, we systematically review the key technologies and design principles of VDBs from the two core dimensions of storage and retrieval, tracing their technological evolution. Next, we conduct an in-depth comparison of several mainstream VDB architectures, summarizing their strengths, limitations, and typical application scenarios. Finally, we explore emerging directions for integrating VDBs with large language models, including open research challenges and trends such as novel indexing strategies. This survey serves as a systematic reference guide for researchers and practitioners, helping readers quickly grasp the technological landscape and development trends in the field of vector databases, and promoting further innovation in both theoretical and applied aspects.

cs.DB↗