Searcharxiv⌕ Search

arXiv subjects

Pengyu Wang

Publications and source records attributed to Pengyu Wang.

At least 19 recordsLinked to original sources
arXiv2026

UniFed-VLM: Federated Instruction Tuning for Vision-Language Models with Multiple Heterogeneity

Vision-Language Models (VLMs) have demonstrated strong performance in multimodal understanding and generation. However, fine-tuning of VLMs typically relies on centralized data, which raises privacy concerns in certain domains (e.g. healthcare). Federated Learning (FL) provides a natural solution by enabling model training without sharing raw data. However, applying FL to VLM instruction tuning is highly challenging. VLMs have substantial parameter scales, and in real-world scenarios, clients exhibit significant heterogeneity in tasks, modalities, and model architectures. Existing methods mainly focus on simplified settings and are unable to handle such multi-dimensional heterogeneous scenarios. In this work, we study federated instruction tuning under joint heterogeneity in tasks, modalities, and model architectures. We propose UniFed-VLM, a unified federated instruction tuning framework for VLMs that addresses multiple types of heterogeneity. It consists of two key components: 1) Federated Compensated Subspace Aggregation (FedCSA), which performs subspace-aligned aggregation of parameter-efficient adapters with dynamic weighting and compensation to mitigate heterogeneity-induced conflicts; 2) Two-stage Collaborative Distillation (TCoD), which enables effective knowledge transfer across heterogeneous models via a Mutual Distillation Adapter (MDA) and a mixture-of-experts-based distillation strategy. We conduct experiments on multiple benchmark datasets, and the results show that UniFed-VLM achieves stronger average performance across diverse tasks compared with existing FL methods. The source code is available at: https://github.com/wangpengyu2004/UniFed-VLM.

cs.LG↗
arXiv2026

MOSS-VL Technical Report

We present MOSS-VL, an open vision-language model family that treats real-time interaction -- perceiving while it speaks -- as a first-class capability. It is co-designed across the stack: the language decoder attends to vision only through gated cross-attention, so the model can naturally see incoming frames while generating; a synthesized interaction corpus supervises when to speak, when to stay silent, and when to revise; and a staged curriculum concentrates all real-time-specific training in one light final stage over a strong offline foundation. Offline, MOSS-VL-Instruct is competitive at comparable scale and leads temporal-reasoning video sets. Across four streaming benchmarks, MOSS-VL-Realtime posts the best average on three (second on the fourth) among open-source streaming models, sweeping the three subsets that squarely test proactive behavior -- 66.0 vs. 37.5 for the best baseline on OmniMMI Proactive Alerting. With 11.3B parameters but visual tokens outside the decoded sequence, MOSS-VL widens its time-to-first-token advantage over same-backbone Qwen3-VL-8B from 2.8x to 5.1x as visual context grows. We release all five checkpoints, the training curriculum, and the real-time inference code at https://github.com/OpenMOSS/MOSS-VL.

cs.CV↗
arXiv2026

BM25 Wins at Scale: A Scaling Study of Retrieval-Augmented Generation Paradigms

Retrieval-augmented generation (RAG) spans lexical and dense retrieval, graph-based indexing, and agentic search, but these paradigms are usually evaluated on different benchmarks at one corpus size, leaving their accuracy-cost scaling unclear. To bridge this gap, we present a controlled study that varies corpus size along 28 strictly nested tiers spanning roughly 450-fold, while holding questions and a fixed bedrock of relevant and adversarial documents unchanged. Under one reader model and one judging protocol, we measure official accuracy, construction and query tokens, and latency. The results reveal a scale-dependent crossover rather than an unconditional winner. File-System Agent leads at the smallest shared tiers, but its sequential exploration costs 39 times more query tokens at the bedrock and becomes less effective as the search space grows. Around 10 million corpus tokens, BM25 overtakes it and leads at every larger shared tier, with a margin approaching 20 points at full scale. BM25 also anchors the low-cost end of the Pareto frontier without LLM-based construction. Dense retrieval remains efficient but less accurate, whereas graph-based RAG encounters construction walls before deployment scale and its scalable variants remain below BM25 at shared tiers. Overall, corpus growth increasingly favors global candidate ranking: lexical retrieval is the strongest scalable default, while agentic reasoning works best after ranked discovery rather than in place of it.

cs.CL↗
arXiv2026

WaveLander: A Generalizable Hierarchical Control Framework for UAV Landing on Wave-Disturbed Platforms via Reinforcement Learning

Autonomous landing of unmanned aerial vehicles (UAVs) on wave-disturbed marine platforms remains challenging due to stochastic platform motion, time-varying platform attitude, and uncertain touchdown conditions. Existing model-based methods often require accurate motion prediction and online optimization, while end-to-end learning approaches may suffer from high training complexity and limited interpretability. This paper presents WaveLander, a hierarchical control framework via reinforcement learning (RL) that decouples vertical landing decision-making from low-level flight stabilization. The RL policy maps a compact platform-relative observation to a scalar vertical velocity reference, while a conventional low-level flight controller maintains attitude stability and lateral tracking. This formulation reduces dynamic platform landing to a low-dimensional, timing-aware control problem and enables smooth landing behavior without explicit switching rules. Simulation results under randomized wave-induced platform motions show that WaveLander achieves robust landing performance and generalizes to unseen disturbance conditions, demonstrating the potential of hierarchical learning-based control for marine UAV recovery.

cs.RO↗
arXiv2026

Brownian Bridge Diffusion-Based Joint Channel Estimation and Data Detection for Jamming-Resilient Receivers

In next-generation wireless networks, the growing density of devices and limited spectrum resources pose severe jamming challenges to fragile legitimate communication links in the wireless electromagnetic environment. Crucially, when jamming overlaps with pilot and data symbols in both time and frequency domains, it inflicts a severe bottleneck on receiver-side joint estimation and detection. Existing schemes often lack an effective framework to combat such jamming contamination, thereby failing to guarantee reliable transmission. To address this issue, we propose a Brownian bridge diffusion-based joint channel estimation and data detection framework (BBD-JCED) for jamming-resilient receivers. Specifically, the proposed framework comprises two core modules: the first extracts jamming features in the short-time Fourier transform (STFT) domain and suppresses jamming samples, thereby improving the signal-to-jamming-plus-noise ratio (SJNR) of the received signal; the second introduces a Brownian bridge diffusion (BBD) process to model the evolution of the suppressed signal and the encoded bits in the presence of channel estimation errors, thereby enabling enhanced joint channel estimation and data detection. To alleviate the computational burden of the BBD process in the second module, we further derive a fast ordinary differential equation (ODE) solver that enables its low-complexity iterative evolution. Finally, we design a multi-module training algorithm to improve the data recovery capability of the proposed framework. Simulation results demonstrate that the proposed framework achieves superior bit recovery performance compared with baseline schemes while maintaining a lower number of model parameters and competitive computational complexity.

cs.IT↗
arXiv2026

Blind Room Impulse Response Identification via Reverberant Speech Spectrum Reconstruction

This paper proposes Rec-RIR for blind room impulse response (RIR) identification. Based on the convolutive transfer function (CTF) approximation, we propose a multi-task deep neural network, which sequentially removes noise and reverberation from speech recording, and estimates the CTF filter by reverberant speech spectrum reconstruction. Subsequently, a pseudo intrusive measurement process is employed to convert the CTF filter into RIR by simulating a common intrusive RIR measurement procedure. Experimental results demonstrate that Rec-RIR achieves state-of-the-art (SOTA) performance in blind RIR identification.

eess.AS↗
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

Prism: Spectral-Aware Block-Sparse Attention

Block-sparse attention is promising for accelerating long-context LLM pre-filling, yet identifying relevant blocks efficiently remains a bottleneck. Existing methods typically employ coarse-grained attention as a proxy for block importance estimation, but often resort to expensive token-level searching or scoring, resulting in significant selection overhead. In this work, we trace the inaccuracy of standard coarse-grained attention via mean pooling to a theoretical root cause: the interaction between mean pooling and Rotary Positional Embeddings (RoPE). We prove that mean pooling acts as a low-pass filter that induces destructive interference in high-frequency dimensions, effectively creating a "blind spot" for local positional information (e.g., slash patterns). To address this, we introduce Prism, a training-free spectral-aware approach that decomposes block selection into high-frequency and low-frequency branches. By applying energy-based temperature calibration, Prism restores the attenuated positional signals directly from pooled representations, enabling block importance estimation using purely block-level operations, thereby improving efficiency. Extensive evaluations confirm that Prism maintains accuracy parity with full attention while delivering up to $\mathbf{5.1\times}$ speedup.

cs.CL↗
arXiv2026

Sparser Block-Sparse Attention via Token Permutation

Scaling the context length of large language models (LLMs) offers significant benefits but is computationally expensive. This expense stems primarily from the self-attention mechanism, whose $O(N^2)$ complexity with respect to sequence length presents a major bottleneck for both memory and latency. Fortunately, the attention matrix is often sparse, particularly for long sequences, suggesting an opportunity for optimization. Block-sparse attention has emerged as a promising solution that partitions sequences into blocks and skips computation for a subset of these blocks. However, the effectiveness of this method is highly dependent on the underlying attention patterns, which can lead to sub-optimal block-level sparsity. For instance, important key tokens for queries within a single block may be scattered across numerous other blocks, leading to computational redundancy. In this work, we propose Permuted Block-Sparse Attention (\textbf{PBS-Attn}), a plug-and-play method that leverages the permutation properties of attention to increase block-level sparsity and enhance the computational efficiency of LLM prefilling. We conduct comprehensive experiments on challenging real-world long-context datasets, demonstrating that PBS-Attn consistently outperforms existing block-sparse attention methods in model accuracy and closely matches the full attention baseline. Powered by our custom permuted-FlashAttention kernels, PBS-Attn achieves an end-to-end speedup of up to $2.75\times$ in long-context prefilling, confirming its practical viability. Code available at https://github.com/xinghaow99/pbs-attn

cs.CL↗
arXiv2026

LoViF 2026 Challenge on Real-World All-in-One Image Restoration: Methods and Results

This paper presents a review for the LoViF Challenge on Real-World All-in-One Image Restoration. The challenge aimed to advance research on real-world all-in-one image restoration under diverse real-world degradation conditions, including blur, low-light, haze, rain, and snow. It provided a unified benchmark to evaluate the robustness and generalization ability of restoration models across multiple degradation categories within a common framework. The competition attracted 124 registered participants and received 9 valid final submissions with corresponding fact sheets, significantly contributing to the progress of real-world all-in-one image restoration. This report provides a detailed analysis of the submitted methods and corresponding results, emphasizing recent progress in unified real-world image restoration. The analysis highlights effective approaches and establishes a benchmark for future research in real-world low-level vision.

cs.CV↗
arXiv2026

MRG-R1: Reinforcement Learning for Clinically Aligned Medical Report Generation

Medical report generation aims to automatically produce radiology-style reports from medical images, supporting efficient and accurate clinical decision-making.However, existing approaches predominately rely on token-level likelihood training, which favors local lexical matching and leaves clinical correctness under-specified in the training objective. This behavior can be attributed to token-level likelihood optimization, which rewards surface-form agreement and therefore fails to directly encode constraints on medically accurate findings. To address this objective mismatch, we introduce a semantic-driven reinforcement learning (SRL) framework for medical report generation, named MRG-R1, which directly optimizes report-level clinical correctness rather than token-level likelihood. The key module is a clinically grounded report-level reward function, which reinforces semantic agreement in clinically relevant findings between generated and reference reports, thereby enabling learning signals that explicitly constrain medical correctness beyond surface linguistic alignment. Our evaluations show that the proposed framework improves the accuracy and coverage of clinically relevant findings in generated reports, and that MRG-R1 achieves state-of-the-art clinical efficacy on the IU X-Ray and MIMIC-CXR benchmark datasets.

cs.CL↗
arXiv2026

Not All Queries Need Deep Thought: CoFiCot for Adaptive Coarse-to-fine Stateful Refinement

Scaling test-time computation enhances LLM reasoning ability but faces a uniform computation paradox. Allocating identical resources leads to over-correction on simple tasks and insufficient refinement on complex ones. To address this, we propose CoFiCot, a coarse-to-fine adaptive framework that dynamically tailors inference strategies to problem difficulty. Specifically, we implement a multi-metric classifier that triages queries by synthesizing semantic entropy, consensus reliability, and predicted reasoning depth . This enables a differentiated refinement stage that applies efficient aggregation for simple queries while routing complex ones to a context-aware correction loop . We formalize correction as a stateful sequential propagation process , where each repair is strictly conditioned on the verified history of prior rectifications. By integrating Process Reward Models (PRMs) within this state-dependent trajectory, CoFiCot effectively bridges the gap between granular error localization and global logical coherence, preventing the context fragmentation typical of stateless refinement methods.

cs.CL↗
arXiv2026

T-linear specific heat in pressurized and magnetized Shastry-Sutherland Mott insulator SrCu2(BO3)2

The pressurized Shastry-Sutherland Mott insulator SrCu2(BO3)2 has been found to host a plaquette-singlet phase and an antiferromagnetic phase that break different symmetries spontaneously.The recent experiment showed that their transition is of a first order nature, which seems against the pursuit of exotic and deconfined degrees of freedom in this famous frustrated quantum magnet. We found a new direction in this study. By applying a magnetic field to the material, we discover that SrCu2(BO3)2 exhibits a universal and metallic T-linear specific heat behavior in a large magnetitic field range close to the pressure of zero-field first order transition between plaquette-singlet and antiferromagnetic phases. Such an unexpected gapless response from an electronically gapped Mott insulator could be attributed to magnetized Dirac spinons liberated by the combined effect of magnetic field and pressure, consistently seen from our quantum many-body thermal tensor network computation of the Shastry-Sutherland model under magnetic field. Such a robust and universal T-linear specific heat phase points out the richness of the phase diagram of the material expanded by the axes of pressure and magnetic field and is calling for new theoretical frameworks to its full explanation.

cond-mat.str-el↗
arXiv2026

WildGraphBench: Benchmarking GraphRAG with Wild-Source Corpora

Graph-based Retrieval-Augmented Generation (GraphRAG) organizes external knowledge as a hierarchical graph, enabling efficient retrieval and aggregation of scattered evidence across multiple documents. However, many existing benchmarks for GraphRAG rely on short, curated passages as external knowledge, failing to adequately evaluate systems in realistic settings involving long contexts and large-scale heterogeneous documents. To bridge this gap, we introduce WildGraphBench, a benchmark designed to assess GraphRAG performance in the wild. We leverage Wikipedia's unique structure, where cohesive narratives are grounded in long and heterogeneous external reference documents, to construct a benchmark reflecting real-word scenarios. Specifically, we sample articles across 12 top-level topics, using their external references as the retrieval corpus and citation-linked statements as ground truth, resulting in 1,100 questions spanning three levels of complexity: single-fact QA, multi-fact QA, and section-level summarization. Experiments across multiple baselines reveal that current GraphRAG pipelines help on multi-fact aggregation when evidence comes from a moderate number of sources, but this aggregation paradigm may overemphasize high-level statements at the expense of fine-grained details, leading to weaker performance on summarization tasks. Project page:https://github.com/BstWPY/WildGraphBench.

cs.CL↗
arXiv2026

A-RAG: Scaling Agentic Retrieval-Augmented Generation via Hierarchical Retrieval Interfaces

Frontier language models have demonstrated strong reasoning and long-horizon tool-use capabilities. However, existing RAG systems fail to leverage these capabilities. They still rely on two paradigms: (1) designing an algorithm that retrieves passages in a single shot and concatenates them into the model's input, or (2) predefining a workflow and prompting the model to execute it step-by-step. Neither paradigm allows the model to participate in retrieval decisions, preventing efficient scaling with model improvements. In this paper, we introduce A-RAG, an Agentic RAG framework that exposes hierarchical retrieval interfaces directly to the model. A-RAG provides three retrieval tools: keyword search, semantic search, and chunk read, enabling the agent to adaptively search and retrieve information across multiple granularities. Experiments on multiple open-domain QA benchmarks show that A-RAG consistently outperforms existing approaches with comparable or lower retrieved tokens, demonstrating that A-RAG effectively leverages model capabilities and dynamically adapts to different RAG tasks. We further systematically study how A-RAG scales with model size and test-time compute. We will release our code and evaluation suite to facilitate future research. Code and evaluation suite are available at https://github.com/Ayanami0730/arag.

cs.CL↗
arXiv2026

Joint single-shot ToA and DoA estimation for VAA-based BLE ranging with phase ambiguity: A deep learning-based approach

Conventional direction-of-arrival (DoA) estimation methods rely on multi-antenna arrays, which are costly to implement on size-constrained Bluetooth Low Energy (BLE) devices. Virtual antenna array (VAA) techniques enable DoA estimation with a single antenna, making angle estimation feasible on such devices. However, BLE only provides a single-shot two-way channel frequency response (CFR) with a binary phase ambiguity issue, which hinders the direct application of VAA. To address this challenge, we propose a unified model that combines VAA with BLE two-way CFR, and introduce a neural network based phase recovery framework that employs row / column predictors with a voting mechanism to resolve the ambiguity. The recovered one-way CFR then enables super resolution algorithms such as MUSIC for joint time of arrival (ToA) and DoA estimation. Simulation results demonstrate that the proposed method achieves superior performance under non-uniform VAAs, with mean square errors approaching the Cramer Rao bound at SNR $\geq$ 5 dB.

eess.SP↗
arXiv2025

Signatures of a Lifshitz transition in pressurized electron-doped cuprate

It is well known that the electronic structure of hole-doped cuprate superconductors is tunable through both chemical doping and external pressure, which frequently offer us new insights of understanding on the high-Tc superconducting mechanism. While, for electron-doped cuprate superconductors, although the chemical doping effects have been systematically and thoroughly investigated, there is still a notable lack of experimental evidence regarding the pressure-driven coevolution of Tc and electronic structure. In this study, we report the first observation on the signatures of pressure-induced Lifshitz transition in Pr0.87LaCe0.13CuO4+delta (PLCCO) single crystal, a typical electron-doped cuprate superconductor, through the comprehensive high-pressure measurements of electrical resistance, Hall coefficient (RH) and synchrotron X-ray diffraction (XRD). Our results reveal that, at 40 K, the ambient-pressure RH with a significantly negative value decreases with increasing pressure until it reaches zero at a critical pressure (Pc ~ 10 GPa). Meanwhile, the corresponding Tc exhibits a slight variation within this pressure range. As pressure is further increased beyond Pc, RH changes its sign from negative to positive and then shows a slight increase, while Tc displays a continuous decrease. Our XRD measurements at 40 K demonstrate that no crystal structure phase transition occurs across the Pc. These results reveal that applying pressure to PLCCO can induce a Lifshitz transition at Pc, manifesting the reconstruction of the Fermi surface (FS), which turns the superconductivity toward fading out. Our calculation further reinforces the Fermi surface reconstruction from electron-dominated to hole-dominated ones at around Pc. These findings provide new evidence that highlights the strong correlation between the superconductivity and the Fermi surface topology in the electron-doped cuprates.

cond-mat.supr-con↗
arXiv2025

NEFT: A Unified Transformer Framework for Efficient Near-Field CSI Feedback in XL-MIMO Systems

Extremely large-scale multiple-input multiple-output (XL-MIMO) systems, operating in the near-field region due to their massive antenna arrays, are key enablers of next-generation wireless communications but face significant challenges in channel state information (CSI) feedback. Deep learning has emerged as a powerful tool by learning compact channel features for feedback. However, existing methods struggle to capture the intricate structure of near-field CSI and incur prohibitive computational overhead on practical mobile devices. To overcome these limitations, we propose the near-field efficient feedback Transformer (NEFT) family for accurate near-field CSI feedback with reduced overhead under diverse hardware constraints. NEFT builds on a hierarchical vision Transformer backbone with progressive token reduction and multi-scale feature extraction, enabling compact and effective modeling of near-field channel characteristics. Furthermore, NEFT is extended with lightweight variants: NEFT-Compact applies multi-level knowledge distillation (KD) to reduce model complexity while preserving accuracy; NEFT-Hybrid adopts an attention-free CNN encoder to reduce encoder-side computation; and NEFT-Edge combines NEFT-Hybrid with KD to enable deployment on highly resource-constrained edge devices. Extensive simulations show that NEFT achieves a 15--21dB improvement in normalized mean-squared error over state-of-the-art methods, NEFT-Compact and NEFT-Edge reduce total FLOPs by 25-36% with negligible accuracy loss, while NEFT-Hybrid reduces encoder-side complexity by up to 64%, enabling deployment in highly asymmetric device scenarios. These results establish NEFT as a practical and scalable solution for near-field CSI feedback in XL-MIMO systems.

eess.SP↗