Searcharxiv⌕ Search

arXiv subjects

Wenjun Zhang

Publications and source records attributed to Wenjun Zhang.

At least 73 records · Page 4Linked to original sources
arXiv2026

WISV: Wireless-Informed Semantic Verification for Distributed Speculative Decoding in Device-Edge LLM Inference

While distributed device-edge speculative decoding enhances resource utilization across heterogeneous nodes, its performance is often bottlenecked by conventional token-level verification strategies. Such rigid alignment leads to excessive rejections, significantly diminishing the accepted sequence length and increasing interaction rounds under fluctuating wireless conditions. In this paper, we propose WISV (Wireless-Informed Semantic Verification), a novel distributed speculative decoding framework that goes beyond strict token-level matching via a channel-aware semantic acceptance policy. WISV integrates a lightweight decision head into the edge-side target LLM to dynamically evaluate speculative tokens by synthesizing high-dimensional hidden representations with instantaneous channel state information (CSI). To optimize the trade-off between verification fidelity and communication overhead, we further design two tailored communication protocols: full-hidden upload and mismatch-first selective-hidden upload. Extensive simulations using a 1B drafter and an 8B target model demonstrate that WISV achieves up to a 60.8% increase in accepted length, a 37.3% reduction in interaction rounds, and a 31.4% improvement in end-to-end latency compared to vanilla speculative decoding across tested settings, while maintaining a negligible task accuracy drop (<1%). Finally, we validate WISV on a hardware testbed comprising an NVIDIA Jetson AGX Orin and an A40-equipped server, confirming its real-world efficacy in accelerating edge-deployed LLM inference.

cs.IT↗
arXiv2026

Low-Complexity Soft-Feedback Detector for AFDM Systems

Affine frequency division multiplexing (AFDM), an emerging multi-carrier modulation scheme, has garnered significant attention due to its resilience to Doppler shifts and capability to achieve full diversity in doubly dispersive channels. However, existing data detection algorithms for AFDM systems face a significant trade-off between computational complexity and accuracy. In this paper, a novel low-complexity data detection scheme, termed the soft-feedback detector (SFD), is proposed. Particularly, building upon a maximum ratio combining (MRC) estimator framework, the SFD leverages the a priori symbol distribution to mitigate error propagation during iterative detection. Specifically, soft-decision feedback is incorporated as extrinsic information derived from the log-likelihood ratios of the transmitted symbols. As a result, the proposed detector significantly enhances detection accuracy while maintaining low computational complexity. Simulation results demonstrate that the SFD consistently outperforms benchmark decision-feedback detectors. In particular, compared with the conventional MRC detector, the proposed scheme achieves approximately a 3 dB signal-to-noise ratio (SNR) gain at the bit error rate (BER) of $10^{-3}$.

eess.SP↗
arXiv2026

AlphaEval: Evaluating Agents in Production

The rapid deployment of AI agents in commercial settings has outpaced the development of evaluation methodologies that reflect production realities. Existing benchmarks measure agent capabilities through retrospectively curated tasks with well-specified requirements and deterministic metrics -- conditions that diverge fundamentally from production environments where requirements contain implicit constraints, inputs are heterogeneous multi-modal documents with information fragmented across sources, tasks demand undeclared domain expertise, outputs are long-horizon professional deliverables, and success is judged by domain experts whose standards evolve over time. We present AlphaEval, a production-grounded benchmark of 94 tasks sourced from seven companies deploying AI agents in their core business, spanning six O*NET (Occupational Information Network) domains. Unlike model-centric benchmarks, AlphaEval evaluates complete agent products -- Claude Code, Codex, etc. -- as commercial systems, capturing performance variations invisible to model-level evaluation. Our evaluation framework covers multiple paradigms (LLM-as-a-Judge, reference-driven metrics, formal verification, rubric-based assessment, automated UI testing, etc.), with individual domains composing multiple paradigms. Beyond the benchmark itself, we contribute a requirement-to-benchmark construction framework -- a systematic methodology that transforms authentic production requirements into executable evaluation tasks in minimal time. This framework standardizes the entire pipeline from requirement to evaluation, providing a reproducible, modular process that any organization can adopt to construct production-grounded benchmarks for their own domains.

cs.CL↗
arXiv2026

Content-Aware Mamba for Learned Image Compression

Recent learned image compression (LIC) leverages Mamba-style state-space models (SSMs) for global receptive fields with linear complexity. However, the standard Mamba adopts content-agnostic, predefined raster (or multi-directional) scans under strict causality. This rigidity hinders its ability to effectively eliminate redundancy between tokens that are content-correlated but spatially distant. We introduce Content-Aware Mamba (CAM), an SSM that dynamically adapts its processing to the image content. Specifically, CAM overcomes prior limitations with two novel mechanisms. First, it replaces the rigid scan with a content-adaptive token permutation strategy to prioritize interactions between content-similar tokens regardless of their location. Second, it overcomes the sequential dependency by injecting sample-specific global priors into the state-space model, which effectively mitigates the strict causality without multi-directional scans. These innovations enable CAM to better capture global redundancy while preserving computational efficiency. Our Content-Aware Mamba-based LIC model (CMIC) achieves state-of-the-art rate-distortion performance, surpassing VTM-21.0 by 15.91%, 21.34%, and 17.58% in BD-rate on the Kodak, Tecnick, and CLIC datasets, respectively. Code will be released at https://github.com/UnoC-727/CMIC.

cs.CV↗
arXiv2026

CEI-3D: Collaborative Explicit-Implicit 3D Reconstruction for Realistic and Fine-Grained Object Editing

Existing 3D editing methods often produce unrealistic and unrefined results due to the deeply integrated nature of their reconstruction networks. To address the challenge, this paper introduces CEI-3D, an editing-oriented reconstruction pipeline designed to facilitate realistic and fine-grained editing. Specifically, we propose a collaborative explicit-implicit reconstruction approach, which represents the target object using an implicit SDF network and a differentially sampled, locally controllable set of handler points. The implicit network provides a smooth and continuous geometry prior, while the explicit handler points offer localized control, enabling mutual guidance between the global 3D structure and user-specified local editing regions. To independently control each attribute of the handler points, we design a physical properties disentangling module to decouple the color of the handler points into separate physical properties. We also propose a dual-diffuse-albedo network in this module to process the edited and non-edited regions through separate branches, thereby preventing undesired interference from editing operations. Building on the reconstructed collaborative explicit-implicit representation with disentangled properties, we introduce a spatial-aware editing module that enables part-wise adjustment of relevant handler points. This module employs a cross-view propagation-based 3D segmentation strategy, which helps users to edit the specified physical attributes of a target part efficiently. Extensive experiments on both real and synthetic datasets demonstrate that our approach achieves more realistic and fine-grained editing results than the state-of-the-art (SOTA) methods while requiring less editing time. Our code is available on https://github.com/shiyue001/CEI-3D.

cs.CV↗
arXiv2026

NEP-CG and NEP-AACG: Efficient coarse-grained and multiscale all-atom-coarse-grained neuroevolution potentials

Machine-learned coarse-grained (CG) models often suffer from noisy training data, limiting their accuracy and transferability. We propose a method to generate low-noise training data based on the potential of mean force by constraining CG beads during atomistic simulations and accumulating time-averaged forces. Implemented within the neuroevolution potential (NEP) framework, our approach achieves training accuracy comparable to atomistic models trained on density functional theory data. For liquid water, the NEP-CG model accurately reproduces densities from 1 bar to 1 GPa, successfully extrapolating beyond the 0.5 GPa training limit, with a virial correction essential for the correct equation of state. For an anisotropic C$_{60}$ monolayer, distinguishing crystallographically distinct bead types reduces stress errors by an order of magnitude and captures directional thermal conductivity. We further introduce a multiscale NEP-AACG model integrating all-atom (AA) and CG degrees of freedom, demonstrated for gold nanowire fracture at an experimentally relevant strain rate. Computational speeds for NEP-CG models reach hundreds to thousands of ns/day using a single consumer-grade GPU. This work provides a robust framework for constructing accurate, transferable, and efficient CG models across diverse systems.

physics.comp-ph↗
arXiv2026

Free-GVC: Towards Training-Free Extreme Generative Video Compression with Temporal Coherence

Building on recent advances in video generation, generative video compression has emerged as a new paradigm for achieving visually pleasing reconstructions. However, existing methods exhibit limited exploitation of temporal correlations, causing noticeable flicker and degraded temporal coherence at ultra-low bitrates. In this paper, we propose Free-GVC, a training-free generative video compression framework that reformulates video coding as latent trajectory compression guided by a video diffusion prior. Our method operates at the group-of-pictures (GOP) level, encoding video segments into a compact latent space and progressively compressing them along the diffusion trajectory. To ensure perceptually consistent reconstruction across GOPs, we introduce an Adaptive Quality Control module that dynamically constructs an online rate-perception surrogate model to predict the optimal diffusion step for each GOP. In addition, an Inter-GOP Alignment module establishes frame overlap and performs latent fusion between adjacent groups, thereby mitigating flicker and enhancing temporal coherence. Experiments show that Free-GVC achieves an average of 93.29% BD-Rate reduction in DISTS over the latest neural codec DCVC-RT, and a user study further confirms its superior perceptual quality and temporal coherence at ultra-low bitrates.

cs.CV↗
arXiv2026

Joint Lossy Compression for a Vector Gaussian Source under Individual Distortion Criteria

This paper investigates the joint compression problem of a vector Gaussian source, where an individual distortion constraint is imposed on each source component. It is known that the rate-distortion function (RDF) is lower-bounded by the rate derived from the Hadamard inequality, which becomes exact when the semidefinite condition (SDC) holds. However, existing works often overlook the case where the SDC is not satisfied. Moreover, even when the SDC holds, a quantitative characterization of how correlations enable more efficient compression is lacking. In this work, we refine the results when the SDC is satisfied and derive new theoretical results when the SDC is not satisfied, thereby establishing theoretical limits for practical source compression with correlations. Specifically, we examine the properties of optimal source reconstruction and provide upper bounds on its dimension, showing that lower-dimensional reconstructions are essential for efficient compression when the SDC does not hold. Within a scalable two-type correlation (2TC) covariance framework, we prove that the probability of satisfying the SDC decays exponentially with source length, emphasizing the importance of exploring theoretical limits when the SDC is not met. Additional, we determine the component-wise correlations that a vector source should possess to achieve the Hadamard compression rate, revealing the trade-off between distortion constraints and correlations. More importantly, by deriving an explicit RDF with correlations incorporated, we quantitatively characterize the gain in compression efficiency achieved by fully leveraging source correlations.

cs.IT↗
arXiv2026

FDD CSI Feedback under Finite Downlink Training: A Rate-Distortion Perspective

This paper establishes the theoretical limits of channel state information (CSI) feedback in frequency-division duplexing (FDD) multi-antenna orthogonal frequency-division multiplexing (OFDM) systems under finite-length training with Gaussian pilots. The user employs minimum mean-squared error (MMSE) channel estimation followed by asymptotically optimal uplink feedback. Specifically, we derive a general rate-distortion function (RDF) of the overall CSI feedback system. We then provide both non-asymptotic bounds and asymptotic scaling for the RDF under arbitrary downlink signal-to-noise ratio (SNR) when the number of training symbols exceeds the antenna dimension. A key observation is that, with sufficient training, the overall RDF converges to the direct RDF corresponding to the case where the user has full access to the downlink CSI. More importantly, we demonstrate that even at a fixed downlink SNR, the convergence rate is inversely proportional to the training length. The simulation results show that our bounds are tight, and under very limited training, the deviation between the overall RDF and the direct RDF is substantial.

cs.IT↗
arXiv2026

Dual-Representation Image Compression at Ultra-Low Bitrates via Explicit Semantics and Implicit Textures

While recent neural codecs achieve strong performance at low bitrates when optimized for perceptual quality, their effectiveness deteriorates significantly under ultra-low bitrate conditions. To mitigate this, generative compression methods leveraging semantic priors from pretrained models have emerged as a promising paradigm. However, existing approaches are fundamentally constrained by a tradeoff between semantic faithfulness and perceptual realism. Methods based on explicit representations preserve content structure but often lack fine-grained textures, whereas implicit methods can synthesize visually plausible details at the cost of semantic drift. In this work, we propose a unified framework that bridges this gap by coherently integrating explicit and implicit representations in a training-free manner. Specifically, We condition a diffusion model on explicit high-level semantics while employing reverse-channel coding to implicitly convey fine-grained details. Moreover, we introduce a plug-in encoder that enables flexible control of the distortion-perception tradeoff by modulating the implicit information. Extensive experiments demonstrate that the proposed framework achieves state-of-the-art rate-perception performance, outperforming existing methods and surpassing DiffC by 29.92%, 19.33%, and 20.89% in DISTS BD-Rate on the Kodak, DIV2K, and CLIC2020 datasets, respectively.

cs.CV↗
arXiv2026

SurfSplat: Conquering Feedforward 2D Gaussian Splatting with Surface Continuity Priors

Reconstructing 3D scenes from sparse images remains a challenging task due to the difficulty of recovering accurate geometry and texture without optimization. Recent approaches leverage generalizable models to generate 3D scenes using 3D Gaussian Splatting (3DGS) primitive. However, they often fail to produce continuous surfaces and instead yield discrete, color-biased point clouds that appear plausible at normal resolution but reveal severe artifacts under close-up views. To address this issue, we present SurfSplat, a feedforward framework based on 2D Gaussian Splatting (2DGS) primitive, which provides stronger anisotropy and higher geometric precision. By incorporating a surface continuity prior and a forced alpha blending strategy, SurfSplat reconstructs coherent geometry together with faithful textures. Furthermore, we introduce High-Resolution Rendering Consistency (HRRC), a new evaluation metric designed to evaluate high-resolution reconstruction quality. Extensive experiments on RealEstate10K, DL3DV, and ScanNet demonstrate that SurfSplat consistently outperforms prior methods on both standard metrics and HRRC, establishing a robust solution for high-fidelity 3D reconstruction from sparse inputs. Project page: https://hebing-sjtu.github.io/SurfSplat-website/

cs.CV↗
arXiv2026

ProxyImg: Towards Highly-Controllable Image Representation via Hierarchical Disentangled Proxy Embedding

Prevailing image representation methods, including explicit representations such as raster images and Gaussian primitives, as well as implicit representations such as latent images, either suffer from representation redundancy that leads to heavy manual editing effort, or lack a direct mapping from latent variables to semantic instances or parts, making fine-grained manipulation difficult. These limitations hinder efficient and controllable image and video editing. To address these issues, we propose a hierarchical proxy-based parametric image representation that disentangles semantic, geometric, and textural attributes into independent and manipulable parameter spaces. Based on a semantic-aware decomposition of the input image, our representation constructs hierarchical proxy geometries through adaptive Bezier fitting and iterative internal region subdivision and meshing. Multi-scale implicit texture parameters are embedded into the resulting geometry-aware distributed proxy nodes, enabling continuous high-fidelity reconstruction in the pixel domain and instance- or part-independent semantic editing. In addition, we introduce a locality-adaptive feature indexing mechanism to ensure spatial texture coherence, which further supports high-quality background completion without relying on generative models. Extensive experiments on image reconstruction and editing benchmarks, including ImageNet, OIR-Bench, and HumanEdit, demonstrate that our method achieves state-of-the-art rendering fidelity with significantly fewer parameters, while enabling intuitive, interactive, and physically plausible manipulation. Moreover, by integrating proxy nodes with Position-Based Dynamics, our framework supports real-time physics-driven animation using lightweight implicit rendering, achieving superior temporal consistency and visual realism compared with generative approaches.

cs.CV↗
arXiv2026

MACEval: A Multi-Agent Continual Evaluation Network for Large Models

Hundreds of benchmarks dedicated to evaluating large models have been presented over the past few years. However, most of them remain closed-ended and are prone to overfitting due to the potential data contamination. Moreover, the increasing scale and scope of current benchmarks with transient metrics, as well as the heavily human-dependent curation procedure, pose significant challenges for timely maintenance and adaptation. In this paper, we introduce MACEval, a Multi-Agent Continual Evaluation network for dynamic evaluation of large models, and define new metrics to quantify performance longitudinally. MACEval employs an interactive and autonomous evaluation mode, utilizing role assignment, in-process data generation, and evaluation routing through a cascaded agent network. Extensive experiments on 23 large models demonstrate the effectiveness of MACEval, which also lightens the evaluation process and reduces a considerable amount of overhead. We hope that MACEval can broaden future directions of large model evaluation. Project page: https://github.com/zijianchen98/MACEval.

cs.CV↗
arXiv2026

Evaluating from Benign to Dynamic Adversarial: A Squid Game for Large Language Models

The potential data contamination issue in contemporary large language models (LLMs) benchmarks presents a fundamental challenge to establishing trustworthy evaluation frameworks. Meanwhile, they predominantly assume benign, resource-rich settings, leaving the behavior of LLMs under pressure unexplored. In this paper, we introduce \textsc{Squid Game}, a dynamic and adversarial evaluation environment with resource-constrained and asymmetric information settings elaborated to evaluate LLMs through interactive gameplay against other LLM opponents. Squid Game consists of six elimination-style levels, focusing on multi-faceted abilities, including instruction-following, code, reasoning, planning, and safety alignment. We evaluate over 50 LLMs on Squid Game, presenting the largest behavioral evaluation study of general LLMs on dynamic adversarial scenarios. We observe a clear generational phase transition in performance in the same model lineage and find evidence that some models resort to speculative shortcuts to win the game, indicating the possibility of higher-level evaluation paradigm contamination in static benchmarks. We also compare prominent LLM benchmarks and \textsc{Squid Game}, highlighting that dynamic evaluation can serve as a complementary part for static evaluations. Project page: https://github.com/zijianchen98/LLM_Squid_Game.

cs.CL↗
arXiv2026

Multi-hop Parallel Image Semantic Communication for Distortion Accumulation Mitigation

Existing semantic communication schemes primarily focus on single-hop scenarios, overlooking the challenges of multi-hop wireless image transmission. As semantic communication is inherently lossy, distortion accumulates over multiple hops, leading to significant performance degradation. To address this, we propose the multi-hop parallel image semantic communication (MHPSC) framework, which introduces a parallel residual compensation link at each hop against distortion accumulation. To minimize the associated transmission bandwidth overhead, a coarse-to-fine residual compression scheme is designed. A deep learning-based residual compressor first condenses the residuals, followed by the adaptive arithmetic coding (AAC) for further compression. A residual distribution estimation module predicts the prior distribution for the AAC to achieve fine compression performances. This approach ensures robust multi-hop image transmission with only a minor increase in transmission bandwidth. Experimental results confirm that MHPSC outperforms both existing semantic communication and traditional separated coding schemes.

cs.IT↗
arXiv2026

WVSC: Wireless Video Semantic Communication with Multi-frame Compensation

Existing wireless video transmission schemes directly conduct video coding in pixel level, while neglecting the inner semantics contained in videos. In this paper, we propose a wireless video semantic communication framework, abbreviated as WVSC, which integrates the idea of semantic communication into wireless video transmission scenarios. WVSC first encodes original video frames as semantic frames and then conducts video coding based on such compact representations, enabling the video coding in semantic level rather than pixel level. Moreover, to further reduce the communication overhead, a reference semantic frame is introduced to substitute motion vectors of each frame in common video coding methods. At the receiver, multi-frame compensation (MFC) is proposed to produce compensated current semantic frame with a multi-frame fusion attention module. With both the reference frame transmission and MFC, the bandwidth efficiency improves with satisfying video transmission performance. Experimental results verify the performance gain of WVSC over other DL-based methods e.g. DVSC about 1 dB and traditional schemes about 2 dB in terms of PSNR.

cs.MM↗
arXiv2026

Joint Source-Channel-Generation Coding: From Distortion-oriented Reconstruction to Semantic-consistent Generation

Conventional communication systems, including both separation-based coding and AI-driven joint source-channel coding (JSCC), are largely guided by Shannon's rate-distortion theory. However, relying on generic distortion metrics fails to capture complex human visual perception, often resulting in blurred or unrealistic reconstructions. In this paper, we propose Joint Source-Channel-Generation Coding (JSCGC), a novel paradigm that shifts the focus from deterministic reconstruction to probabilistic generation. JSCGC leverages a generative model at the receiver as a generator rather than a conventional decoder to parameterize the data distribution, enabling direct maximization of mutual information under channel constraints while controlling stochastic sampling to produce outputs residing on the authentic data manifold with high fidelity. We further derive a theoretical lower bound on the maximum semantic inconsistency with given transmitted mutual information, elucidating the fundamental limits of communication in controlling the generative process. Extensive experiments on image transmission demonstrate that JSCGC substantially improves perceptual quality and semantic fidelity, significantly outperforming conventional distortion-oriented JSCC methods.

cs.IT↗
arXiv2026

When Wires Can't Keep Up: Reconfigurable AI Data Centers Empowered by Terahertz Wireless Communications

The explosive growth of artificial intelligence (AI) workloads in modern data centers demands a radical transformation of interconnect architectures. Traditional copper and optical wiring face fundamental challenges in latency, power consumption, and rigidity, constraining the scalability of distributed AI clusters. This article introduces a vision for Terahertz (THz) Wireless Data Center (THz-WDC) that combines ultra-broadband capacity, one-hop low-latency communication, and energy efficiency in the short-to-medium range (1-100m). Performance and technical requirements are first articulated, including up to 1 Tbps per link, aggregate throughput up to 10 Tbps via spatial multiplexing, sub-50 ns single-hop latency, and sub-10 pJ/bit energy efficiency over 20m. To achieve these ambitious goals, key enabling technologies are explored, including digital-twin-based orchestration, low-complexity beam manipulation technologies, all-silicon THz transceivers, and low-complexity analog baseband architectures. Moreover, as future data centers shift toward quantum and chiplet-based modular architectures, THz wireless links provide a flexible mechanism for interconnecting, testing, and reconfiguring these modules. Finally, numerical analysis is presented on the latency and power regimes of THz versus optical and copper interconnects, identifying the specific distance and throughput domains where THz links can surpass conventional wired solutions. The article concludes with a roadmap toward wireless-defined, reconfigurable, and sustainable AI data centers.

cs.IT↗