Papers

Document retrieval is a crucial component of many modern AI systems, directly influencing their effectiveness, robustness, and fairness in downstream tasks. While recent years have seen a growing number of retrievers, comparative studies in the literature are typically limited in scope or focused on singular benchmarks, domains, or model families. This fragmentation makes it difficult to draw reliable conclusions about the relative strengths, weaknesses, and trade-offs of document retrievers. To address this gap, we conduct a large-scale empirical evaluation of document retrievers, covering three families (sparse, dense, and expansion-based) and evaluating 33 retrievers across seven IR datasets, analyzing retrieval quality, runtime, and failure points. Rather than tuning each model individually, we evaluate every retriever off the shelf, under the configuration reconstructable from its public documentation and a uniform compute budget. Our results show that NV-Embed-v2 achieves the strongest performance on four of the seven datasets, albeit at the cost of substantial query latencies. Among sparse retrievers, we find that SPLADE-v3 rivals the top-performing approach despite much lower latency, and even achieves top scores on MS MARCO. On instruction-following datasets, GritLM delivers the best performance. Finally, an analysis of the retrievers' failure points reveals contrasts between models and families that indicate potential for unrealized gains in retrieval performance.

LLM agents, notably Claude Code and OpenAI Codex, are emerging as versatile tools beyond coding agents only. These agents can be enhanced with skills—reusable artifacts that package domain knowledge, workflows, and tool-use instructions. To date, however, little is known about how such skills are designed or how they affect agent effectiveness and behavior in practice. In this paper, we investigate these questions in smart contract security auditing, a domain in which agents have shown substantial promise. We systematically collect 83 smart contract audit skills from the wild and evaluate them on EVMBench across seven agent–model configurations. Our study examines three dimensions: (i) the design characteristics of audit skills, including their structure, knowledge representations, workflows, and tool dependencies; (ii) their effectiveness in improving vulnerability detection; and (iii) their influence on agent execution trajectories. We find that audit skills are mostly lightweight but heterogeneous in design, covering a broad yet imbalanced range of vulnerability types. Their effectiveness is determined primarily by the model rather than the agent harness: Codex/GPT-5.5 achieves the largest gains, improving detection score by 22.8% and captured award by 43.2%. We further find that skill triggering is a key bottleneck. When triggered, skills preserve a shared six-stage audit workflow while exhibiting distinct loading patterns and differential effects on agent behavior across configurations. We release our skill corpus and artifacts to support future research.

We formalize slate recommendation as a randomized score learner followed by deterministic selection. First, an appropriately scoped differential-privacy guarantee passes through selection and its audit trace by post-processing. End-to-end privacy holds only when selector inputs are public or independent, previous private outputs, or separately privacy-accounted; fixing raw state or candidate information instead yields only a conditional guarantee. Second, we derive a logged margin certificate: bounded score-induced objective movement below half the smallest greedy decision margin guarantees that the ordered slate is unchanged. Controlled fixed-margin tests show near-linear exponent scaling, with an empirical slope of $-0.220$ (95% CI $[-0.231,-0.210]$) against the independent-noise reference $-1/4$. Real-anchor experiments on OULAD, MovieLens-25M, and Amazon Musical Instruments show that greater anchor weight reduces score-noise-induced ranking churn. OULAD and EdNet certificate checks validate the implementation of the logged inequality, while closed-loop simulations show bounded target drift and setting-dependent downstream utility. The contribution is therefore a privacy-scope contract and a certifiable score-to-slate stability mechanism, not a universal utility claim.

Voice agents must complete users' tasks despite noise, reverberation, and competing speech. Evaluating agents' robustness therefore requires following how acoustic conditions affect the conversation and the actions taken on the user's behalf. This overview examines what existing benchmarks reveal about agents' ability to complete tasks under acoustic stress and where further task-based evaluation is required. We then introduce TRACE, a practical workflow for designing, running, and interpreting evaluations of acoustic robustness in task-oriented human-agent interactions: the same agent attempts a specified task with an original recording and an acoustically stressed copy, and the resulting conversations are scored for task completion, wrong actions, recovery, and user effort. Finally, we explain how results from these evaluations can guide changes to an agent to prevent wrong actions and improve recovery.

We study a variation of the classical Maximum (Minimum) Leaf Spanning Tree problem. In many applications, Depth-First Search (DFS) is used to compute a spanning tree of a graph. Such a search tree is constructed by connecting each vertex $v$ with the last vertex the search has visited before $v$ and we call this a last-in tree. By restricting the Maximum (Minimum) Leaf Spanning Tree problem to last-in trees of a graph search, we ask for a search ordering that leads to the largest (smallest) number of leaves in its search tree. Recently, Bergougnoux et al. [Journal of Computer and System Sciences 154 (2025)] have studied the parameterized complexity of these problems for DFS. They showed that the minimization problem is para-$\mathsf{NP}$-hard and the maximization problem is $\mathsf{W}[1]$-hard when parameterized by the number of leaves. When parameterized by the number of internal vertices, both problems have polynomial kernels. Here, we examine whether these results also hold for the variant Lexicographic DFS (LDFS). We show that the hardness results of DFS can be transferred to LDFS. We also present exponential kernels for the number of internal vertices as the parameter. We complement this by showing that polynomial kernels do not exist, unless $\mathsf{NP} \subseteq \mathsf{coNP} / \mathsf{poly}$. We also consider last-in trees of searches that do not follow the DFS scheme. In contrast to (L)DFS, minimizing the number of internal vertices is para-$\mathsf{NP}$-hard for several searches including Breadth-First Search.

We show that non-Fuchsian geometrically finite hyperbolic structures on a given hyperbolizable 3-manifold M are uniquely determined, up to isotopy, by their bending laminations. Consequently, the bending map from the space of non-Fuchsian geometrically finite structures on M, endowed with the strong topology, to the space of bending laminations, endowed with Lecuire's tubular topology, is a homeomorphism. This extends the convex co-compact case established with Schlenker. The proof combines hyperbolic Dehn filling, continuity and properness of the bending map (Lecuire) and real-analyticity of its fibres (Bonahon). We also establish a criterion for contractibility of the boundary fibres of a continuous extension of a homeomorphism, extending Finney's theorem to a boundary setting. This topological result may be of independent interest.

Building on the work of Pach, Rubin, and Tardos, we prove that every set of $n\ge2$ points in the real plane with no three collinear contains at least $cn$ pairwise crossing segments with distinct endpoints, for an absolute constant $c>0$.

We present YODAS v3, a weakly-labeled speech corpus containing over 1.1 million hours of 48kHz multi-channel audio in 147 languages, released under a CC BY 3.0 license. YODAS v3 is not only the largest open speech dataset to date, but also the first truly large-scale speech corpus with high-fidelity stereo audio. We first provide the collection methodology for the corpus, where we introduce new techniques for gathering language-balanced speech data. The effectiveness of our approach is shown by the language distribution of the crawled data: 22 languages in YODAS v3 have over 10K hours and 73 languages have over 5K hours of data. We then conduct extensive analyses on the composition of the data, such as the distribution of languages, audio quality, and transcription quality. Finally, we train baseline speech recognition and neural codec models to show the effectiveness of the dataset. Download at https://huggingface.co/datasets/espnet/yodas3.

Smart maritime infrastructures provide continuous access to heterogeneous sensing streams, enabling repeated experimentation, digital-twin development, and AI-based maritime services. However, sensing hardware alone is not sufficient for scene-specific model development: historical video streams must also be spatially indexed, contextualized, and reduced to informative subsets for annotation. This paper presents a frame-to-panorama localization and context-aware sampling pipeline for ship detection in historical PTZ maritime video lacking reliable pan, tilt, and zoom metadata. The main contribution is an end-to-end data-curation approach that recovers camera-view information from historical PTZ video and combines it with environmental context and visual diversity to construct compact, scene-specific training sets. Specifically, frames are localized on a reference panorama using SuperPoint and LightGlue, enriched with weather and solar-state metadata, and selected through diversity sampling to preserve variation across camera view and environmental conditions. A second context-aware stage targets under-represented distant-vessel cases near the horizon using tile-level visual embeddings and Gaussian Mixture Model clustering. Applied within the CMMI MDigi-I Smart Marina testbed, the proposed pipeline reduces 40,718 candidate frames to 220 images for annotation, corresponding to a 99.5% reduction. A YOLO26-m detector fine-tuned on this subset achieves a mean AP50 of 94.78% $\pm$ 0.51% and a mean AP50-95 of 75.10% $\pm$ 1.73% under sequence-grouped five-fold cross-validation. These results demonstrate that highly redundant infrastructure video streams can be transformed into compact, spatially and contextually diverse training sets for scene-specific detector adaptation while substantially reducing annotation effort.

Reducing communication in derivative-free decentralized learning requires controlling the disagreement accumulated over multiple local updates. This paper develops SPADE-DFL, a primal–dual method that allows the number of local function-value updates between neighbor exchanges to grow with the computation budget while preserving the nonprivate convergence order. For smooth nonconvex objectives under uniform query-moment bounds, the prescribed nonprivate schedule achieves a time-averaged stationarity and consensus bound of $\mathcal{O}(T^{-1/3})$ using only $\Theta(T^{2/3})$ communication rounds, where $T$ is the number of local updates per client. For private training, the accumulated data-dependent increment is isolated from the graph correction, allowing one protected state per client and round to generate all outgoing messages. We prove client-level differential privacy for the full interactive transcript and quantify the resulting optimization error over a finite horizon. Experiments on four classification tasks show that SPADE-DFL achieves higher mean test accuracy than existing decentralized learning methods.

We audit a multilingual affective generation benchmark eight instruction-tuned LLMs producing emoji summaries for 17,100 Bangla, English and Hindi sentences, with 6,960 human judgements and find its headline conclusions to be artefacts of the measurement instrument rather than properties of the systems. Treating annotators as a random rather than a fixed factor, no system differs significantly from any other ($F(7,14)=0.59$, $p=0.76$), although the conventional analysis declares 19 of 28 pairwise differences significant. Annotator identity explains far more rating variance than system identity, and the winning system changes whenever any single annotator is removed. The ordering that does emerge tracks output length: mean emoji count explains 78.7% of between-system variance, and a within-item length-matched comparison over 2,599 pairs reverses the leaderboard. We further show that cross-provider anisotropy differences vanish under mean-centring, that per-language token costs change sign with the normalising unit, and that multi-view row-wise splits inflate macro-F1 by $3.1$ points and change the top-ranked system. In place of preference scoring we propose **emoji-affect decodability**, a reference-based probe whose rankings are stable to $\pm0.003$ macro-F1 across seeds.

Deep search requires LLM agents to decompose complex queries, search for evidence, and synthesize grounded answers, yet existing ReAct-style agents suffer from two limitations: role coupling, where one policy must handle planning, evidence use, and synthesis; and context accumulation, where growing search histories introduce noise and obscure useful information. To address these issues, we propose IterSynth, a role-decoupled and summary-based paradigm that alternates between a Planner for identifying information needs and a Synthesizer for integrating evidence into an evolving summary state. This design separates planning from synthesis while using the summary as the persistent state of search, reducing both capability coupling and context noise. To train IterSynth effectively, we further introduce Role-Decoupled Policy Optimization (RDPO) for reinforcement learning, which combines terminal outcome rewards with turn-level rubric evaluations and computes role-specific advantages for more precise credit assignment. Experiments on five long-horizon deep-search benchmarks such as BrowseComp and Xbench-DS show that IterSynth-8B achieves an average score of 50.7, surpassing the strongest prior $\leq$8B agent by +4.2%. Moreover, IterSynth serves as a model-agnostic prompting paradigm, delivering substantial zero-shot gains over ReAct and similar prompting paradigms on frontier proprietary models.

Head-pose variation introduces substantial appearance transformations in visual speech recognition (VSR), making pose-aware feature modulation desirable. However, performance degradation and unwanted feature interactions may result from using numerous Feature-wise Linear Modulation (FiLM) circuits with fixed modulation intensity. We propose a Pose Adaptive Dynamic FiLM framework with a Dynamic Residual FiLM (DR-FiLM) modulator that predicts input-dependent weights to adaptively control the strength of pose-conditioned modulation. Experiments on LRS2 and LRS3 demonstrate that unweighted multi-pathway modulation substantially degrades phoneme recognition, increasing PER to 20.33% and 29.42%, respectively, compared with 16.20% and 20.96% for the single ResFiLM configuration. In contrast, the proposed DR-FiLM with dynamic Deep-Res weighting reduces PER to 15.74% on LRS2 and 23.91% on LRS3, substantially mitigating the adverse effects of unweighted modulation. The analysis of the learned weights further reveals a consistent tendency to assign greater weight to the deeper FiLM pathway as head-pose variation increases. These results show that merging pose-conditioned FiLM circuits is more efficient when the modulation strength is dynamically controlled.

The Method of Regions is a systematic way to derive asymptotic expansions of Feynman integrals. For Euclidean integrals there is a well-established algorithm to determine the complete set of regions as facets of a Newton polytope defined by the graph polynomials. In many physically relevant Minkowski limits, additional regions known as hidden regions are needed to obtain the correct asymptotic expansion. By making a precise connection with pinch singularities in parameter space and the associated Landau conditions, we identify the general mechanism giving rise to these regions and devise an algorithm, the Hidden Region Finder, to determine them. We show that hidden regions arise through a delicate interplay between asymptotic scaling of the edge parameters, which enhances certain monomials of the graph polynomial, and cancellations amongst these monomials, which reduce their collective contribution to the same order as other terms in the polynomial. We explore massless four-, five- and six-point integrals in a variety of wide-angle mass-shell, planar, collinear, double-collinear and Regge limits. Across these different expansions, hidden regions recur in the same small set of seed topologies. In all cases studied, they can be traced to wide-angle configurations with multiple hard-scattering subdiagrams. Distinct kinematic limits of a given seed inherit a common singular locus, while each limit fixes its own scaling vector and cancellation depth. This exploratory study showcases the potential of the proposed algorithm both to uncover missing contributions to asymptotic expansions and to organise them through their underlying singular geometry.

Nearly all galaxies are believed to host a supermassive black hole at their center. In the context of galaxy mergers, the SMBHs from the progenitor galaxies are expected to form a bound binary system. Given the high frequency of such mergers throughout cosmic time, binary SMBHs should be relatively common in active galactic nuclei (AGNs). Yet, their direct detection remains a major observational challenge, with most candidates inferred only through indirect signatures. The quasar Markarian 231 presents unique ultraviolet-optical properties, including an atypical drop in the total flux around 2500 A and a strong wavelength-dependent polarization. In this work, we explore the potential of spectropolarimetry as a tool to detect and characterize SMBH binaries in AGNs. We construct a simple analytical model in order to reproduce the spectral energy distribution (SED) of Mrk 231 taking into account a binary SMBHs scenario and we compute the resulting total and polarized flux. Our goal is to test if the presence of two distinct sources can reproduce the unique ultraviolet-optical properties observed in Markarian 231. Such investigations support the case for future high-sensitivity UV spectropolarimeters in order to probe the inner structure of quasars. In particular, the next-generation instrument POLLUX, proposed for the Habitable Worlds Observatory, would be particularly suited to provide new insights into AGNs physics and supermassive black hole binaries in quasars.

Studying heterogeneous treatment effects has become essential in experimental and observational studies. A critical assumption for obtaining reliable treatment effect estimates is overlap, which requires that treated and control units have sufficiently similar covariate distributions. Poor overlap may limit the effectiveness of estimators, especially those based on propensity scores, potentially leading to unreliable results. We investigate the effectiveness of kernel balancing (KBal) (Hazlett, 2020) as an alternative to propensity score methods for conditional average treatment effect (CATE) estimation, particularly in settings with overlap violations. Building on optimization-based balancing approaches, we integrate KBal weights into tree-based methods, specifically, causal forests (Athey et al., 2019) and the X-Learner (XRF) (Künzel et al., 2019), to assess their impact on bias reduction and estimation precision. Monte Carlo evidence shows that KBal achieves near-exact balance in a transformed feature space, thereby improving treatment effect estimation in cases where traditional reweighting methods struggle due to extreme weights, finite-sample bias, or insufficient removal of pre-existing confounding bias. We apply the proposed methods to the semi-synthetic IHDP benchmark dataset. Overall, the results indicate that KBal leads to performance improvements, especially in settings with nonlinear treatment effects and limited overlap, making it a useful alternative to propensity score methods.

The Poincaré recurrence theorem for a holomorphic correspondence provides a natural framework for studying the first return time and a dynamical system thus induced, in this setting. After establishing a version of the Kac's theorem on the average of the first return times for a holomorphic correspondence, we introduce the concept of an induced correspondence, akin to the case of dynamics of maps and prove that it inherits the ergodicity of the holomorphic correspondence. Additionally, we characterise the ergodic measures with respect to the considered holomorphic correspondence, establish the density of forward orbits under mild conditions and prove a Rokhlin-Kakutani type lemma.

The bifurcation phenomenon in a singularly perturbed one-phase free boundary problem governed by the g-Laplacian is established under prescribed boundary conditions. This bifurcation is characterized by the existence of a third solution, obtained via the Mountain Pass Lemma, when the boundary data decreases below a suitable threshold. Moreover, we analyze the asymptotic behavior of the associated evolution problem, proving convergence to stable stationary solutions and showing that the Mountain Pass solution is unstable in this dynamical sense.

Geometry-dependent PDEs define operators whose inputs and outputs may each consist of an oriented interface and a function on that interface, whereas most existing neural operators are formulated on fixed domains. In this paper, we first establish an approximation theorem for continuous operators between general state sets. We then instantiate the theorem with the Interface Autoencoder (IAE), which represents interface–function states in a fixed box using finite projection codes and decodes them by zero-set extraction and restriction. For tensor cosine codes, we prove a uniform $C^0$ reconstruction rate and tube-local gradient convergence. These estimates yield Hausdorff and graph-Hausdorff error bounds, together with a reference-based topology certificate. The IAE accommodates both cross-space geometric operators, mapping domain geometries to the associated PDE solution fields, and within-space geometric operators, mapping between interface–function states. Its effectiveness is demonstrated through numerical experiments on the Poisson equation, interface merging in Hele–Shaw flow, and two-phase Stokes flow with surfactant.

We present two new applications of Simmons' point-free Cantor-Bendixson coderivative operator in intuitionistic logic. First, we use it to give a simplified proof of the recent result of Xu and Ye that the free Heyting algebra on two generators does not occur as the Heyting algebra of subterminal objects in any elementary topos. Then we use it to prove that complete Heyting algebra semantics is not strongly complete for intuitionistic second-order propositional logic: semantic consequence from an arbitrary set of assumptions does not coincide with ordinary syntactic consequence.

Aluminum is a reference standard in high-energy-density research and serves as a liner material in megampere Z-pinch facilities. Using mixed deterministic-stochastic finite-temperature density functional theory, we compute the equation of state and transport coefficients of liquid aluminum up to temperatures of 1000 eV and compare the results with model-based approaches. We find that for T<200 eV the density-functional results differ from the models by more than 10% at high density, with a maximum deviation of over 30% (at T = 10 eV), leading to significant discrepancies in the Hugoniot curve at high compression; the calculated electrical and thermal conductivities deviate from model predictions by 26% to 63% at T~100 eV. We further employ the Drude model and the Epperlein-Haines framework to examine the magnetic-field dependence of the conductivities. These results can provide valuable input data for radiation-hydrodynamics codes.

We present a new Plebanski formulation of General Relativity. The key difference is that we work with a gauge vector field derived from the Lanczos potential of the Weyl tensor. We show that, on shell, the gauge vector field encodes all nontrivial gravitational degrees of freedom and, in a suitable Lorentz gauge, Bianchi identities ensure that this gauge vector field satisfies a source-free wave equation for Ricci-flat spacetimes. This description suggests new ways of interpreting the true geometrical degrees of freedom with applications in both classical and quantum gravity.

Background: Artificial intelligence (AI)-based glaucoma detection from colour fundus photographs (CFP) offers scalable screening, but performance may decline on external datasets because of differences in ground-truth definitions, populations, and coexisting conditions such as high myopia (HM). We developed and validated a Vision Transformer-based deep learning (DL) model for glaucoma detection across multi-ethnic cohorts with and without HM. Methods: A ViT-B/16 model with predictive uncertainty estimation was developed using 56,483 CFPs (57.1% with myopia; 14.4% with HM). Glaucoma labels were standardised using clinical, imaging, and perimetry data. The model was validated on 16 independent datasets across three continents, including four datasets with explicit HM labels. Findings: Internal AUROC was 98.7% (95% CI 98.2-99.1%), with sensitivity 94.5% and specificity 97.3%. Across 16 external datasets from eight countries, AUROCs ranged from 86.4% to 99.6%. In HM eyes, internal AUROC was 97.8% (95% CI 96.1-99.2%), with sensitivity 94.8% and specificity 93.7%. External HM AUROCs were 86.5% in the Beijing Eye Study and 93.3%, 91.8%, and 85.5% in hospital-based datasets from Taiwan, Thailand, and South Korea. In an exploratory HM clinical evaluation, the model had higher CFP-only diagnostic accuracy than ophthalmologists and trained graders (92.0% vs 70.0%; p=0.008) and performed comparably to glaucoma specialists using full clinical information. Interpretation: The model showed robust glaucoma detection across myopic and non-myopic multi-ethnic populations and may support AI-assisted screening in settings with high HM prevalence.

Building on recent work of Hom and Wan, we construct additional examples of exotic knot traces, providing partial progress toward their conjecture. Our approach uses the immersed-curve techniques in Heegaard Floer homology developed by Chen and Hanselman.

Measuring how successful a conversation is remains difficult, even for humans judging spoken dialogue. We evaluate state-of-the-art LLMs as pointwise and pairwise judges of conversational success on CANDOR, finding pointwise scoring correlates moderately with human ratings, while pairwise comparison suffers from long transcripts and positional bias. Since this leaves judge scores incomparable across models, we propose a small anchor set and a calibration function that calibrates any judge onto a shared, interpretable scale. We further release the Voice Arena Goal Dataset (VA), 200 task-oriented human-AI and human-agent conversations with pairwise annotations, revealing a substantial gap between current judges and human-level discrimination. Using VA, we test whether CANDOR-fitted calibration transfers to human-AI conversations, finding it brings judges onto a shared scale despite never observing VA during fitting.

Quantum heat transport governs energy exchange processes and statistical laws in non-equilibrium quantum systems, and plays a pivotal role in quantum thermodynamics. We investigate the steady-state thermal transport of a noncommuting coupled spin system. We employ the quantum dressed master equation approach within the framework of open quantum system theory to accurately analyze the non-equilibrium dynamics, ensuring the validity of transport results in the strong coupling regime. Our results demonstrate that noncommuting spin coupling serves as a significant resource for modulating the nonlinearity of the heat current. Specifically, in the weak spin-coupling regime, the system exhibits robust negative differential thermal conductance (NDTC) across various spin numbers. By deriving analytical expressions for the heat current in both the single-spin and large-spin limits, we reveal that this NDTC behavior is governed by microscopic cycle fluxes. Physically, this arises because spin excitation channels induced by the cold reservoir are suppressed under a large temperature bias, thereby blocking energy exchange cycles. Conversely, in the strong spin-coupling and large temperature bias regime, the quantum system demonstrates pronounced thermal rectification. This high rectification efficiency originates from the unidirectional saturation of the heat current, rendering the system a promising candidate for high-performance thermal diodes. Furthermore, we extend the model to a three-terminal configuration to construct a quantum thermal transistor. By manipulating the temperature of the gate reservoir, we achieve efficient modulation and amplification of heat flow between the source and drain. The heat amplification factor is shown to far exceed unity in specific operating regions, confirming significant thermal amplification.

Detectors of alignment failures screen deployed language models and score alignment benchmarks. Most are generative judges that spend a decoding pass on every criterion, and classifiers that read token probabilities, such as Llama Guard, still score one fixed label per call. Jev, a model trained with reinforcement learning for calibrated decisions (RLCD), answers many typed questions about one input with calibrated probabilities in a single call. Whether it detects alignment failures has not been measured. We present RLCDAlignBench, which benchmarks Jev on ten alignment failures: sycophancy, jailbreaks, deception, prompt injection, hallucination, privacy violation, social bias, reward hacking, concealing uncertainty, and power seeking. It spans 44 benchmarks and five target models, labelled by each benchmark's scorer and, on two, by humans. Many of these failures are relational, defined against a reference, such as the user's belief or an injected instruction, that the response alone does not reveal. Our key idea is therefore to vary what Jev is asked separately from what it sees: the question's wording and answer type on one side, the fields of the input on the other. A single generic question reaches a median AUROC of 0.886 zero-shot and beats supervised baselines on most benchmarks. Question wording matters little, while context matters more, mostly through fields that encode the label. Jev matches the reference scorer's agreement with human labels, surfaces label defects in existing benchmarks, and costs 63x less than LLM-judge scorers. Code and data: https://github.com/sumleo/RLCDAlignBench.

Recent advances in speech language models have improved automatic speech recognition (ASR) for long-form audio. However, accurately and consistently transcribing domain-specific terminology remains challenging. Motivated by the world knowledge and contextual capability of large language models (LLMs), we propose Agentic-GER, an LLM-based agent for terminology correction in long-form speech. The agent uses global context from the full transcript to identify suspicious terms and resolve ambiguous hypotheses. It selectively re-transcribes the source speech to check candidate corrections, and uses accepted edits to guide subsequent decisions. Experiments with four LLMs and two ASR systems on GigaSpeechBench show consistent terminology improvements in both Chinese and English, with and without thinking. On Chinese speech, Agentic-GER achieves up to a 36.8% relative reduction in biased character error rate (B-CER) over the Whisper baseline.

We analyze the transport of rod-like particles by diffusion and drift in a three-dimensional channel with varying circular or elliptic cross section. Applying the Fick-Jacobs approximation to the transport equation of the particles' probability distribution, we derive an effective one-dimensional substitute model and the associated free energy profile. Our results show that the data for the mean first passage time of rods, once expressed as a function of the effective free energy barrier, collapse onto the same master curve as obtained for point or spherical particles. The observed universality provides a simple framework for predicting transport times of anisotropic particles in confined geometries without resolving the full multidimensional dynamics.

We construct a smooth Riemannian metric with strictly positive sectional curvature on every smooth homotopy seven-sphere. For each smooth structure, we construct positively curved metrics on two seven-dimensional disks whose induced boundary metrics agree under the prescribed attaching map. Under this identification, a gluing theorem then yields the required smooth metric on the closed manifold. The construction applies to all 28 oriented diffeomorphism classes.