arXiv subjects

Wen Wu

Publications and source records attributed to Wen Wu.

At least 19 recordsLinked to original sources

arXiv2026

A Pretrained Model for Bayesian Filtering in Heterogeneous Dynamical Systems

Filtering has not yet had its foundation-model moment: state estimation is still dominated either by online Bayesian filtering algorithms that require known dy- namics as inputs, or by learning-based neural filters trained from scratch for a single system or parametric family. The core challenge is structural heterogene- ity: dynamical systems differ in their governing dynamics, physical scales, and noise levels, precluding a shared representation space for a reusable filter. To bridge this gap, we introduce FilterFM, the first pretrained model for Bayesian filtering that estimates latent states across heterogeneous dynamical systems with a single checkpoint. FilterFM employs a hierarchical two-level observation se- quence encoder and a prior-to-posterior state-reconstruction module that together handle varying observation and state dimensions across heterogeneous systems, while modeling state uncertainty via quantile prediction. Trained on a mixture of heterogeneous filtering tasks, FilterFM exhibits superior in-domain performance, delivering significant gains in estimation accuracy, runtime efficiency, and noise robustness compared to existing baselines. Beyond in-domain performance, it marks the first model to successfully enable zero-shot filtering on unseen dynamics, bridging the generalization gap in state estimation. Furthermore, we uncover clear scaling behaviors in foundation-model-based Bayesian filtering, which combined with comprehensive component ablation studies, validate the structural optimality of our design and broaden the generalization scope for this research field.

cs.LG↗

arXiv2026

MENTOR: A Metacognition-Driven Self-Evolution Framework for Uncovering and Mitigating Implicit Domain Risks in LLMs

Ensuring the safety of Large Language Models (LLMs) is critical for real-world deployment. However, current safety measures often fail to address implicit, domain-specific risks. To investigate this gap, we introduce a dataset of 3,000 annotated queries spanning education, finance, and management. Evaluations across 14 leading LLMs reveal a concerning vulnerability: an average jailbreak success rate of 57.8\%. In response, we propose MENTOR, a metacognition-driven self-evolution framework. MENTOR performs metacognitive self-assessment, using strategies such as perspective-taking and consequential reasoning to uncover latent model misalignments. MENTOR couples single-pass rule-guided inference for routine requests with a selectively invoked metacognitive evolution cycle that revises residual unsafe responses, distills successful corrections into a dynamic rule graph, and compiles validated rules into activation-level steering signals for future inference. Experiments demonstrate that MENTOR substantially reduces attack success rates across all tested domains and outperforms existing safety alignment methods. The code and dataset for MENTOR are available at: https://anonymous.4open.science/r/MENTOR-Evo.

cs.AI↗

arXiv2026

MMTA: Benchmarking Multimodal Temporal Analysis with Time Series, Text, and Vision

Time-series analysis is increasingly studied in multimodal settings, yet most benchmarks pair temporal signals with text or with visualizations derived from the same signal. This leaves open whether multimodal large language models (MLLMs) can integrate time series with real-world visual evidence. We introduce MMTA (multimodal temporal analysis), a benchmark of tri-modal tasks with diverse cross-modal complementarity patterns among time series, text, and vision. MMTA spans nine domains -- robotics, autonomous driving, human action, manufacturing, AR assistance, healthcare, recommendation, finance, and climate -- under video+TS and image+TS layouts, with a 3,524-sample test split covering classification and forecasting. The benchmark standardizes heterogeneous real-world sources into a shared schema with domain-specific prompts, provenance documentation, and success-weighted metrics that penalize both wrong answers and failed generations. Evaluating eight open-source and proprietary MLLMs reveals a shared bottleneck: representing time series as text creates very long token sequences and weakens fine-grained alignment with visual evidence, especially in long-video settings. We then introduce TimeOmni-v, a multimodal LLM extension with a dedicated time-series encoder, temporal positional alignment, modality-aware token layout, and a forecasting head. TimeOmni-v improves long-video classification and time-series inference efficiency, showing that MMTA not only exposes representation and grounding failures but also motivates targeted model design.

cs.LG↗

arXiv2026

Target Sensing in Full-Dimensional MIMO-OFDM ISAC Systems Via Coupled Tensor Decomposition

Integrated sensing and communication (ISAC) has been recognized as a key enabler for future wireless communication systems. However, existing ISAC designs based on MIMO-OFDM generally fail to fully exploit the spatial degrees of freedom and the communication data payload, resulting in restricted target sensing performance. To cope with this issue, in this paper we propose a novel target sensing method for full-dimensional (FD) MIMO-OFDM ISAC systems based on coupled tensor decomposition. We first characterize the system model of FD MIMO-OFDM ISAC, where the base station employs a large number of antennas in two dimensions and facilitates both sensing and communication via the data payload of OFDM signals. Then, we formulate the problem of estimating target parameters as a coupled canonical polyadic decomposition (CPD) problem, which jointly leverages the echo signals of communication data payload over multiple subcarriers in the frequency domain. The theoretical analysis indicates that the formulated coupled CPD problem possesses a more relaxed uniqueness condition compared with the ordinary CPD problem, thereby enabling more stable target sensing performance through the pure communication data payload in FD MIMO-OFDM ISAC. In addition, we develop a coupled CPD-based target sensing algorithm, where the target parameters, including azimuth and elevation angles, ranges, and velocities, are extracted from the decomposed factor matrices. Numerical results show that the proposed target sensing method achieves a higher success rate and more accurate parameter estimation than existing tensor decomposition methods.

eess.SP↗

arXiv2026

TVPara: Paralinguistic Captioning of Expressive TV-Series Speech in Complex Acoustic Environments

Large audio-language models (LALMs) struggle to parse paralinguistics (e.g., emotion, pace, pitch) in speech that combines intense emotion with complex acoustic environments. Strengthening this ability calls for training data that reflects such conditions. However, existing corpora rely heavily on studio-recorded speech, offering limited emotional and acoustic diversity. To address this gap, we introduce TVPara, a paralinguistic captioning dataset built from TV series. Speech in TV series often carries diverse emotional expression and co-occurs with scene-specific background sound, providing expressive emotion and complex acoustic environments at scale. Constructed via LALM pre-annotation and human correction, TVPara includes two tasks: Single-Utterance (~492h) and Multi-Utterance (~213h) Paralinguistic Captioning. Fine-tuning open-source LALMs on TVPara yields an average relative improvement of 31.26% in paralinguistic understanding under complex acoustic environments. The gains further generalize to an unseen TV series and to cleaner, cross-domain speech.

cs.SD↗

arXiv2026

Time-JEPA: Grounding Video World Models with Physical Time Series

Video world models learn how visual environments evolve by predicting missing or future content. However, physical-state information can provide useful context for predicting how environments evolve. Prior work incorporates this information through different mechanisms: domain-specific physical laws can complicate multi-domain training, while supervision on individual frames or frame pairs gives limited attention to how physical state evolves over time. We introduce Time-JEPA, which fine-tunes a video world model using low-dimensional, temporally aligned physical time series as auxiliary supervision. Physical time series connect video representations with physical state at corresponding temporal positions, encouraging the model to preserve how physical state evolves. In multiple domains including autonomous driving, robot manipulation, and human motion, Time-JEPA improves current-state decoding, multi-horizon future-state prediction, and kinematic reconstruction over matched video-fine-tuned controls across both condition and state-variable distribution shifts. Multi-domain training provides aggregate gains over matched single-domain training in all three tasks. External evaluations further suggest that time-series supervision improves sensitivity to physical anomalies and the representation of physical interactions, including in unseen scenarios. Aggregate gains on a second video backbone indicate that the method can be applied to more backbones. These results suggest that physical time series help video world models better understand the physical world.

cs.CV↗

arXiv2026

ECHO: Correcting Time Series Foundation Model Forecasts with Historical Analogues and Errors

Time Series Foundation Models (TSFMs) provide general-purpose forecasting capabilities, yet their forecasts can remain misaligned with the temporal characteristics of a target domain. This motivates lightweight forecast correction even when only black-box access to the TSFM is available. Our key observation is that two forms of historical information provide complementary correction signals: historical analogues reveal how similar contexts evolved, while past forecast errors reveal a TSFM's characteristic deviations in the target domain. Based on this observation, we propose ECHO, a black-box forecast correction framework that learns from both sources. ECHO uses AnalogE to assess retrieved historical analogues together with the current context and base forecast, learning transferable corrections through cross-domain pretraining. In parallel, RecentE learns target-specific residual corrections from past forecast errors and refines them as causally available feedback accumulates. ECHO dynamically combines the two corrections according to their observed forecasting performance. Across 5 representative TSFMs and 9 real-world datasets, ECHO reduces MSE in all 45 backbone--dataset settings, yielding an average relative reduction of 9.26\% in macro-average MSE. Its gains also extend to long-horizon forecasting.

cs.LG↗

arXiv2026

Bayesian Speech Synthesizers Can Learn from Multiple Teachers

Text-to-speech (TTS) is a "one-to-many" problem: one text can be spoken in many valid ways, so its realizations form a distribution whose spread depends on the input. Most TTS models do not model this spread explicitly: they leave it implicit in the sampling process or fix it in advance. To model this distribution explicitly, we propose BELLE (Bayesian evidential learning with language modelling), which places a Bayesian prior over the mean and variance of each mel-spectrogram frame and predicts this prior from the input, while keeping the architecture and single-pass sampling of common continuous-valued autoregressive TTS models such as MELLE. Estimating this variance requires several recordings of each text, whereas standard TTS corpora provide one. We therefore construct two parallel-speech corpora: ParallelVox from real human recordings and LS+6TTS, which adds speech from six pre-trained TTS teachers to LibriSpeech (LS). Trained on these corpora, BELLE outperforms MELLE under the same training conditions, is competitive with larger TTS systems on the objective metrics of LibriSpeech test-clean, and supports streaming generation. Analyses of repeated human readings show that real speech has the properties the Bayesian prior assumes and that the variance predicted by BELLE follows the variation across these readings. Parallel speech improves BELLE, and synthetic parallel speech performs comparably to real parallel speech at the same volume and is far easier to scale.

cs.SD↗

arXiv2026

Self-Calibration DOA Estimation for Moving Dilated Arrays with Gain and Phase Errors

Moving dilated arrays (MDAs) can increase the uniform degrees of freedom (uDOFs) while alleviating mutual coupling, but their performance is susceptible to unknown gain and phase errors (GPEs). To deal with this problem, we investigate self-calibration direction-of-arrival (DOA) estimation for MDAs with unknown GPEs. First, multi-position observations are phase aligned to construct an equivalent virtual array with an enlarged aperture. Next, the motion-induced relationships among the GPEs are exploited to establish explicit calibration constraints. By leveraging the orthogonality between the steering vectors and the noise subspace, the joint estimation of the DOAs and GPEs is formulated as a constrained optimization problem. To solve this problem, a two-stage alternating optimization (AO) algorithm is developed. In the first stage, the GPEs are estimated in closed form with the DOAs fixed. In the second stage, the estimated GPEs are used to calibrate the covariance matrix, after which the DOAs are refined using the multiple signal classification (MUSIC) algorithm. To obtain reliable initial estimates, a rotational invariance-like (RIL) initialization method is further developed by exploiting two translated virtual subarrays with identical GPEs. Furthermore, the Cramer-Rao bound (CRB) is derived to characterize the theoretical performance limit of the considered system. Simulation results demonstrate that the proposed method converges reliably and achieves accurate DOA estimation under different array apertures, numbers of snapshots, GPE levels, and numbers of sources.

eess.SP↗

arXiv2026

OWLET: On-Policy Weighted Learning via Evidence Transfer for Video Large Language Model Post-Training

Post-training has become a key stage for eliciting reasoning capabilities in Video Large Language Models (Video-LLMs). Existing post-training methods mainly follow two paradigms: reinforcement learning (RL) and on-policy distillation (OPD). However, they do not fully exploit visual evidence for effective training. RL supervises the policy with a single trajectory-level reward, which is too coarse-grained for video reasoning: the same response-level advantage is applied to every generated token, disregarding that the tokens of a video response are grounded in the visual stream to very different degrees. OPD provides token-level supervision through output-distribution matching, but lacks direct supervision of the spatiotemporal representations that produce those outputs. We observe across extensive experiments that the same model reasons more accurately when its video view combines global temporal context with question-relevant details. Comparing the same student-generated response under this view and uniform sampling reveals changes in token likelihoods and internal representations, providing a training signal for evidence use. Building on this observation, we propose OWLET, an evidence-privileged RL method that derives training signals from a richer teacher view. This view couples uniform global context with question-conditioned local windows, providing visual evidence without requiring a larger model. To better exploit this privileged visual evidence, OWLET introduces (i) \textbf{evidence-aware credit redistribution}, which compares token likelihoods across the two views and redistributes positive response-level advantages according to their relative evidence support; and (ii) \textbf{spatiotemporal state alignment}, which aligns the two views' hidden states on verifier-accepted responses.

cs.CV↗

arXiv2026

Explainable and Generalisable LLM-based Cognitive Decline Detection with Spontaneous Speech

Alzheimer's disease (AD) and mild cognitive impairment (MCI), which may precede AD, manifest early through subtle linguistic and acoustic alterations. Traditional diagnostics, however, are often resource-intensive and lack scalability for mass screening. To address these challenges, we introduce a novel bilingual speech large language model framework for automated, explainable cognitive screening. Unlike conventional pipelines that rely on error-prone automatic speech recognition, our system directly processes raw speech to learn joint acoustic-semantic representations, preserving critical prosodic cues often lost in transcription. Utilising our newly collected PUTH-AD dataset alongside multiple open-source corpora, we implemented a multi-task learning objective that simultaneously performs cognitive status classification and generates clinician-understandable natural language explanations. Our system achieved the highest average accuracy and AUROC across six dataset/task conditions, comparing three representative baselines. The system demonstrated cross-task transfer to held-out PUTH-AD task subsets, maintaining classification accuracy on an entirely unseen cognitive task without task-specific fine-tuning. Furthermore, clinician evaluation confirms that the generated explanations are both clinically relevant and largely consistent with the underlying speech evidence, supporting their potential utility in clinical interpretation. This study provides a scalable, objective, and explainable framework for speech-based cognitive screening, combining cognitive status classification with natural language explanations that clinicians can assess and verify, bridging the gap between advanced AI and clinical utility.

eess.AS↗

arXiv2026

Self-Evolving Time-Series Forecasting Agents with Episodic Memory and Online Policy Learning

LLM-based agents are increasingly used for time-series forecasting because they can organise contextual information, perform multi-step analysis, and guide the sequence of actions required to complete forecasting tasks. Most existing agents focus only on the current forecasting instance. However, in real-world deployments, forecasting commonly operates online, with new forecasts issued from the currently available history as the forecast origin advances and the ground-truth targets of earlier instances progressively become available. These targets provide feedback on the actions taken in earlier instances, yet existing agents generally do not preserve or utilise this information to adapt their subsequent actions. To address this limitation, we introduce FASE, a Feedback-Aware Self-Evolving forecasting agent that converts such feedback into task-specific experience for subsequent forecasting instances. FASE combines episodic memory, which retrieves relevant completed instances, with online policy learning, which summarises the feedback accumulated across instances into ranking guidance. The proposed framework is evaluated on 29 dataset configurations selected from the GIFT-Eval benchmark. Across these 29 configurations, FASE attains the strongest aggregate point forecasting performance among the evaluated methods and reduces the normalised MAE by 9.1% relative to the best individual foundation model baseline. The results further indicate that the cumulative advantage of FASE increases as delayed feedback accumulates. Together, these findings demonstrate that FASE can continually self-evolve through feedback from completed forecasting instances without updating the parameters of the LLM.

cs.LG↗

arXiv2026

ERR+: Sequential Entropy Resolution for Efficient and Decisive LLM Reasoning

Large reasoning models achieve strong performance on complex tasks by generating extended chain-of-thought (CoT) traces via reinforcement learning with verifiable rewards (RLVR). While current RLVR methods have achieved strong results with correctness-based reward signals, they provide limited guidance on the quality of the reasoning process itself, leaving the internal reasoning structure largely unoptimized. Through empirical analysis across multiple model families, we identify a consistent pattern: correct reasoning trac es exhibit more frequent and larger token-level entropy drops within the thinking phase than incorrect ones. We propose ERR+, a two-phase RLVR framework grounded in this observation. The first phase trains with the Entropy Relief Reward (ERR), a bonus proportional to cumulative token-level entropy drops in the thinking phase, log-normalized by response length. Unlike prior methods that suppress entropy, ERR rewards the resolution of uncertainty while leaving exploratory high-entropy states unconstrained. The second phase introduces the Robust Relative Efficiency Reward, which scores each response's length against co-generated peers via a $\tanh$-transformed within-group $z$-score. We provide a formal analysis showing that joint optimization of the two objectives induces gradient conflict in early training, motivating the sequential design . Experiments on five datasets demonstrate consistent improvements in both accuracy and response conciseness across model backbones. Our code is available at https://github.com/XrkArul/err_response

cs.LG↗

arXiv2026

From Time to Channels: Robust and Efficient Local Flaw Detection in Steel Wire Ropes Using Tri-Axis MFL Signals

Steel wire ropes (SWRs) are critical load-bearing components whose local flaws (LFs) pose serious safety risks. Magnetic flux leakage (MFL) inspection commonly detects LFs from temporal or axial morphology, which can change with the sensing axis and operating condition. We show instead that LF responses remain localized over neighboring channels of a circular array. This observation motivates a channel-feature-oriented (CFO) detector that removes smooth channel backgrounds, applies circular matched filtering, and fuses spatially co-located tri-axis responses. Experiments performed on real-world equipment show that CFO attains the highest localization performance among three representative baselines, reaching F1@0.5/F1@0.7 scores of 73.9%/59.5%. It attains the best temporal localization performance under all four conditions and achieves at least 19.4 times the throughput of the evaluated signal-processing baselines. These results demonstrate that circular channel locality provides a robust LF representation across the evaluated operating conditions.

eess.SP↗

arXiv2026

Partial Superimposed Pilot-Aided Sparse Vector Transmission for High-Mobility URLLC

A partial superimposed pilot-aided sparse vector transmission (PSP-SVT) scheme is proposed for short-packet ultra-reliable and low-latency communications in high-mobility scenarios. Unlike conventional full superimposed pilot-aided SVT schemes, the proposed PSP-SVT scheme deploys only a few pilots over a subset of subcarriers. This sparse pilot structure is sufficient for basis expansion model based channel tracking while effectively reducing pilot-data interference. Based on the PSP pattern, an iterative receiver is developed to jointly perform channel estimation and data decoding. The reduced pilot interference in PSP-SVT provides more accurate initial channel estimation and data detection, thereby improving the subsequent iterative refinement and mitigating their error propagation. Moreover, the impacts of the number of PSPs and power allocation ratio on block error rate (BLER) performance are investigated to reveal the near-optimal pilot configuration. Simulation results show that the proposed PSP-SVT scheme outperforms existing full superimposed pilot-aided SVT schemes in terms of BLER with fast convergence speed.

eess.SP↗

arXiv2026

Self-Calibration DOA Estimation for Movable Antenna Systems with Antenna Position Errors

In this letter, we investigate the direction-of-arrival (DOA) estimation problem for wireless sensing with movable antenna (MA) systems in the presence of unknown antenna position errors (APE). To achieve robust wireless sensing, we transform the DOA estimation problem with APE into an optimization problem via the orthogonality between the steering vector and the noise subspace. Then we propose an alternating optimization (AO)-based self-calibration estimation, which consists of two stages and iteratively estimates the APE and DOA. Specifically, in the first stage, by fixing the APE, the problem reduces to the classical DOA estimation problem, which is solved using the multiple signal classification (MUSIC) algorithm. In the second stage, we fix the DOA to estimate the APE. By applying the Lagrange multiplier technique to the subproblem, we obtain a closed-form expression for the APE estimation. Simulation results demonstrate the superior DOA estimation performance of the proposed self-calibration algorithm for MA systems compared to the existing approaches.

eess.SP↗

arXiv2026

Tensor Decomposition-Based Wireless Sensing for MIMO-OFDM ISAC via Flexible Spatial-Temporal-Spectral Optimization

Integrated sensing and communication (ISAC) is regarded as a key enabling technique in future 6th-generation (6G) mobile communication systems. However, existing multi-input multi-output (MIMO) orthogonal frequency division multiplexing (OFDM) ISAC designs generally rely on the fixed-position antennas and fixed allocation of time-frequency resources, thereby limiting the degrees of freedom of wireless sensing along the spatial-temporal-spectral dimensions. In this paper, we propose a novel wireless sensing framework for MIMO-OFDM ISAC systems with flexible spatial-temporal-spectral optimization and propose a tensor decomposition-based approach to estimate target parameters, including azimuth/elevation angles, ranges, and velocities. Specifically, we first establish a monostatic wireless sensing model for MIMO-OFDM ISAC systems, where the positions of antenna elements, the allocation of OFDM symbols and subcarriers can be flexibly configured. Then, we formulate the problem of estimating target parameters as a tensor decomposition problem admitting to the canonical polyadic format, which enables the parallel target parameters estimation process from corresponding factor matrices along the spatial, temporal, and spectral dimensions, respectively. Based on the decomposed factor matrices, we derive the Cramer-Rao Bound (CRB) for the unknown target parameters and reveal that the estimation accuracy of azimuth/elevation angles, velocities and ranges is fundamentally determined by the array geometry, the distribution of OFDM symbols and subcarriers. Building on this insight, we obtain an optimized solution for the positions of antenna elements, and optimal solutions for the subcarrier allocation and OFDM symbol allocation to minimize the CRB, as well as the mean square error of target parameters estimation.

eess.SP↗

arXiv2026

Intern-S2-Preview: Scientific Agentic Foundation Model

Scientific discovery increasingly requires AI systems that can reason over scientific evidence of heterogeneous modalities, interact with scientific tools and environments, and sustain progress across long task horizons. We present Intern-S2-Preview, a series of scientific agentic foundation models designed to support multimodal scientific understanding, reasoning, generation, and long-horizon tasks. The training pipeline begins with scientific multimodal pre-training over rendered scientific documents, interleaved image-text data, and diverse scientific corpora. Starting from the pretrained checkpoint, we apply a unified post-training pipeline consisting of supervised fine-tuning, scalable multi-task reinforcement learning (RL), black- and white-box agentic RL, and on-policy distillation. This pipeline is supported by practical techniques that improve rollout and training stability and efficiency, including partial rollout with off-policy correction, adaptive length regularization, online speculative decoding, robust multi-task optimization, and trace-aware experience assembly for agentic tasks. At the architecture level, Intern-S2-Preview-397B extends time series modelling from efficient long-sequence understanding to numerical forecasting, while Memory Decoder is studied as a separate memory-augmented path for rapid scientific specialization without modifying the frozen 397B backbone. Evaluations across scientific, multimodal, agentic, and general-purpose benchmarks show that Intern-S2-Preview-397B achieves competitive or leading results in multiple settings. The time series modules improve scientific signal understanding and forecasting on SciTS, while the separate Intern-MemDec-4B extension improves the Biology-Instructions average score from 56.92 to 60.32 without modifying the frozen 397B backbone.

cs.LG↗