Searcharxiv⌕ Search

arXiv subjects

Haijun Wu

Publications and source records attributed to Haijun Wu.

At least 19 recordsLinked to original sources

CARE: Contrastive-Anchored Residual Encoding with Listwise Reconstruction for Retrieval-Oriented Semantic IDs

Generative retrieval replaces nearest-neighbor search with the generation of a document's semantic ID (SID), so the SID tokenizer bounds what the retriever can recall. Common SID constructions (residual k-means, RQ-VAE, finite scalar quantization, capacity balancing) optimize criteria defined on document embeddings alone and do not control whether a query can still find its relevant documents after quantization. We propose two retrieval-oriented SID tokenizers. CARE-SID (Contrastive-Anchored Residual Encoding for Semantic IDs) quantizes in a retrieval-aligned space: a zero-initialized residual alignment network is trained with a query-document contrastive loss before quantization, an anchor loss and a reconstruction target fixed to the original embedding prevent drift and collapse, and reference-vector rating residual quantization assigns the codes. CARE-LR (CARE with Listwise Reconstruction) then distills, for each training query, the similarity distribution that the original embeddings induce over a candidate list into the SID reconstructions, through a function-preserving layer-resolved head. Both keep a document-only encoding interface. We judge SIDs without training a generator, by row-budget recall in a bucketing view (the partition) and a reconstruction view (the SID with its reconstruction). In offline evaluation on industrial video search data, both tokenizers achieve higher reconstruction-view recall than residual k-means, and CARE-LR improves further on CARE-SID, mainly through the listwise distillation. A comparison of two document embeddings shows that contrastive alignment helps most when the embedding has not been trained on search data, and that the embedding affects SID recall far more than the quantizer. CARE-SID is deployed in the generative retrieval stage of an industrial video search system.

cs.IR↗

A pure source transfer domain decomposition method for Helmholtz equations in 3D unbounded domain

We develop and analyze a three-dimensional pure source transfer domain decomposition method (PSTDDM) for the Helmholtz equation in free space. Extending the two-dimensional PSTDDM proposed by [Du and Wu, J. Sci. Comput., 83(2020),83:68], the proposed method transfers sources successively along the $x_1$, $x_2$, and $x_3$ directions, yielding layer-wise, bar-wise, and block-wise PSTDDMs. The main challenge in the convergence analysis is deriving wavenumber-explicit truncation-error estimates. To address this challenge, we establish inf--sup conditions for the uniaxial perfectly matched layer (UPML) problem under relaxed restrictions on the PML parameters and reformulate the source-transfer operators. We then introduce a positive-definite reconstruction step that produces a globally conforming approximation. In addition, a new quantitative lower bound is established for the imaginary part of the complex PML distance, more easily yielding exponential decay estimates in three dimensions. These lead to wavenumber-explicit convergence estimates for all three PSTDDMs. Numerical experiments demonstrate the effectiveness of the proposed methods both as direct solvers and as preconditioners for GMRES.

math.NA↗

WeFair: Fidelity-Aware Alignment from Human Feedback for Personalized Query Rewriting in Short-Video Search

Short-video search needs compact query rewriters that recover users' personalized intent from ambiguous queries under strict latency constraints. Existing LLM-based methods have improved rewriting by using synthetic reasoning and subsequent consumption-behavioral feedback. However, overemphasizing subsequent consumption can favor engaging content over the user's immediate need, causing intent drift even when rewritten queries are grounded in behavioral history. We propose WeFair, Fidelity-Aware Alignment from Human Feedback for Personalized Query Rewriting in WeChat short-video search. WeFair uses human feedback according to how directly it reflects personalized intent, assigning feedback fidelity to a progression from demonstration to preference to verification. We first mine actual queries and user reformulations from search sessions, and then screen them for contextual support. In the demonstration stage, a supervised fine-tuning curriculum progresses from large-scale context-to-query learning to reasoning and direct prediction on human reformulation and preservation. In the preference stage, direct preference optimization learns relative preferences from subsequent consumption feedback. In the verification stage, group relative policy optimization rewards exact agreement with reformulation or preservation targets to constrain intent drift. Further analysis shows that broad supervision benefits the full curriculum supervised training, while strict verification achieves better performance than auxiliary rewards. In offline experiments, WeFair-2B outperforms strong rewriting baselines in overall accuracy and intent preservation. An online A/B test shows relative increases of 2.593% in effective video-view rate and a 2.350% relative reduction in query reformulation rate. WeFair-2B has been deployed in the WeChat short-video search system, which serves over 100 million queries per day.

cs.AI↗

WeGR: Towards Unified Generative Retrieval via Behavior-aware Tokenization and Preference Alignment

Generative retrieval (GR) offers a promising paradigm for industrial search by unifying candidate generation within an autoregressive framework. However, effectively leveraging user behaviors in industrial GR remains challenging: existing SID tokenization methods struggle to preserve collaborative signals while maintaining balanced codebook utilization, whereas reinforcement learning (RL) for prioritizing high-value candidates suffers from sparse rewards when generated SIDs fail to match positive items. To address these challenges, we propose WeGR, a unified GR framework with two complementary components. Specifically, the Collaborative and Collapse-free Tokenizer (CoCo-VAE) injects query-to-item and item-to-item collaborative signals into continuous representations before quantization, preserving behavioral information without disrupting codebook optimization. It further couples first-layer codebook refresh to reactivate underutilized codewords with fidelity anti-collapse regularization, which broadens gradient coverage over relevant codewords and penalizes collisions in deeper layers to yield behavior-aware SIDs with balanced utilization. For preference alignment, Dynamic Horizon Policy Optimization (DHPO) progressively expands reward evaluation and policy optimization horizons to alleviate reward sparsity, enabling coarse-to-fine preference learning from SID prefixes to complete items without auxiliary reward models. Extensive offline experiments validate that WeGR improves retrieval and ranking performance. Moreover, a six-week online A/B test on a large-scale e-commerce search platform achieves relative gains of 2.539% in orders, 1.508% in CTR, and 1.381% in CVR, while reducing CPU and GPU usage by 25.5% and 26.1%. These results highlight the effectiveness and efficiency of WeGR in industrial deployment, supporting multiple search scenarios and tens of millions of daily page views.

cs.IR↗

Spherical Perfectly Matched Layer Method for Electromagnetic Scattering Problems with Large Wave Number

We study a spherical perfectly matched layer (PML) method for the time-harmonic Maxwell scattering problem with large wavenumber. Using vector spherical harmonics and a modal transformation for the coupled transverse magnetic components, we establish the sharp wavenumber-explicit stability estimate for the truncated PML problem and prove that its inf-sup constant is of order $O(κ^{-1})$. We then analyze the first-order Nédélec edge element method and its continuous interior penalty counterpart. Under the condition that $κ^3h^2$ is sufficiently small, the preasymptotic errors in the energy norm and the $κ$-scaled $\mathbf{L}^2$ norm are of orders $O(κh+κ^3h^2)$ and $O(κ^2h^2+κ^3h^2)$, respectively. Numerical experiments confirm the theoretical results.

math.NA↗

VR-JEPA: Learning Contrastive-State Latent Guidance for Generation-based Video Reasoning

Reasoning through video generation offers a promising path toward visual intelligence by modeling latent visual states and their dynamics. However, current video generation models often lack explicit guidance on how these states should evolve, leaving generated trajectories prone to physical and structural inconsistencies that undermine reasoning reliability. While the Video Joint-Embedding Predictive Architecture (V-JEPA) provides rich spatiotemporal priors learned through latent prediction, these general priors do not naturally adapt to the logical reasoning capabilities required for complex visual tasks. To bridge this gap, we propose VR-JEPA, a framework that aligns the V-JEPA predictor with task-specific reasoning logic through localized contrastive-state learning and uses its predicted latent trajectories to guide video generation for visual reasoning. Specifically, (i) we pair successful trajectories with generated alternatives under the same input conditions and use discrepancies in their V-JEPA representations to identify informative states and tokens for localized contrastive supervision. (ii) We further equip the V-JEPA predictor with skill-specific experts trained on anchor-task data, allowing the model to adaptively specialize its shared spatiotemporal priors across diverse cognitive domains. Together with skill-specific experts, this contrastive supervision enables VR-JEPA to predict latent trajectories that provide task-specific logical guidance for video generation. Comprehensive experiments on the large-scale VBVR-Pro-Bench dataset demonstrate that VR-JEPA achieves an $11.33\%$ relative improvement over the cutting-edge generation-based reasoning baseline, significantly mitigating physical artifacts and enhancing logical consistency.

cs.CV↗

Beyond Solo and Consistency: Vindicating Multi-Agent Debate via Conditional Progressive Pruning

Large Language Model (LLM) based Multi-Agent Debate (MAD) is one of the most effective test time scaling techniques. Through multi-round communication, agents complement each other in knowledge and reasoning and solve tasks that no single member can solve. However, existing MAD frameworks fail to beat strong Single Agent and Consistency-based baselines under the same strict cost limit, which shakes the foundation of the MAD field. We propose Conditional Progressive Pruning (CPP), a lightweight pruning framework that fully exploits multi-round MAD. CPP outperforms all existing MAD frameworks on multiple dominated benchmarks. It is also the first to fully outperform consistency methods. Our code, detailed agent interaction records will be released soon.

cs.CL↗

Efficient Multimodal Generative Recommendation with Latent Narrative Reasoning

Generative recommendation reformulates item prediction as semantic identifier generation, yet episodic content introduces a fundamentally different setting where the target is determined by narrative evolution rather than user preference. This task requires models to understand multimodal storyline progression while addressing the efficiency challenges caused by redundant visual contexts and costly explicit reasoning generation. We propose \textbf{NarraLite}, an efficient multimodal generative recommendation framework that jointly compresses perception and reasoning. Specifically, Progressive Spectral Compression selectively distills long visual contexts into compact narrative-relevant evidence, preserving transition-critical information while reducing redundant visual computation. Latent Narrative Reasoning introduces context-routed latent reasoning tokens and aligns their contextualized representations with future continuation semantics, enabling implicit narrative inference without autoregressively decoding textual rationales. We further establish a user-agnostic multimodal benchmark for short-form drama continuation across UGC, PGC, and OOD settings. Extensive experiments demonstrate that NarraLite consistently improves continuation accuracy, narrative coherence, and robustness over existing approaches, while achieving a favorable accuracy--efficiency trade-off.

cs.CL↗

Mr.LHDR: A Benchmark for Multimodal Real-World Long-Horizon Deep Research Agents

Deep research agents are increasingly capable of web search, tool use, multimodal evidence analysis, and information synthesis. However, existing benchmarks mainly evaluate medium-horizon exploration and rarely test whether agents can sustain long, dependency-heavy research processes. We introduce Mr. LHDR (Multimodal real-world Long-Horizon Deep Research), a benchmark for evaluating real-world deep research over long, irreducible chains of interdependent evidence across eight categories. Each question is constructed from a hidden Node-Relation graph and requires an average of 12.1 necessary intermediate conclusions with a mean dependency depth of 10.4 before reaching a short, unique, and verifiable answer. Questions incorporate multimodal evidence, including images, maps, PDFs, logos, charts, tables, and video frames, with at least one non-text element that changes the reasoning state. Mr. LHDR evaluates both final answers and the correctness of intermediate conclusions under annotated dependencies. We evaluate general models, deep research systems, and agent frameworks using Overall Accuracy (OA), Strict Accuracy (SA), Checklist Score (CS), and Dependency-Aware Checklist Score (DACS). Results show that even the strongest system achieves only 43.1% OA and 34.3% SA, indicating that final-answer accuracy substantially overestimates complete research success. Removing images reduces DACS by 12.6 points, demonstrating the importance of multimodal evidence, while SA consistently declines as reasoning chains become longer. These findings reveal sustained, dependency-consistent evidence integration, rather than isolated fact retrieval, as a key bottleneck for current deep research agents.

cs.AI↗

Polynomial preserving recoveries of edge element method on Cartesian grids for the time-harmonic Maxwell equations with large wave number

This paper considers the lowest-order first type Nédélec edge element method (EEM) on Cartesian grids for the three-dimensional time-harmonic Maxwell equations with a large wave number. New polynomial preserving recovery (PPR) operators are proposed for the curl of the edge element solution and for the solution itself, respectively. Under the condition that $κ^3 h^2 C_{\mathrm{sol}}$ is sufficiently small, second-order superconvergence estimates are proved for both the recovered curl and the recovered solution, where $κ$ is the wave number, $h$ is the mesh size, and $C_{\mathrm{sol}}$ is a stability constant associated with the Maxwell solution operator. In particular, the analysis shows that the proposed PPR procedures cannot mitigate the well-known pollution effect inherent to the EEM. To reduce the pollution error, we further propose a new continuous interior penalty edge element method (CIP-EEM) that incorporates an additional normal-jump penalty term. It is shown that by appropriately choosing the penalty parameters, the new CIP-EEM can improve the phase error by two orders in $κh$. Numerical experiments are presented to confirm the theoretical superconvergence results and to demonstrate that the CIP-EEM can effectively reduce the pollution error in the high-frequency regime.

math.NA↗

The effect of numerical integration in the FEM for elliptic problems with mixed boundary conditions

This paper investigates the impact of quadrature accuracy for volume and face integrals in the finite element method using $p$-th order polynomial shape functions for elliptic problems with mixed Dirichlet and Robin boundary conditions. The optimal $p$-th order $H^1$-convergence is maintained when numerical integration with algebraic precision at least $2p-2$ for volume terms and at least $2p-1$ for face terms is adopted. For $L^2$-error, we achieve optimal $O(h^{p+1})$ convergence when using quadrature rules of precision no less than $\max\{p,2p-2\}$ for volume terms and no less than $2p-1$ for face terms. Of particular significance, we present two examples to show that the above result on $L^2$-error is sharp for the linear FEM ($p=1$). When reduced to the case of Dirichlet boundary condition, our results yield improved dependence on the given data compared to the classical results established by Ciarlet, \textit{et al}. Numerical experiments are provided to illustrate the theoretical findings and confirm the necessity of specified quadrature accuracy and data regularity.

math.NA↗

HeadRank: Decoding-Free Passage Reranking via Preference-Aligned Attention Heads

Decoding-free reranking methods that read relevance signals directly from LLM attention weights offer significant latency advantages over autoregressive approaches, yet suffer from attention score homogenization: middle-context documents receive near-identical scores, destroying the fine-grained distinctions required for ranking. We propose HeadRank, a framework that lifts preference optimization from discrete token space into the continuous attention domain through entropy-regularized head selection, hard adjacent-level preference pairs, and a distribution regularizer that jointly sharpen discriminability in the homogenized middle zone. Depth truncation at the deepest selected layer further reduces inference to $\mathcal{O}(1)$ forward passes. Across 14 benchmarks on three Qwen3 scales (0.6B--4B) using only 211 training queries, HeadRank achieves the highest average NDCG@10 at every scale, outperforming both generative and decoding-free baselines on the majority of benchmarks with 100\% formatting success. At 4B, 57.4\% of relevant middle-zone documents reach the top quartile versus 14.2\% for irrelevant ones -- a 43-percentage-point selectivity gap that demonstrates the effectiveness of attention-space preference alignment for listwise reranking.

cs.IR↗

When & How to Write for Personalized Demand-aware Query Rewriting in Video Search

In video search systems, user historical behaviors provide rich context for identifying search intent and resolving ambiguity. However, traditional methods utilizing implicit history features often suffer from signal dilution and delayed feedback. To address these challenges, we propose WeWrite, a novel Personalized Demand-aware Query Rewriting framework. Specifically, WeWrite tackles three key challenges: (1) When to Write: An automated posterior-based mining strategy extracts high-quality samples from user logs, identifying scenarios where personalization is strictly necessary; (2) How to Write: A hybrid training paradigm combines Supervised Fine-Tuning (SFT) with Group Relative Policy Optimization (GRPO) to align the LLM's output style with the retrieval system; (3) Deployment: A parallel "Fake Recall" architecture ensures low latency. Online A/B testing on a large-scale video platform demonstrates that WeWrite improves the Click-Through Video Volume (VV$>$10s) by 1.07% and reduces the Query Reformulation Rate by 2.97%.

cs.IR↗

Super-resolution optical trapping of multiple cold atoms

Arrays of optical tweezers form the backbone of neutral atoms analog and digital quantum processors. However, the inter-trap distance remains generally much larger than the size of the tweezers to avoid interference-induced trap distortions, limiting the trap density. Here, we report single-atom trapping in four super-resolved tweezers, meaning with a separation below the Sparrow diffraction limit. The optical pattern is generated using superoscillatory phenomenon leading to subwavelength traps with full control of the trap relative phases. We investigate two sets of relative phases that impede or allow the hopping and the reshuffling of atoms. We envision that superoscillatory light structuring will bridge the gap between large-distance traps generated by tweezer arrays and short-distance traps formed with optical lattices.

physics.atom-ph↗

Preasymptotic error estimates of higher-order EEM for the time-harmonic Maxwell equations with large wave number

The time-harmonic Maxwell equations with impedance boundary condition and large wave number are discretized using the second-type Nédélec's edge element method (EEM). Preasymptotic error bounds are derived, showing that, under the mesh condition $κ^{2p+1}h^{2p}$ being sufficiently small, the error of the EEM of order $p$ in the energy norm is bounded by $\mathcal{O}\big(κ^{p}h^p + κ^{2p+1}h^{2p}\big)$, while the error in the $κ$-scaled $\boldsymbol{L}^2$ norm is bounded by $\mathcal{O}\big((κh)^{p+1} + κ^{2p+1} h^{2p}\big)$. Here, $κ$ is the wave number and $h$ is the mesh size. Numerical tests are provided to illustrate our theoretical results.

math.NA↗

An Unfitted Interface Penalty DG--FE Method for Elliptic Interface Problems

We propose an unfitted interface penalty Discontinuous Galerkin-Finite Element Method (UIPDG-FEM) for elliptic interface problems. This hybrid method combines the interior penalty discontinuous Galerkin (IPDG) terms near the interface-enforcing jump conditions via Nitsche method-with standard finite elements away from the interface. The UIPDG-FEM retains the flexibilities of IPDG, particularly simplifying mesh generation around complex interfaces, while avoiding its drawback of excessive number of global degrees of freedom. We derive optimal convergence rates independent of interface location and establish uniform flux error estimates robust to discontinuous coefficients. To deal with conditioning issues caused by small cut elements, we develop a robust two-dimensional merging algorithm that eliminates such elements entirely, ensuring the condition number of the discretized system remains independent of interface position. A key feature of the algorithm is a novel quantification criterion linking the threshold for small cuts to the product of the maximum interface curvature and the local mesh size. Numerical experiments confirm the theoretical results and demonstrate the effectiveness of the proposed method.

math.NA↗

Tailoring ultra-high-order optical skyrmions

Skyrmions, as quasiparticles with topological spin textures, has recently garnered great attention for both condensed matter and structured wave communities, promising next-generation large-density robust information technologies. However, a big challenge to this end is that the generation of high-order skyrmions is elusive in any physical systems. Here, we propose the method to create and control ultra-high-order skyrmions (skyrmion number up to $400^{th}$) in a structured light system. We also experimentally control the topological state transition between bimeron and skyrmion, arbitrarily tailor the transverse size of an arbitrary-order skyrmionic beam independent of topological number, and ensure the topological stability upon propagation. Our work offers solutions for topologically resilient communication and memory with much enhanced information capacity.

physics.optics↗

Optical skyrmion lattices accelerating in free space

Generation and propagation of optical skyrmions provide a versatile plalform for topologically nontrivial optical informatics and light-matter interactions, but their acceleration along curved trajectories is to be studied. In this study, we experimentally demonstrate the first accelerating skyrmion lattices conveyed by Airy structured light, characterized by topologically stable skyrmion textures with self-acceleration along parabolic trajectories. We show that the skyrmion unit cell can maintain a Skyrme number $|N_\text{sk}|>0.9$ within a propagation range of $\pm1.22\ z_R$ upon parabolic acceleration. Notably, the meron structure remains $|N_\text{sk}|$ stable within $0.5\pm0.02$ over a significantly extended range of $\pm3.06\ z_R$. Our work provides a new potential carrier for topologically robust information distribution, particle sorting and manipulation.

physics.optics↗