arXiv:2601.05169v2 Announce Type: replace Abstract: We show that a general class of social impact models with higher-order interactions on hypergraphs can be exactly reduced to an equivalent model with pairwise interactions on a weighted projected network. This reduction is made by a mapping that preserves the microscopic probabilities of changing the state of the nodes. As a particular case, we introduce hypergraph-voter models, for which we compute the weights of the projected network both analytically and numerically across several hypergraph ensembles, and we characterize their ordering dynamics through simulations of both higher-order and reduced dynamics. For a linear social impact function (hypergraph-linear voter model) the weights of the projected network are static, allowing us to develop a pair approximation that describes with accuracy the time evolution of macroscopic observables, which turn out to be independent of those weights. The macroscopic dynamics is thus equivalent to that of the standard voter model on the unweighted projected network. For a power-law social impact function (hypergraph-nonlinear voter model) the weights of the projected network depend on the instantaneous system configuration. Nevertheless, the nonlinear voter model on the unweighted projected network still reproduces the main macroscopic trends for well connected hypergraphs.
Science Journals
arXiv:2603.16131v2 Announce Type: replace Abstract: The explosive growth of AI research has created unprecedented information overload, increasing the demand for scientific summarization at multiple levels of granularity beyond traditional abstracts. While LLMs are increasingly adopted for summarization, existing benchmarks remain limited in scale, target only a single granularity, and predate the LLM era. Moreover, since the release of ChatGPT in November 2022, researchers have rapidly adopted LLMs for drafting manuscripts themselves, fundamentally transforming scientific writing, yet no resource exists to analyze how this writing has evolved. To bridge these gaps, we introduce SciZoom, a benchmark comprising 44,946 papers from four top-tier ML venues (NeurIPS, ICLR, ICML, EMNLP) spanning 2020 to 2025, explicitly stratified into Pre-LLM and Post-LLM eras. SciZoom provides three hierarchical summarization targets (Abstract, Contributions, and TL;DR) achieving compression ratios up to 600:1, enabling both multi-granularity summarization research and temporal mining of scientific writing patterns. Our linguistic analysis reveals striking shifts in phrase patterns (up to 10x for formulaic expressions) and rhetorical style (23% decline in hedging), suggesting that LLM-assisted writing produces more confident yet homogenized prose. SciZoom serves as both a challenging benchmark and a unique resource for mining the evolution of scientific discourse in the generative AI era. Our code and dataset are publicly available on GitHub (https://github.com/janghana/SciZoom) and Hugging Face (https://huggingface.co/datasets/hanjang/SciZoom), respectively.
arXiv:2603.16346v3 Announce Type: replace Abstract: For over a century, the relationship between Indian summer monsoon rainfall and El Nino-Southern Oscillation has been the foundation of 'long-range' prediction of Indian monsoon. This relation is estimated from correlations between Indian summer monsoon rainfall and a Pacific sea surface temperature-based index of El Nino-Southern Oscillation. However, a prominent multi-decadal variability in the correlation raises doubts on the realism of El Nino-Monsoon relation and stability of Indian monsoon predictability. Previous studies discussed that Pacific-based El Nino-Southern Oscillation indices do not represent El Nino's global influence completely, making their correlation with Indian monsoon unreliable. To address this limitation, a Global El Nino-Southern Oscillation framework based on the depth of the 20 degree Celsius isotherm is developed, integrating subsurface signal from all three tropical ocean basins and maximizing Indian monsoon teleconnections. Contrary to previous findings, the 20 degree Celsius isotherm-based Global El Nino-Southern Oscillation exhibits a strong and stable correlation (greater than 0.8) with Indian monsoon at an 18-month lead. Through a re-examination of the El Nino-Monsoon relationship with the superior Global El Nino-Southern Oscillation predictor, we show that the true relationship is independent of global warming and stationary in time. We discover that the stationarity in the El Nino-Monsoon relationship emerges as a natural consequence of chaotic synchronization between Indian summer monsoon rainfall and 20 degree Celsius isotherm at an 18-month lead. Such synchronization between chaotic climate variables provides a new physical basis for climate predictability beyond the conventional deterministic limit set by chaos. Our findings provide the foundation and renewed confidence in long range climate prediction.
arXiv:2603.19957v2 Announce Type: replace Abstract: Pathology reports are structured, multi-granular documents encoding diagnostic conclusions, histological grades, and ancillary test results across one or more anatomical sites; yet existing pathology vision-language models (VLMs) reduce this output to a flat label or free-form text. We present HiPath, a lightweight VLM framework built on frozen UNI2 and Qwen3 backbones that treats structured report prediction as its primary training objective. Three trainable modules totalling 15M parameters address complementary aspects of the problem: a Hierarchical Patch Aggregator (HiPA) for multi-image visual encoding, Hierarchical Contrastive Learning (HiCL) for cross-modal alignment via optimal transport, and Slot-based Masked Diagnosis Prediction (Slot-MDP) for structured diagnosis generation. Trained on 749K real-world Chinese pathology cases from three hospitals, HiPath achieves 68.9% strict and 74.7% clinically acceptable accuracy with a 97.3% safety rate, outperforming all baselines under the same frozen backbone. Cross-hospital evaluation confirms generalisation with only a 3.4pp drop in strict accuracy while maintaining 97.1% safety.
arXiv:2603.29237v3 Announce Type: replace Abstract: Enforcing prescribed global integral constraints in mesh-free neural PDE solvers is challenging in high-dimensional domains. Existing projection methods for spatial integrals are often tied to fixed grids or uniform quadrature, which can conflict with randomly sampled physics-informed neural networks (PINNs) and scale poorly with dimension. High-order differential operators also increase reverse-mode automatic differentiation memory costs. We propose Stochastic Dimension Implicit Functional Projection (SDIFP), a quadrature-level framework for enforcing prescribed first and second spatial moments. SDIFP replaces tensor-product nodal projection by a global affine correction of the neural-network output, with two scalar coefficients determined from a weighted quadrature rule. Under positive target variance and nonzero empirical raw variance, this correction is the nearest-point projection, in the weighted quadrature norm, onto the empirical two-moment constraint set. Thus, the prescribed moments are exact for the selected quadrature rule, while continuum errors are quadrature errors of the corrected field. For decomposable high-dimensional linear operators, SDIFP combines affine moment correction with stochastic operator-subset sampling. With independent residual and derivative sampling and conditionally unbiased coefficient-gradient estimation, the resulting estimator is unbiased for the specified quadrature-based residual objective; the shared-subset fast mode is biased in general. SDIFP avoids tensor-product quadrature for moment enforcement, separates forward quadrature evaluation from the reverse-mode graph, and retains pointwise inference efficiency once the affine coefficients are fixed or precomputed.
arXiv:2603.29578v2 Announce Type: replace Abstract: Facial Expression Recognition (FER) is essential for human-machine interaction, as it enables machines to interpret human emotions and internal states from facial affective behaviors. Although deep learning has significantly advanced FER performance, most existing deep-learning-based FER methods rely heavily on discriminative classifiers for fast predictions. These models tend to learn shortcuts and are vulnerable to even minor distribution shifts. To address this issue, we adopt a conditional generative diffusion model and introduce the Emotion Diffusion Classifier (EmoDC) for FER, which demonstrates enhanced adversarial robustness. However, retraining EmoDC using standard strategies fails to penalize incorrect categorical descriptions, leading to suboptimal recognition performance. To improve EmoDC, we propose margin-based discrepancy training, which encourages accurate predictions when conditioned on correct categorical descriptions and penalizes predictions conditioned on mismatched ones. This method enforces a minimum margin between noise-prediction errors for correct and incorrect categories, thereby enhancing the model's discriminative capability. Nevertheless, using a fixed margin fails to account for the varying difficulty of noise prediction across different images, limiting its effectiveness. To overcome this limitation, we propose Adaptive Margin Discrepancy Training (AMDiT), which dynamically adjusts the margin for each sample. Extensive experiments show that AMDiT significantly improves the accuracy of EmoDC over the baseline model with standard denoising diffusion training under 100-step evaluations. Additionally, AMDiT-enhanced EmoDC has better generalization and robustness than state-of-the-art discriminative classifiers.
arXiv:2601.04181v2 Announce Type: replace Abstract: Reliable long-term decoding of gestures from surface electromyography (EMG) is hindered by signal drift caused by electrode displacement, muscle fatigue, and/or posture changes. Although modern models achieve high intra-session accuracy, their performance often degrades substantially across recording sessions. Existing approaches to mitigate this problem typically rely on large training datasets or computationally intensive pipelines that are unsuitable for energy-efficient wearable devices. We propose a lightweight test-time adaptation framework for EMG decoding. The framework includes three complementary adaptation strategies: (i) causal adaptive batch normalization for online statistical alignment, (ii) Gaussian Mixture Model alignment with experience replay to mitigate forgetting, and (iii) meta-learning for rapid few-shot calibration. We evaluate these methods on the multi-session NinaPro DB6 dataset. All approaches substantially improve inter-session robustness relative to a non-adaptive baseline while maintaining low computational overhead. Replay-regularized statistical alignment provides the most stable adaptation under limited data, while meta-learning achieves the highest accuracy when sparse calibration labels are available. Overall, our self-supervised test-time adaptation methods reach up to 82% inter-session accuracy, significantly improving upon prior approaches while maintaining resource-efficient operation. These results demonstrate that lightweight test-time adaptation can enable robust, long-term EMG decoding for wearable or prosthetic applications.
arXiv:2603.04560v2 Announce Type: replace Abstract: Recent works use a neuro-symbolic framework for general manipulation policies. The advantage of this framework is that -- by applying off-the-shelf vision and language models -- the robot can break complex tasks down into semantic subtasks. However, the fundamental bottleneck is that the robot needs skills to ground these subtasks into embodied motions. Skills can take many forms (e.g., trajectory snippets, motion primitives, coded functions), but regardless of their form skills act as a constraint. The high-level policy can only ground its language reasoning through the available skills; if the robot cannot generate the right skill for the current task, its policy will fail. We propose to address this limitation -- and dynamically expand the robot's skills -- by leveraging user feedback. When a robot fails, humans can intuitively explain what went wrong (e.g., ``no, go higher''). While a simple approach is to recall this exact text the next time the robot faces a similar situation, we hypothesize that by collecting, clustering, and re-phrasing natural language corrections across multiple users and tasks, we can synthesize more general text guidance and coded skill templates. Applying this hypothesis we develop Memory Enhanced Manipulation (MEMO). MEMO builds and maintains a retrieval-augmented skillbook gathered from human feedback and task successes. At run time, MEMO retrieves relevant text and code from this skillbook, enabling the robot's policy to generate new skills while reasoning over multi-task human feedback. Our experiments demonstrate that using MEMO to aggregate local feedback into general skill templates enables generalization to novel tasks where existing baselines fall short. See supplemental material here: https://collab.me.vt.edu/memo
arXiv:2606.24202v1 Announce Type: new Abstract: A recent unified treatment of PID tuning for time-delay plants (An, Tang, Sun, Zhang and Chen, Automatica, 2026) combines the D-partition method with a boundary gradient vector (BGV) to orient the boundaries of stabilizing, relative-stability and stability-margin regions. That method answers a feasibility question, namely where admissible gains lie, and it leaves a manual interior-point test to fix the unstable-pole count in each cell, with the choice of a single controller left to the user. This note makes three contributions. First, the one operation the BGV leaves manual, the absolute unstable-pole count, is available analytically: exactly for delay-free designs through a companion-matrix or Routh count, and through an argument-principle (Mikhailov) evaluation for retarded-type delay loops. Labelling every cell with its analytic count removes the interior-point test and decides the whole partition. Second, we add the step the BGV framework cannot reach, a time-domain selection rule that returns one certified controller: among monotone step responses we choose the minimum-settling-time PI gains, characterized by a tangency condition, with monotonicity guaranteed by external positivity (a nonnegative closed-loop impulse response). Third, we flag a neutral-type pitfall that the unified analysis never delimits: an ideal PID with derivative action on a first-order-plus-dead-time (FOPTD) plant is of neutral type, with a root chain on the imaginary axis when k Kd = T. We reproduce the authors' delay-free benchmark exactly, recovering both admissible Kp intervals, and demonstrate the full pipeline on a FOPTD plant, delivering a certified monotone, fast-settling PI controller that the region-only method can neither locate nor justify; the selected gains match an independent closed-form tangency rule to within one percent. All claims are validated numerically.
arXiv:2604.17473v4 Announce Type: replace Abstract: Vision-Language Navigation(VLN) requires an agent to navigate through 3D environments by following natural language instructions. While recent Video Large Language Models(Video-LLMs) have largely advanced VLN, they remain highly susceptible to State Drift in long scenarios. In these cases, the agent's internal state drifts away from the true task execution state, leading to aimless wandering and failure to execute essential maneuvers in the instruction. We attribute this failure to two distinct cognitive deficits: Progress Drift, where the agent fails to distinguish completed sub-goals from remaining ones, and Memory Drift, where the agent's history representations degrade, making it lose track of visited landmarks. In this paper, we propose a Dual-Anchoring Framework that explicitly anchors the instruction progress and history representations. First, to address progress drift, we introduce Instruction Progress Anchoring, which supervises the agent to generate structured text tokens that delineate completed versus remaining sub-goals. Second, to mitigate memory drift, we propose Memory Landmark Anchoring, which utilizes a Landmark-Centric World Model to retrospectively predict object-centric embeddings extracted by the Segment Anything Model, compelling the agent to explicitly verify past observations and preserve distinct representations of visited landmarks. Facilitating this framework, we curate two extensive datasets: 3.6 million samples with explicit progress descriptions, and 937k grounded landmark data for retrospective verification. Extensive experiments in both simulation and real-world environments demonstrate the superiority of our method, achieving a 15.2% improvement in Success Rate and a remarkable 24.7% gain on long-horizon trajectories. To facilitate further research, we will release our code, data generation pipelines, and the collected datasets.
arXiv:2602.15245v2 Announce Type: replace Abstract: Reinforcement learning (RL)-based biomechanical simulations have the potential to revolutionise HCI research and interaction design, but currently lack usability and interpretability. Using the Human Action Cycle as a design lens, we identify key limitations of biomechanical RL frameworks and develop MyoInteract, a novel framework for fast prototyping of biomechanical HCI tasks. MyoInteract allows designers to setup tasks, user models, and training parameters from an easy-to-use GUI within minutes. It trains and evaluates muscle-actuated simulated users within minutes, reducing training times by up to 98%. A workshop study with 12 interaction designers revealed that MyoInteract allowed novices in biomechanical RL to successfully setup, train, and assess goal-directed user movements within a single session. By transforming biomechanical RL from a days-long expert task into an accessible hour-long workflow, this work significantly lowers barriers to entry and accelerates iteration cycles in HCI biomechanics research.
arXiv:2602.20592v2 Announce Type: replace Abstract: Speech signals encode emotional, linguistic, and pathological information within a shared acoustic channel; however, disentanglement is typically assessed indirectly through downstream task performance. We introduce an information-theoretic framework to quantify cross-dimension statistical dependence in handcrafted acoustic features by integrating bounded neural mutual information (MI) estimation with non-parametric validation. Across six corpora, cross-dimension MI remains low, with tight estimation bounds ($< 0.15$ nats), indicating weak statistical coupling in the data considered, whereas Source--Filter MI is substantially higher (0.47 nats). Attribution analysis, defined as the proportion of total MI attributable to source versus filter components, reveals source dominance for emotional dimensions (80\%) and filter dominance for linguistic and pathological dimensions (60\% and 58\%, respectively). These findings provide a principled framework for quantifying dimensional independence in speech.
arXiv:2606.24644v1 Announce Type: new Abstract: Millions of people use machine translation (MT) tools daily, yet little is known about their perception of what systems can and cannot do. This paper studies users' mental models of speech translation systems through a new framework based on cross-lingual question answering, where users either accept MT output or request professional re-translation to answer questions based on the information presented in a foreign language. By analyzing user behavior and accuracy trends across varying translation qualities, we examine to what extent they can predict where the system is likely to be wrong, and how this mental model evolves. Users develop stronger mental models with practice, especially when they have some knowledge of the source language, primarily by relying on surface-level error cues. Moreover, providing speech transcriptions can help users develop better mental models. Our results show the promise of cross-lingual question answering as a downstream task for studying MT mental models and advancing our understanding of human-AI collaboration.
arXiv:2606.24655v1 Announce Type: new Abstract: The explosive growth and complexity of product data within the dynamic Brazilian e-commerce landscape demand robust and specialized methods for structured information extraction. Traditional approaches to Product Attribute Value Extraction (PAVE) often struggle with the linguistic nuances and sheer diversity of product descriptions in Portuguese. To address this critical gap, this paper introduces two major contributions. First, we present AI-PAVEBr, a specialized system engineered with Large Language Models (LLMs) to perform high-accuracy PAVE specifically for Brazilian e-commerce catalogs. Second, to facilitate reproducible research and provide a definitive benchmark, we introduce and share the Golden Set, a new, meticulously curated, and manually annotated dataset for PAVE in Portuguese. We detail the creation process and structure (Entity, Category, Subcategories) of this high-quality reference set. Our experiments conclusively show that AI-PAVE-Br, leveraging targeted prompt engineering, dramatically outperforms conventional Named Entity Recognition (NER) baselines. This work not only delivers a superior, scalable solution for a major non-English market but also enriches the NLP community with a valuable, publicly available resource for future PAVE research.
arXiv:2606.24226v1 Announce Type: new Abstract: Crypter-as-a-Service (CraaS) has become a key enabling layer of the contemporary malware economy by providing on-demand evasion capabilities through underground service markets. In this paper, we present a longitudinal characterization of the CraaS ecosystem on exploit.in, a major Russian-language cybercrime forum with a presence on both the clear web and the dark web. From a collection of approximately 1,000,000 posts, we combine keyword filtering, LLM-assisted annotation, and manual validation to extract a corpus of 491 threads and 2,949 posts spanning January 2020 to August 2025. Our analysis shows that crypters on exploit.in are not merely sold as static tools, but as continuously maintained operational services whose value depends on recurring stub renewal - sometimes on a daily basis - sustained antivirus evasion, and trust-based delivery. We develop a taxonomy of five seller types and four buyer profiles, and map the buyer-seller correspondences that structure market transactions. We further document pricing models ranging from low-cost per-build Telegram bot services to high-end custom development and salaried recruitment. Using social-network analysis, we find that the market is hierarchically structured around a small core of highly central actors, many of whom appear to function as trust brokers or other influential intermediaries, while its stability relies on a broader trust and governance infrastructure including escrow, guarantors, reputation systems, and security deposits. Finally, we discuss differences between the CraaS model observed on exploit.in and that reported on HackForums. Although both forums share similar service logics, our corpus suggests that exploit.in exhibits a more professionalized and service-oriented CraaS configuration.
arXiv:2606.24230v1 Announce Type: new Abstract: This paper presents an extension of the Flower federated learning framework to support Semi-Asynchronous Federated Learning. The proposed approach adapts the traditional synchronous paradigm to better handle client heterogeneity and straggler effects. By introducing a semi-asynchronous training strategy, the system allows partial synchronization among clients while maintaining training efficiency and scalability. We implement and evaluate the proposed modification within Flower, instantiated as the FedSaSync strategy, demonstrating improved robustness and reduced idle time compared to fully synchronous baselines in heterogeneous environments. The results show that SAFL can balance convergence stability and system efficiency in heterogeneous environments typical of edge and distributed learning scenarios.
arXiv:2606.24251v1 Announce Type: new Abstract: Large language models exhibit a growing range of misaligned behaviors such as strategic deception, sandbagging, and self-preservation. As they are increasingly deployed in high-stakes settings, it is critical to reliably detect such behaviors to ensure safe and responsible use. In this work, we propose to monitor misalignment by decomposing it into fine-grained cognitive processes -- misalignment indicators -- and detecting their presence in a model's internal activations via linear probes. We develop a taxonomy of 18 indicators spanning different misaligned behaviors, paired with an automated, meta-plan-guided pipeline that generates multi-turn training conversations. To rigorously evaluate generalization, we construct an out-of-distribution suite combining automated behavioral elicitation, established misalignment benchmarks, and natural benign conversations. Across 5 misaligned behaviors, our probes match a strong LLM judge with 0.935 AUROC on out-of-distribution benchmarks while keeping a low false positive rate on benign traffic. We further perform in-depth analysis to understand the probes and the model's internal representations of misalignment indicators.
arXiv:2606.24276v1 Announce Type: new Abstract: Full convexity is an interesting alternative to classical digital convexity since it guarantees connectedness and even simple connectedness in digital spaces Z d , for any dimension d. This paper aims at giving a better understanding of the monotonicity properties of fully convex digital sets, since earlier works showed that the question was difficult for thin fully convex sets. To decipher the hierarchy of fully convex sets ordered by inclusion, we study how we can peel a fully convex set progressively while keeping its full convexity. We provide a characterization of peelable points and fast algorithms to identify them. Furthermore we show that fully convex set can be peeled one point at a time till reduced to the empty set, similarly to digitally convex sets in the classical sense. The peeling of a fully convex set can be seen as an analog to homotopic thinning processes, but with an additional geometric property.
arXiv:2602.23324v2 Announce Type: replace Abstract: Navigation up a smooth sensory gradient is one of the simplest behavioural tasks, and some organisms solve it by making continuous adjustments to their course. Bacteria instead employ a variety of discrete strategies, including run and tumble motion, direction reversals, and turns by specific angles. Here we ask what drives the choice of these strategies, framing the problem as maximising up-gradient speed with a given amount of sensory information per unit time. We find that, without directional information on which way to turn, behavioural strategies that take discrete actions perform better than gradual steering. As the amount of information is increased, we see a series of transitions between optimal strategies, including a shift from direction reversals to fully re-orienting tumbles. Among more complex re-orientation strategies, we show that discrete turn angles are best, and observe transitions in the number of angles employed by the optimal strategy. More broadly, such emergent simplicity in behaviour is a tractable example of a widespread phenomenon in which biology chooses a discrete solution, despite the underlying physics being continuous.
arXiv:2604.23989v2 Announce Type: replace Abstract: Recent work on large language models (LLMs) has emphasized the importance of scaling inference compute. From this perspective, the state-of-the-art method Scattered Forest Search (SFS) has been proposed, employing Monte Carlo Tree Search with carefully crafted initial seeds and textual optimization for multi-turn program correction. However, its complexity makes it unclear what factors contribute to improvements in inference performance. To address this problem, we analyze SFS and propose a simpler method, \textsc{Iterative Refinement of Repair Instructions} (IRRI), which fixes initial programs and iteratively refines repair instructions. Because of the simplicity of IRRI, we theoretically establish the non-elimination of IRRI using Oracle-Guided Inductive Synthesis (OGIS). Experiments on several program generation benchmarks suggest that IRRI achieves inference performance comparable to state-of-the-art methods. These results indicate that, even without complex search structures, refining initial programs with high-quality repair instructions alone can effectively improve inference performance.
arXiv:2605.07565v2 Announce Type: replace Abstract: We study Bayesian Optimisation (BO) in settings where the objective function is influenced by uncontrollable environmental contexts governed by an unknown probability distribution. In practice, the contextual distribution must be estimated from empirical data, a process that inherently introduces distributional mismatch, producing sub-optimal results. While Distributionally Robust Optimisation (DRO) provides a framework to mitigate these risks, existing robust BO methods frequently suffer from high computational complexity, rely on discretisation of continuous context spaces, or impose restrictive assumptions on the structure of the ambiguity set. To overcome these limitations, we propose Ensemble Distributionally Robust Bayesian Optimisation (EDRBO). Our framework leverages the expressive power of ensemble surrogate models to approximate the black-box function while simultaneously accounting for contextual uncertainty. By utilising Wasserstein ball as ambiguity sets, EDRBO provides a robustified acquisition function that remains computationally tractable and natively handles continuous context spaces. We establish a rigorous theoretical foundation for our approach by proving sublinear cumulative regret guarantees of order $\mathcal{O}(\gamma_T \sqrt{T})$, where $\gamma_T$ represents the maximum information gain within the ensemble. Finally, we provide extensive empirical evaluations that corroborate our theory and demonstrate the state-of-the-art performance of EDRBO.
arXiv:2606.24745v1 Announce Type: new Abstract: Generative models, particularly diffusion and score-based approaches, have recently achieved strong performance in speech enhancement, but their iterative sampling process limits real-time deployment. Flow Matching offers an efficient alternative by transporting noisy speech toward clean speech through an ordinary differential equation with few function evaluations. In this work, we propose a skip-free encoder-decoder backbone for flow-matching speech enhancement, guided by Latent Representation Alignment (LRA). Instead of relying on U-Net skip connections, which may transfer noise-correlated low-level features to the decoder, the proposed model aligns its bottleneck and decoder representations with clean latent features extracted from a frozen Descript Audio Codec encoder-decoder without quantization. This codec-aligned supervision promotes compact clean-speech representations while preserving efficient few-step inference. Experiments on WSJ0-CHiME3 and VoiceBank-DEMAND show improved PESQ and perceptual quality, especially on VoiceBank-DEMAND, using only five function evaluations.
arXiv:2605.09102v2 Announce Type: replace Abstract: We study finite element approximations of elliptic and parabolic interface problems with discontinuous coefficients and nonlinear jump conditions. We introduce a scalar interface reduction in which the solution is decomposed into a continuous component and a unit-jump response mode. This representation isolates the interface nonlinearity into a single scalar variable while the bulk problem remains linear. From this perspective, the nonlinear interface condition is reduced to a scalar nonlinear equation, which may be interpreted as a nonlinear Schur complement associated with the interface degree of freedom. The resulting formulation leads to a simple computational procedure consisting of linear solves combined with a low-dimensional nonlinear update. Numerical results for representative elliptic and parabolic problems confirm second-order accuracy for interface quantities and demonstrate the effectiveness of the proposed approach.
arXiv:2605.08245v4 Announce Type: replace Abstract: Vision-Language Models (VLMs) increasingly power high-stakes applications, from medical imaging to autonomous systems, yet they routinely hallucinate, confidently describing content not present in the input. We investigate the root causes of these failure modes with a mechanistic analysis focusing on the decoder-based VLMs. We trace these failure modes to a geometric over-alignment: to bridge the modality gap required by attention mechanisms, decoder-based VLMs over-align visual embeddings with the text manifold, injecting a statistical linguistic bias that systematically overshadows fine-grained visual evidence. While prior work either aggressively closes this gap or suppresses hallucinations through expensive black-box decoding strategies, none addresses the underlying geometric cause. We provide the first quantitative characterization of this over-alignment, demonstrating that linguistic bias concentrates in the top principal components of a universal, dataset-agnostic text subspace. Building on this insight, we propose two complementary remedies: a training-free inference strategy and a bias-aware fine-tuning paradigm, both of which explicitly project out this subspace from visual representations. Our methods significantly reduce hallucinations across POPE, CHAIR, and AMBER benchmarks, and improve CLAIR scores on long-form captioning tasks, with the training-free variant adding no computational overhead over the base model.
arXiv:2605.11305v4 Announce Type: replace Abstract: Dynamic alignment in magnetohydrodynamic turbulence is often interpreted as scale-dependent alignment of counterpropagating Els"asser increments (\delta_r z^\pm), with consequences for inertial-range spectra. We show that standard amplitude-weighted measurements do not establish progressive alignment of typical fluctuations. We separate angular statistics from Els"asser-amplitude weighting and interpret the signal as finite-time retention of amplitude--angle states, tested with Johns Hopkins Turbulence Database simulations and NASA Wind measurements. In the simulations, the unweighted folded angle (\theta_r) between (\delta_r z^+) and (\delta_r z^-), with alignment and anti-alignment folded together, remains only moderately below the random 3D baseline and shows no monotonic decrease across inertial-range separations. Smaller angles in weighted diagnostics are produced mainly by large (A_r=|\delta_r z^+||\delta_r z^-|) events, giving a negative covariance between (A_r) and (\sin\theta_r) that is removed by shuffled controls. Transition measurements show that high-amplitude large-angle states deplete faster than high-amplitude small-angle states. The source--depletion balance reconstructs second-order Els"asser amplitudes and gives an effective rms increment scaling close to (\ell_\perp^{1/4}), although the typical folded angle is nearly scale independent. Mean-log increment-amplitude checks give larger slopes than second-order-amplitude fits in both simulation and Wind data, consistent with stronger intermittent-event weighting of second-order statistics. Wind measurements reproduce the same amplitude--angle hierarchy and negative covariance under Taylor sampling. Conventional dynamic-alignment diagnostics therefore measure selective retention of intense Els"asser fluctuations, not progressive alignment of typical fluctuations.