arXiv subjects

Wen Wu

Publications and source records attributed to Wen Wu.

At least 37 records · Page 2Linked to original sources

arXiv2026

Movable Antenna Enhanced Wireless Sensing via Steering Vector Correlation and CRB Optimization

In this paper, we investigate the angle-of-arrival (AoA) estimation problem for wireless sensing systems equipped with movable antennas (MA). To achieve high estimation performance and accuracy, we formulate a joint optimization problem integrating the sidelobes of steering vector correlation (SVC) and the Cramér-Rao bound (CRB). We first mathematically transform the SVC and the CRB into tractable objective functions. Specifically, we introduce a proxy variable and apply a discrete grid search strategy to overcome the intractability of optimizing the SVC with unknown target angles. Concurrently, we derive a generalized lower bound for the CRB, which yields a scalar function of the MA positions. Guided by the transformed objective, we propose a successive convex approximation-based position optimization algorithm. The proposed algorithm handles the non-convex terms by employing first order Taylor expansions within a defined trust region, which allows the MA positions to be updated incrementally in each iteration. Simulation results demonstrate that the proposed algorithm achieves superior AoA estimation performance.

eess.SP↗

arXiv2026

Rotatable Antenna-Enhanced Wireless Sensing with Uniform Sparse Array via Tensor Decomposition

In this letter, we propose a new wireless sensing system equipped with a rotatable antenna (RA) array to enhance the sensing performance of a uniform sparse array (USA). To tackle the severe spatial undersampling issues, we propose a novel tensor decomposition-based direction-of-arrival (DOA) estimation algorithm. Specifically, we introduce a synchronous multiple rotation pattern for active target probing such that the received signals across multiple rotations to capture the diverse spatial degree of freedoms. Subsequently, we mathematically formulate the received signals across successive rotations as a third-order tensor, and leverage the canonical polyadic decomposition to obtain the factor matrices incorporating the DOA of targets. By analyzing the extrema distribution laws of array steering vector correlation (SVC) and gain SVC of RAs, we propose to combine the array and gain factor matrices via the Kronecker product, which theoretically guarantees the unambiguous DOA estimation. Simulation results demonstrate that the proposed RA-enhanced tensor decomposition-based algorithm achieves high-precision and unambiguous sensing performance compared to conventional uniform dense arrays and omnidirectional antenna systems.

eess.SP↗

arXiv2026

MMAG: A Multi-Control Mixed Audio Generation Benchmark

Recent audio generation systems have progressed from single-modality synthesis to generating complex acoustic scenes containing speech, music, and sound effects. Therefore, evaluating these models requires assessing multiple interacting capabilities, including semantic fidelity, speaker consistency, and temporal control, yet existing benchmarks focus on isolated domains or coarse-grained descriptions. To address this gap, we introduce the Multi-control Mixed Audio Generation (MMAG) benchmark. MMAG contains approximately 4,000 manually verified audio clips with rich annotations covering speech content, speaker identity, music attributes, sound events, and temporal relationships, together with dedicated subsets for voice cloning and timestamp-conditioned generation. We further propose a systematic evaluation protocol that measures acoustic fidelity, speech quality, semantic alignment, and temporal accuracy. Benchmarking representative agentic orchestrators, unified audio-visual generation models, and native mixed-audio generators reveals substantial performance trade-offs across these capabilities, with no existing model performing consistently well. Our results highlight the remaining challenges of controllable mixed audio generation and establish MMAG as a comprehensive benchmark for future research.

cs.SD↗

arXiv2026

Leveraging unstructured grids for direct numerical simulations of wall turbulence

Towards computational cost saving for direct numerical simulations (DNSs) of wall turbulence, we formulate an unstructured grid-generation framework, termed $η$-grid, where the wall-normal ($y$) and spanwise ($z$) grid sizes are proportional to the local Kolmogorov scale $η$. The framework consists of an inner layer, with a thickness $\sim 50$ viscous units, with viscous-scaled grid sizes similar to a conventional DNS grid: $0.3 \lesssim Δy^+ \lesssim 4, Δz^+ \simeq 5$ over a smooth wall, and $\ell^+/30 \lesssim Δy^+, Δz^+ \lesssim 4$ over uneven surfaces, where $\ell^+$ is the smallest surface wavelength. Above the inner layer, $Δy^+ \simeq Δz^+ \simeq 2η^+$. We test $η$-grid with finite volume and spectral element solvers, and conduct DNSs of turbulent channel flows and boundary layers over smooth wall and various streamwise-aligned riblets, up to friction Reynolds number $δ^+_0 = 1000$. We assess the accuracy of $η$-grid against the conventional Cartesian grids, through comparison with the reference DNS and experimental data. Results from $η$-grid and the Cartesian grids differ by less than $1\%$, in terms of turbulence statistics up to second-order, and the energy spectra. For turbulent channel flows with $10^3 \lesssim δ^+_0 \lesssim 10^4$, the number of grid points with $η$-grid ($N_η$) scales $\propto {δ^+_0}^{2.47}$ over a smooth wall, and $\propto {δ^+_0}^{2.0-2.47}$ over riblets, whereas the number of grid points with a Cartesian grid and hyperbolic-tangent $y$-grid ($N_\mathrm{Tanh}$) scales $\propto {δ^+_0}^{3.0}$. By $δ^+_0 = 6000$, $N_η/N_\mathrm{Tanh} \simeq 0.1$ over a smooth wall, and $N_η/N_\mathrm{Tanh} \simeq 0.04$ over typical drag-reducing riblets, with viscous-scaled spacing $15$.

physics.flu-dyn↗

arXiv2026

SALMONN-2: Advancing General-Purpose Hearing Abilities with Self-Supervised Representations

Recent audio large language models (ALLMs) are typically built upon audio encoders trained with large amounts of supervised data. Since self-supervised learning (SSL) audio encoder models are known to learn general-purpose and transferable representations, we investigate whether general-purpose SSL audio representations can serve as an effective foundation for ALLMs. We present SALMONN-2, an ALLM built upon a unified SSL encoder. To better exploit the hierarchical representations learned by SSL encoders, we propose a multi-layer feature fusion (MLF) adapter that aggregates information from all encoder layers before projecting them into the language model. Beyond conventional audio understanding tasks, we further explore multimodal in-context learning (MICL) in ALLMs and study how this capability can be acquired through contextual biasing training. Experimental results show that a general-purpose SSL encoder achieves performance comparable to, or better than, specialised supervised audio encoders while providing a more balanced capability across speech, audio, music and paralinguistic tasks. SALMONN-2 further achieves state-of-the-art performance among comparable-scale open-weight models on ALLM understanding benchmarks, obtaining the best results on MMAU-Pro, MMAR and MMSU. We also show that MICL does not emerge naturally in ALLMs, but can be effectively acquired through targeted contextual biasing training.

eess.AS↗

arXiv2026

Dual-Mapping Sparse Vector Coding for Phase Noise-Resilient Short-Packet Transmission

Sparse vector transmission (SVT) has emerged as a promising technique for ultra-reliable low-latency short-packet communications. However, existing SVT schemes typically assume negligible phase noise (PN), an assumption that rarely holds in practical wireless systems. In this paper, a dual-mapping sparse vector coding (DM-SVC) scheme is proposed for short-packet communications subject to PN. In DM-SVC, pilot symbols are mapped onto multiple non-zero blocks and data symbols onto isolated non-zero elements within a single sparse vector, thereby enabling pilot-data separation through distinct sparsity patterns rather than explicit resource partitioning. Moreover, the indices of pilot blocks convey additional information bits, further improving spectral efficiency. A basis expansion model is adopted to represent the PN process, substantially reducing the number of parameters to be estimated. Furthermore, an iterative joint PN estimation and data decoding algorithm is developed, where pilot block indices are first detected exploiting block-sparse priors, after which PN estimation and data decoding proceed iteratively. Simulation results show that DM-SVC could achieve block error rate performance close to that of perfect PN compensation, while offering improved spectral efficiency and reduced codebook storage overhead compared to state-of-the-art SVT schemes.

eess.SP↗

arXiv2026

Long-Term Simulation Exposes Cognitive-Developmental Risks in AI Companions

AI companions powered by large language models increasingly interact with cognition-developing users, including children and adolescents, creating risks that may accumulate over time. Existing safety evaluations largely rely on single-turn or short-session tests, which cannot capture risks that emerge only through prolonged interaction. To address this gap, we propose TSJ (Theater-Stage-Judge), a longitudinal framework combining persona-driven user simulation, dynamic psychological-state updating and retrospective evaluation. We evaluate six mainstream models across four developmental stages, twenty-four risk dimensions and three psychological-vulnerability personas, covering 12,960 simulated person-day interactions. TSJ shows that short-horizon testing systematically underestimates developmental risks, for which TSJ yields a stable risk estimate only after 140 turns within prolonged simulated relationships. Applying TSJ further identifies early childhood and emerging adulthood as the most vulnerable stages, with cognitive trust and emotional dependency as the weakest domains. TSJ provides a scalable methodology for longitudinal cognitive developmental risk evaluation in AI companion systems.

cs.AI↗

arXiv2026

SPEAR: A Unified SSL Framework for Learning Speech and Audio Representations

Self-supervised learning (SSL) has significantly advanced acoustic representation learning. However, most existing models are optimised for either speech or audio event understanding, resulting in a persistent gap between these two domains. We address this gap with SPEAR (SPEech and Audio Representations), a self-supervised framework that distils complementary knowledge from a speech-focused SSL teacher and a general-audio SSL teacher into a single unified model. SPEAR applies multi-codebook vector quantisation to continuous teacher representations to produce fine-grained discrete tokens that capture both semantic and acoustic information. To effectively integrate these heterogeneous representations, SPEAR jointly predicts them given a masked input with an asymmetric pre-training loss. We further improve robustness in complex sound scenes through a novel token mixing mechanism. Extensive experiments demonstrate that SPEAR consistently outperforms existing unified speech and audio models. SPEAR establishes a new state-of-the-art on the SUPERB benchmark, surpassing WavLM Large on 12 of 15 tasks, while achieving competitive performance on the HEAR benchmark. These results position SPEAR as a versatile foundation for general-purpose speech and audio representation learning. The code and pre-trained models will be released.

eess.AS↗

arXiv2026

Towards Paradigm-General Suicide Risk Detection via Speech LLM

Suicide risk among adolescents remains a critical public health concern, and speech provides a non-invasive and scalable approach for its detection. Speech-based suicide risk assessment commonly relies on carefully designed speech elicitation paradigms (\textit{e.g.,} verbal fluency, reading, or question answering) to probe cognitive and affective states. Existing approaches, however, typically focus on one single paradigm at a time. This paper, for the first time, investigates cross-paradigm approaches that unify diverse speech elicitation paradigms within a single model. Specifically, we use a speech LLM as backbone with a mixture of DoRA experts (MoDE) to capture complementary cues across assessments dynamically, tested on 1,223 participants across ten speech elicitation paradigms. Results show that MoDE outperforms both paradigm-specific and conventional joint-learning models. Moreover, it can generalise to unseen paradigms and provide better confidence calibration.

eess.AS↗

arXiv2026

MemPro: Agentic Memory Systems as Evolvable Programs

Long-horizon autonomous agents require memory systems to retain historical information, track evolving states, and reuse relevant knowledge beyond finite context windows. Existing agentic memory systems typically follow a memory construction-retrieval (MCR) pipeline, but often adapt mainly the memory bank while keeping the surrounding pipeline fixed after deployment. This fixed-pipeline design struggles to handle heterogeneous task-specific failure modes and can become misaligned with memory banks that evolve in scale and structure over time. To address these limitations, we propose MemPro, a system-level evolution framework that treats the entire MCR pipeline as an evolvable program rather than adapting only the memory bank or prompt text. MemPro maintains a version tree of runnable memory-system implementations, where an Evolving Agent iteratively selects promising versions, diagnoses recurring failures, and creates improved child versions through failure-mode-guided edit-debug refinement. Experiments on LongMemEval, LoCoMo, HotpotQA, and NarrativeQA show that MemPro consistently outperforms strong static and prompt-level evolving baselines within a few iterations, continues to improve with evolution, and achieves a favorable performance-cost trade-off. Code is available at https://github.com/wanghai673/MemPro.

cs.CL↗

arXiv2026

FAB-Bench: A Framework for Adaptive RAG Benchmarking in Semiconductor Manufacturing

Retrieval-Augmented Generation (RAG) has become critical for knowledge-intensive applications, yet evaluating its performance in vertical domains remains difficult due to domain complexity, diverse context scales, and heavy reliance on expert assessments that are costly, inconsistent, and non-scalable. We introduce FAB-Bench, an end-to-end framework for adaptive benchmarking of RAG systems in semiconductor manufacturing. FAB-Bench defines six diagnostic metrics measuring factual accuracy, contextual utilization, completeness, retrieval relevance, technical depth, and reasoning consistency. The framework couples retriever diagnostics with generator-level reasoning analysis across context windows of 4K-32K tokens, quantifying how retrieval precision and generative fidelity co-evolve as contextual scope expands. From over 1,300 generated candidates, we curated a high-quality benchmark of 200 query-answer pairs spanning three synthesis strategies: needle-in-haystack, intra-document multi-topic, and cross-document multi-hop. Systematic evaluation across four LLMs and four RAG frameworks reveals three distinct context-scaling behaviors: logarithmic growth, early saturation, and cold-start dynamics, and identifies attention dilution as the primary mechanism behind performance degradation at extreme context lengths. Cross-framework validation on three additional production RAG systems confirms evaluation portability.

cs.CL↗

arXiv2026

Forget What's Sensitive, Remember What Matters: Token-Level Differential Privacy in Memory Sculpting for Continual Learning

Continual Learning (CL) models, while adept at sequential knowledge acquisition, face significant and often overlooked privacy challenges due to accumulating diverse information. Traditional privacy methods, like a uniform Differential Privacy (DP) budget, indiscriminately protect all data, leading to substantial model utility degradation and hindering CL deployment in privacy-sensitive areas. To overcome this, we propose a privacy-enhanced continual learning (PeCL) framework that forgets what's sensitive and remembers what matters. Our approach first introduces a token-level dynamic Differential Privacy strategy that adaptively allocates privacy budgets based on the semantic sensitivity of individual tokens. This ensures robust protection for private entities while minimizing noise injection for non-sensitive, general knowledge. Second, we integrate a privacy-guided memory sculpting module. This module leverages the sensitivity analysis from our dynamic DP mechanism to intelligently forget sensitive information from the model's memory and parameters, while explicitly preserving the task-invariant historical knowledge crucial for mitigating catastrophic forgetting. Extensive experiments show that PeCL achieves a superior balance between privacy preserving and model utility, outperforming baseline models by maintaining high accuracy on previous tasks while ensuring robust privacy.

cs.AI↗

arXiv2026

AuDirector: A Self-Reflective Closed-Loop Framework for Immersive Audio Storytelling

Despite advances in text and visual generation, creating coherent long-form audio narratives remains challenging. Existing frameworks often exhibit limitations such as mismatched character settings with voice performance, insufficient self-correction mechanisms, and limited human interactivity. To address these challenges, we propose AuDirector, a self-reflective closed-loop multi-agent framework. Specifically, it involves an Identity-Aware Pre-production mechanism that transforms narrative texts into character profiles and utterance-level emotional instructions to retrieve suitable voice candidates and guide expressive speech synthesis, thereby promoting context-aligned voice adaptation. To enhance quality, a Collaborative Synthesis and Correction module introduces a closed-loop self-correction mechanism to systematically audit and regenerate defective audio components. Furthermore, a Human-Guided Interactive Refinement module facilitates user control by interpreting natural language feedback to interactively refine the underlying scripts. Experiments demonstrate that AuDirector achieves superior performance compared to state-of-the-art baselines in structural coherence, emotional expressiveness, and acoustic fidelity. Audio samples can be found at https://anonymous-itsh.github.io/.

cs.SD↗

arXiv2026

Mapping the Turn: An Eulerian Binormal-Axis Diagnostic for Recirculating 3D Flows

Three-dimensional (3D) recirculating flows are often interpreted qualitatively from selected streamline visualizations. In separated flows, such recirculating motion is central to the drag modulation, but the local orientation of recirculation remains difficult to quantify in a field-based form. This work introduces an Eulerian binormal-axis diagnostic that locally evaluates the orientation of streamline turning at each point in the velocity field, yielding a spatially resolved field of the recirculating direction. Motivated by the Frenet-Serret binormal direction of a curved streamline, the diagnostic uses the velocity vector and its convective acceleration to extract the local streamline-turning axis without requiring explicit streamline integration. The resulting direction is encoded with barycentric RGB weights to visualize streamwise, spanwise, and wall-normal turning axis contributions. The diagnostic is first applied to Hill's spherical vortex, which provides a controlled analytic example of 3D recirculating motion for interpreting the binormal-axis direction and the associated barycentric RGB encoding. It is then applied to the mean field of a pressure-gradient-induced 3D separation bubble. The resulting visualizations show that the diagnostic reveals orientation changes that are not apparent from streamline visualization. The proposed diagnostic therefore converts qualitative streamline impressions into a spatially resolved measure of local streamline-turning orientation, providing a quantitative complement to conventional 3D flow visualization.

physics.flu-dyn↗

arXiv2026

Can Heterogeneous Language Models Be Fused?

Model merging aims to integrate multiple expert models into a single model that inherits their complementary strengths without incurring the inference-time cost of ensembling. Recent progress has shown that merging can be highly effective when all source models are \emph{homogeneous}, i.e., derived from the same pretrained backbone and therefore share aligned parameter coordinates or compatible task vectors. Yet this assumption is increasingly unrealistic in open model ecosystems, where useful experts are often built on different families such as Llama, Qwen, and Mistral. In such \emph{heterogeneous} settings, direct weight-space fusion becomes ill-posed due to architectural mismatch, latent basis misalignment, and amplified cross-source conflict. We address this problem with \texttt{HeteroFusion} for heterogeneous language model fusion, which consists of two key components: topology-based alignment that transfers knowledge across heterogeneous backbones by matching functional module structures instead of raw tensor coordinates, and conflict-aware denoising that suppresses incompatible or noisy transfer signals during fusion. We further provide analytical justification showing that preserving the target adapter basis while predicting structured updates leads to a stable and well-conditioned transfer process. Across heterogeneous transfer, multi-source fusion, noisy-source robustness, and cross-family generalization settings, \texttt{HeteroFusion} consistently outperforms strong merging, fusion, and ensemble baselines.

cs.AI↗

arXiv2026

Metric properties of continued fractions with large prime partial quotients

Let $x \in [0,1)$ with continued fraction expansion $[a_1(x),a_2(x),\dots]$, and let $ϕ:\mathbb{N}\to\mathbb{R}^+$ be a non-decreasing function. We consider the numbers whose continued fraction expansions contain at least two partial quotients that are simultaneously large and prime, that is \[ E'(ϕ):=\Big\{x\in[0,1): \exists\, 1\leq k\neq l\leq n, \ a'_{k}(x),\ a'_{l}(x)\geqϕ(n) \ \text{for i.m. } n\in\mathbb{N}\Big\}, \] where $a'_i(x)$ denotes $a_i(x)$ if $a_i(x)$ is prime and $0$ otherwise. We establish a zero-one law for the Lebesgue measure of $E'(ϕ)$ and determine its Hausdorff dimension.

math.NT↗

arXiv2026

AOT-POT: Adaptive Operator Transformation for Large-Scale PDE Pre-training

Pre-training neural operators on diverse partial differential equation (PDE) datasets has emerged as a promising direction for building general-purpose surrogate models in scientific machine learning. However, the inherent complexity and structural diversity of PDE solution operators make multi-PDE pre-training fundamentally challenging. Existing methods mainly address this by increasing model capacity, while leaving the target solution operators unchanged. Inspired by classical numerical analysis, we instead propose to transform complex and diverse solution operators into simpler, better-aligned forms that are easier to model jointly. Since the optimal transformation varies across PDE types, it must be adaptive and input-dependent, allowing a single neural operator to approximate an entire family of operators. We instantiate this idea as AOT-POT (adaptive operator-transformation for pre-training operator transformer), which expands hidden representations into multiple parallel streams, adaptively aggregates and redistributes them before and after each sub-layer, and mixes streams through Sinkhorn-projected doubly stochastic matrices for stable training. These mechanisms together reshape diverse solution operators into a unified form that can be effectively modeled by a single architecture. Empirically, AOT-POT achieves state-of-the-art performance on 12 PDE benchmarks with only 3\% additional parameters, reducing relative L2 error by up to 77.6\% (40.9\% on average). Fine-tuning AOT-POT further reduces L2 error by up to 92\% on in-domain PDEs and 89\% on out-of-domain PDEs (unseen types during pre-training), demonstrating that adaptive operator transformation is an effective and complementary direction for advancing PDE foundation models beyond simply scaling model capacity.

cs.LG↗

arXiv2026

SLAP: Stratified Loss-based Pruning for On-Policy Data-Efficient Instruction Tuning

Instruction tuning has optimized the specialized capabilities of large language models (LLMs), but it often requires extensive datasets and prolonged training times. The challenge lies in developing specific capabilities by identifying useful data and efficiently fine-tuning. High-quality and diverse pruned data can help models achieve lossless performance at a lower cost. In this paper, we propose \textbf{SLAP}, a novel batch-aware data selection framework that evaluates the learnability of entire batch compositions rather than individual. SLAP ensures comprehensive data distribution coverage through distribution-aware stratified sampling while maximizing intra-batch diversity through relative distance optimization. By leveraging Hessian-approximated gradient information for dynamic batch selection, SLAP significantly outperforms existing state-of-the-art methods across multiple model architectures (LLaMA, ChatGLM) and diverse downstream tasks including multi-turn dialogue, multilingual translation, and question answering. Most notably, SLAP achieves superior performance with 20-40\% less training data compared to full dataset training, substantially reducing computational costs while maintaining or improving model capabilities. These results establish SLAP as a powerful approach for efficient and effective instruction tuning of large language models.

cs.CL↗