Papers

Topic modeling is an effective technique for discovering hidden themes within documents and is widely used in text mining and data analysis across a variety of industry sectors. Recently, large language model (LLM)-based topic models have been emerged that prompt LLMs to generate topics then assign the topics to documents, producing more natural and human-readable topics than conventional topic modeling algorithms. However, the nature of topic assignment process causes certain drawbacks, such as the incapability to produce topic distributions over a document, too broad or narrow topics, and high resource consumption, which increases with the number and length of of documents being assigned topics. These issues are particularly critical for industrial applications, which require high-quality, in-depth analysis and the processing of large volumes of documents. In this context, this paper introduces a framework called SeLATM, which addresses these concerns by employing segment-level topic generation and topic refinement through agentic feedback loops. Experimental results on various datasets demonstrate that SeLATM significantly reduces the LLM resources compared to methods based on topic assignment process, while maintaining superior performance.

Nonstabilizerness is a necessary resource for quantum systems to lie beyond the classically simulable regime. With the advent of stabilizer $\alpha$-Rényi entropies, nonstabilizerness has also become a many-body diagnostic, complementary to entanglement. While deciding whether an arbitrary quantum state has nonstabilizerness is provably hard, such states already require a description exponentially large in the number of qubits and are thus out of reach for many-body physics. Here we instead consider 2D tensor network (TN) states, which compactly capture states obeying an entanglement area law, and ask whether they admit a simpler algorithm for the same task. In contrast to the 1D case, we prove that, even at small, constant bond dimension, computing the stabilizer entropy of 2D TN states is $\#\mathrm{P}$-hard for any integer $\alpha\ge 2$, and deciding stabilizer membership is $\mathrm{C_=P}$-complete. The corresponding constant-accuracy problems remain $\mathrm{C_= P}$-hard. We further prove that deciding whether a PEPS can be transformed into a stabilizer state by local unitaries over a specified bipartition is $\mathrm{C_= P}$-hard. Under standard complexity assumptions, these results rule out classical or quantum algorithms with polynomial resources for all three tasks in the worst case.

Transformers have become a central architecture for in-context learning (ICL), particularly through their state-of-the-art performance in large language models. This success motivates understanding how transformers exploit intrinsic geometry of real-world data. However, existing nonparametric ICL theory has largely focused on simplified settings, such as Euclidean domains or single manifold models, failing to fully explain its phenomenal empirical success in complex tasks. To address this gap, we study the prediction problem under unknown heterogeneous geometry, modeled by sample size-dependent mixtures of manifolds with heterogeneous dimensions, smoothness, and sampling masses. Under the growing geometric complexity model, we establish a minimax lower bound capturing the aggregate difficulty of the components and construct an oracle tangent local polynomial estimator with a matching upper bound. This estimator is connected to a structure-informed, two-stage softmax transformer with a geometric preconditioner and local polynomial solvers. The transformer achieves negligible approximation error relative to the minimax rate with logarithmic depth and polynomial size. Finally, we derive an in-context generalization bound for near empirical risk minimizers (ERM) over this class. Together, these results identify conditions under which the resulting predictor exploits local geometry and attains minimax rate.

We present a catalog of sources detected above 50 GeV by the Fermi Large Area Telescope (LAT) on board the Fermi Gamma-ray Space Telescope using 16 years of observations. This is the Fourth Catalog of Hard Fermi-LAT Sources (4FHL), containing 673 objects detected in the 50 GeV-2 TeV energy range. The sensitivity and localization are improved by factors of $\sim$2 relative to the previous LAT catalog at the same energies (i.e. the 2FHL catalog). The increased photon statistics, nearly doubling the number of detections relative to 2FHL, enables more detailed population studies, including the identification of fainter sources detected above 50 GeV. The majority of the sources (84.7%) are associated with extragalactic counterparts, primarily BL Lacertae objects, including 36 objects at redshift $z>1$. Galactic sources account for 10% of the sample, mainly pulsar wind nebulae and supernova remnants, while $\sim$5% remain unassociated or of uncertain classification. Twenty sources are reported with no counterpart in previous LAT catalogs. The 4FHL catalog provides the deepest all-sky survey above 50 GeV to date. By focusing on this high-energy regime, it provides a fundamental link between space-based gamma-ray observations and the TeV range probed by current and future ground-based observatories. It serves as a useful reference for guiding future observations, including new potential targets for follow-up with Cherenkov telescopes.

Vision-language models (VLMs) often struggle with compositional reasoning tasks, but the reasons for this underperformance remain unclear. A common hypothesis is that models struggle to integrate multiple components, leading to training interventions to improve compositional binding. However, this assumption has never been directly quantified. Existing benchmarks evaluate captions only in their composed form, making it impossible to separate the cost of joint reasoning from the cost of recognizing individual components under increasing load. We introduce COMPASS (COMPositional Analysis of SkillS), a controlled evaluation framework designed to isolate and measure the distinct factors underlying compositional failure. By comparing performance on composed captions with their decomposed counterparts , we directly quantify the cost of compositional integration across 87K image-caption pairs. Across multiple VLMs, this gap is real but partial, accounting for only part of the observed degradation. This motivates a finer-grained investigation into what additional factors govern model behavior. We analyze performance at the level of individual skills: object detection, attribute binding, and relation reasoning, using skill-targeted perturbations across 274K image-caption pairs. We find a consistent skill-specific pattern: each skill degrades primarily with the count of its own primitive type (self-load), while cross-load effects are predominantly positive, suggesting that primitives of different types provide useful grounding context. This pattern holds across standard contrastive encoders, explicitly trained compositional reasoning models, and non-contrastive architectures. These findings show that compositional degradation reflects multiple separable factors that cannot be reduced to joint reasoning alone.

Early material degradation can produce weak, local changes in elastic stiffness before a crack or void forms, but conventional ultrasonic measurements dilute these changes through path and thickness averaging. This paper introduces DEPTH (Depth-resolved Elastic Property Tomography using Higher-order modes), which uses mode-to-mode scattering among higher-order guided waves to reconstruct elastic-property variations as a function of axial position and depth through a plate. The formulation is developed using shear-horizontal (SH) modes, for which the reconstructed property is shear modulus. A Born formulation shows that each incident-received mode pair contains at most two cosine depth orders, allowing the through-thickness field to be recovered from diverse modal mixtures across mode pairs and frequency. The matrix-free regularised inverse is evaluated using direct two-dimensional modal data, practical single-surface excitation and reception, and full three-dimensional finite-element data. In a 10 mm plate, the ideal two-dimensional case gives a through-thickness point-spread FWHM of 0.84-1.03 mm, while the single-surface configuration recovers four compact targets at the correct depths. This provides the first demonstration of depth-resolved elastic-property tomography from multimode guided-wave transmission with excitation and reception confined to a single physical plate surface. With full three-dimensional propagation and finite-width transduction, the reconstruction retains the compact-target positions with a whole-image correlation of 0.899 against a matched acquisition-limited reconstruction. These simulation results demonstrate that higher-order guided-wave mode conversion can provide local depth-resolved elastic-property tomography from surface-access measurements, with potential for characterising early-stage damage such as fatigue.

Federated learning (FL) data corruption can affect either inputs or labels, but it remains unclear whether input-conditional uncertainty and prediction-label loss expose these corruption modes equally. This paper compares two corruption-detection signals in FL: input-conditional uncertainty and prediction-label loss. The uncertainty signal is characterised using a learned aleatoric variance estimate together with Monte Carlo (MC) dropout variance and entropy measures, while the loss is computed against the supplied label. We test these signals against additive image noise and persistent random label flips. On ResNet-20 with CIFAR-10 and SVHN under Dirichlet partitions with data that are not independent and identically distributed (non-IID), the two corruption types behave differently. For persistent random label flips, the within-client per-sample area under the receiver operating characteristic curve (AUC) is 0.85 on CIFAR-10 and 0.95 on SVHN for prediction-label loss, while every uncertainty estimator stays at chance (0.49–0.50). This pattern is consistent with the model remaining confident in the underlying image despite the supplied label being wrong. For image noise, expected-entropy uncertainty rises above chance (0.67 on CIFAR-10 and 0.66 on SVHN), while loss responds comparably (0.64 on both). Each signal is therefore the stronger detector for a different corruption: the prediction-label loss for persistent label flips, and expected-entropy uncertainty for image noise, with its advantage becoming apparent as federation-wide corruption prevalence increases. Robust FL data-quality assessment should match the signal to the corruption rather than rely on uncertainty alone across corruption types.

We show that there exists a constant $\epsilon > 0$ such that if $C$ is a conjugacy class in $A_n$ of size at least $|A_n|^{1-\epsilon}$, then $A_n \setminus \{1\} \subseteq C^2$. This proves the robust version of Thompson's conjecture for the alternating group, conjectured by Shalev (Annals of Math., 2009). Our proof combines character bounds and combinatorial cancellation techniques with the recent theory of hypercontractivity for functions over symmetric groups.

Cloth manipulation is a challenging task due to the deformable and high-dimensional nature of cloth, which leads to complex interaction dynamics and perceptual ambiguity arising from frequent occlusions of critical visual cues such as folds, edges, and grasp points. In this work, we tackle cloth unfolding using a regrasping-in-the-air strategy, where one manipulator holds the cloth while the other grasps it at an optimally selected point to unfold it. To this end, we propose CeDiRNet-6DoF, a deep learning framework that jointly predicts effective grasp points and the complete 6-DoF grasp pose from the observed cloth configuration. By integrating dense 3D grasp regression with segmentation and sine-cosine-encoded Euler angles, the proposed method reliably estimates the grasp configuration that maximizes the unfolded cloth area. We extensively evaluated CeDiRNet-6DoF on a bimanual robotic setup within the ICRA 2024 Cloth Competition framework, achieving state-of-the-art performance. An ablation study further validates the benefits of key design components, including joint segmentation, background randomization, and image cropping. These results establish CeDiRNet-6DoF as a robust and versatile foundation for reliable robotic cloth manipulation in unstructured environments.

The growing popularity of short videos has driven demand for one-click content creation. Visual templates turn uploaded images into personalized content with preset effects, but reusable template generation still requires substantial manual effort in asset preparation and tool orchestration. We propose TemplateCraft, a multi-agent system that converts natural-language instructions into client-executable templates through planning, material generation, effect-workflow generation, and protocol compilation. Its Planner-Evaluator loop uses execution feedback for targeted rollback, while stage-level and long-term memory support revision without parameter updates. We evaluate TemplateCraft on TemplateBench, derived from 60 real-world templates. With the same Qwen3-VL backbone, TemplateCraft raises image/video generation success rates from 56.7%/30.0% to 66.7%/50.0% over Planner-only (best-of-three) and improves template adherence and style consistency. With additional evaluation and revision, it matches or exceeds a GPT-4o Planner-only baseline on selected metrics. Persistent assets further improve cross-input style consistency.

Medical vision-language model (VLM) post-training is commonly evaluated through answer accuracy. We examine how changes in accuracy and training objectives relate to image-conditioned decisions in a controlled Qwen2.5-VL-3B study on PMC-VQA. We compare supervised fine-tuning (SFT) with low-rank adaptation (LoRA) restricted to the language model, expanded multimodal adaptation scopes, standard answer-only Group Relative Policy Optimization (GRPO), and a counterfactual evidence objective. On 2,000 clean-test questions, language model LoRA SFT changes correct-image accuracy by +1.10 percentage points (95% paired bootstrap CI:-0.85 to +3.05), while visual-benefit events decrease by 2.40 points and image sensitivity decreases by 5.60 points. Paired records reveal 155 acquired and 203 lost visual-benefit events. Broader adaptation yields lower correct-image accuracy than language-model LoRA SFT. Standard GRPO produces mixed-reward groups and parameter updates, with an uncertain clean test accuracy change. A generation audit reveals that canonical option scores can follow a different token path from generated answers. With scores taken along the greedy generation path, the evidence target improves on the training set; its gains over standard GRPO remain inconsistent on validation data at matched training doses. Sample-level analyses trace how evidence scores, decision margins, and generated answers change during post-training. This empirical and measurement audit identifies gaps between optimization activity, target acquisition, and useful held-out visual behavior.

The goal of this paper is to clarify the relation between the Kerler-Lyubashenko TQFT $J_\mathcal{E}$ associated with the adjoint end $\mathcal{E}$ of a (not necessarily semisimple) modular category $\mathcal{C}$ and its renormalized version $V_\mathcal{E}$ based on the modified trace supported by the ideal $\mathrm{Proj}(\mathcal{C})$ of projective objects of $\mathcal{C}$. More precisely, we construct a $3$-dimensional ETQFT $\boldsymbol{\hat{A}}_\mathcal{E}$ that contains both $J_\mathcal{E}$ and $V_\mathcal{E}$. The ETQFT $\boldsymbol{\hat{A}}_\mathcal{E}$ is given by a $2$-functor whose source is the $2$-category of admissible $3$-dimensional cobordisms, and whose target is the $2$-category of finitely complete linear categories. We improve on earlier versions of the construction by dropping the admissibility condition for surfaces. By doing so, we obtain an ETQFT whose circle category $\boldsymbol{\hat{A}}_\mathcal{E}(\boldsymbol{S}^1)$ is equivalent to the category $\mathcal{C}$, as opposed to the ideal $\mathrm{Proj}(\mathcal{C})$.

Clinical prediction models estimate risk from patient measurements, while large language models support medical text understanding and question answering. Yet their language capabilities do not ensure accurate prediction from structured, high-dimensional clinical time series. Improving this ability would connect risk estimation with flexible questions about a patient's evolving condition. We introduce ViSTA, a compact adapter that incorporates irregular numerical measurements into a pretrained vision-language model's chart representations. It learns corrections to visual tokens while leaving all pretrained parameters unchanged. On MIMIC-IV, ViSTA has the highest mean scores among the compared adaptations on all four metrics for acute kidney injury and mortality prediction across models with 2-9 billion parameters. With 0.516 million trainable parameters, the 2-billion-parameter model reaches an area under the ROC curve of 0.7376 for acute kidney injury, compared with GPT-5.6 Sol's 0.7380 with text input and high reasoning effort. Training for temporal question answering yields 69.27% accuracy at 4 billion parameters with over 90% fewer trainable parameters than low-rank adaptation using charts or numerical text, at a 2.82-4.88 percentage-point accuracy gap. ViSTA extends pretrained language models to numerical prediction and temporal questions.

This paper studies Veech groups of dilation surfaces whose fundamental group is not finitely generated. We prove that if $ S$ is a surface with self-similar end space and every end is accumulated by genus, then every countable subgroup of $SL(2,\mathbb{R}) $ can be realized as the Veech group of a dilation surface homeomorphic to $ S$. Additionally, we can construct this dilation surface such that it realizes horizon saddle connections in a prescribed set of directions. This contrasts with the case of closed dilation surfaces, where horizon saddle connections impose restrictions on the algebraic structure of the Veech group.

Multidisciplinary Design Optimization (MDO) enables the application of optimization algorithms to complex engineered systems, ranging from aerospace and automotive to robotics and microelectronics. Such multi-component systems typically consist of coupled subsystems, each characterized by its own design variables, constraints and objectives. Without adequate coordination of these couplings, subsystems may be optimal in isolation while the overall system remains suboptimal or even infeasible. Distributed design optimization coordinates coupled subsystem optimization problems while allowing them to retain control over their local design variables. Although promising coordination methods exist, their application is hindered by the lack of a comprehensive software framework which supports intuitive problem definition, provides suitable distributed coordination algorithms, is modular and extensible, enables interactive (post-)processing, and supports the distributed computation of subsystem executions. This work derives requirements for such software, assesses existing frameworks against them, and introduces DistributedDesignOptimizer, an open-source Python framework to fill this gap.

We build on the recent development that toposes with enough points can be represented as categories of points equipped with ultraconvergence data (a.k.a. virtual ultracategories), characterizing usual classes of geometric morphisms — surjections, embeddings, hyperconnected maps, and localic maps — in terms of their points. In particular, the characterisation of geometric surjections relies on a notion of "virtual ultraretractions", which generalises usual retractions and filtered colimits, categorifies Johnstone's notion of subclosure for topological spaces, and provides a new characterisation of separating classes of points of a topos.

We study the gravitational action in the teleparallel equivalent of general relativity, where the spin connection contributes to the action through a surface term and, when chosen appropriately, regularizes the action by effectively playing the role of a counterterm. We resolve several apparent ambiguities in this procedure using the Schwarzschild spacetime as an example. We show that the apparent discrepancy between bulk and quasilocal actions originates from contributions associated with spacetime singularities, and that the quasilocal action agrees with the standard results obtained in general relativity. We further clarify the role of the tetrad choice in determining the corresponding spin connection, showing that the proper diagonal and canonical frames give the same Euclidean action, while only the proper diagonal frame reproduces the expected Lorentzian action on the Wheeler–DeWitt patch. Finally, we establish the uniqueness of the regularization scheme and show that imposing an appropriate fall-off behavior of the torsion tensor removes the remaining ambiguity and fixes the Euclidean action uniquely.

Low-order refined (LOR) preconditioning replaces a high-order operator with a low-order discretisation on a refined nodal mesh. For tensor-product elements, the two operators are spectrally equivalent with bounds independent of the polynomial order $P$, but the construction does not extend directly to simplex and mixed-element discretisations. This work makes two contributions: it extends LOR preconditioning to simplex and mixed-element discretisations, including triangular, tetrahedral, and prismatic elements, and establishes a generalised Vandermonde transformation linking the modal and nodal LOR formulations, showing that the resulting preconditioned spectra and Krylov convergence are independent of the high-order basis. Numerical experiments show controlled iteration growth on triangular meshes despite increasing condition number, and controlled iteration counts up to $P=5$ on tetrahedral, prismatic, and mixed-element meshes. A single algebraic multigrid V-cycle per outer iteration gives the best balance of iteration count and cost. The method is applied to a production incompressible Navier-Stokes simulation of a race-car front-wing and wheel configuration, discretised on a mesh of $2.87\times10^6$ mixed prismatic and tetrahedral elements giving $32.2\times10^6$ pressure degrees of freedom at $P=3$. LOR reduces the mean pressure conjugate gradient (CG) iteration count from $235.3$ to $5.5$, and the pressure-solve time over 1000 timesteps by 16.1%, relative to the default production static-condensation diagonal preconditioner in Nektar++. The one-time cost of constructing the LOR preconditioner is amortised over the production simulation.

We study recursive-design wild bootstrap inference for dynamic panel data models with unobserved common factors estimated by Common Correlated Effects. In the large N,T setting, the bootstrap reproduces the biased limiting distribution in pure autoregressive models, but fails to capture all bias and factor-estimation variance components in models with additional regressors, particularly under weak exogeneity. We trace this failure to holding regressors fixed across bootstrap replications. We propose to combine bootstrap procedure with available bias-correction methods to conduct adjusted inference. Monte Carlo evidence shows substantial improvements over conventional strategies of using bias-correction paired with cross-sectional bootstrap methods.

We study the order dimension of Cartesian products of finite posets. Kelly and Trotter conjectured in 1982 that $\dim(P\times Q)\ge\dim P+\dim Q-2$ for all finite posets $P$ and $Q$. For $m\ge3$, let $R_m$ denote the incidence poset of the complete graph on $m$ vertices. We prove that there is a constant $C$ such that, for all sufficiently large $m$, $\dim(R_m\times R_m)\le\left(1+\frac{2}{\log_2 6}\right)\dim R_m+C$. Since $1+2/\log_2 6<2$, this disproves the Kelly–Trotter conjecture and shows that $2\dim R_m-2-\dim(R_m\times R_m)$ can grow linearly with $\dim R_m$. For every integer $d\ge8$, we construct an incidence poset $Q_d$ such that $\dim Q_d=d$ and $Q_d$ has the $(3,d)$-covering property. Consequently, $\dim(P\times Q_d)\le\dim P+d-3$ for every poset $P$ with $\dim P\ge3$. Thus, for every integer $d\ge8$, the poset $Q_d$ violates the Kelly–Trotter conjecture with every poset of dimension at least $3$. Finally, we prove that a poset $Q$ has an $(r,s)$-covering if and only if $\dim(Q\times C^r)\le s$ for every finite chain $C$ with at least two elements.

Developing plastic scintillators that simultaneously achieve high light yield (LY), ultrafast timing, and scalable processability remains a formidable challenge in radiation detection. Here, we report a solution-processable hybrid scintillator that integrates high-Z sensitization with Purcell-enhanced radiative recombination within a conjugated-polymer platform. By embedding HfO2 nanoparticles and resonant Ag-SiO2 plasmonic nanoantennas into a poly(9,9-dioctylfluorene) matrix, we exploit the intrinsic sub-nanosecond kinetics of conjugated polymers while overcoming their characteristic low stopping power. The HfO2 nanocrystals amplify energy deposition and suppress interchain aggregation, preserving high photoluminescence quantum yields in the solid state. Simultaneously, the nanoantennas induce strong plasmon-exciton coupling, accelerating radiative decay to ~150 ps. This synergistic architecture yields LY=6.1 photons/keV, comparable to standard commercial plastics with nanosecond lifetimes, yet over an order of magnitude higher than current sub-nanosecond state-of-the-art materials. This performance represents one of the most favourable combinations of efficiency and timing resolution reported to date. Our versatile strategy provides a framework for next-generation organic scintillators, merging record-breaking temporal response with high efficiency for advanced radiation-sensing architectures.

Policy learning and compliant control offer a promising route to reliable autonomous assembly under pose errors and contact uncertainty. However, combining them does not ensure coordination: the policy may continue pushing against contact while the controller yields, producing sustained loading with limited progress. To address this problem, we propose LeCo (Leverage Compliance), a policy-admittance learning framework that guides a visual policy through execution-time interaction under fixed admittance. A multirate feedback mechanism aggregates high-rate contact-interaction records into policy-transition rewards. An integrated conflict cost then characterizes sustained policy-loading/controller-unloading opposition, while a directional high-force tail cost captures continued-loading events within a transition. Together with task completion, these costs encourage the policy to leverage compliance with less unproductive loading. We evaluate LeCo on four real connector-assembly tasks, obtaining an aggregate success rate of 94%. Across tasks, mean successful-trial resultant-force and torque peaks decrease by approximately 30% and 64% relative to the comparison baseline. Reward ablation further shows that adding conflict shaping reduces median successful-trial contact-conditioned conflict density by approximately 53%. These results support learning to leverage fixed compliance by turning multirate policy-admittance interaction into complementary reward signals for effective, lower-load insertion.

The two-sphere partition function of timelike Liouville theory admits two complementary semi-classical descriptions whose relation presents a non-perturbative puzzle. A direct path-integral expansion around the round two-sphere saddle reproduces the perturbative small-$\beta$ expansion, with $\beta$ the Liouville coupling, of the analytically continued Dorn-Otto-Zamolodchikov-Zamolodchikov (DOZZ) formula, while the latter contains an additional oscillatory factor suggestive of a second saddle contribution. We investigate the origin of this factor in the Liouville zero-mode sector using Picard-Lefschetz theory. Taking as input the Hankel-type contour motivated by the inverse-Gamma representation of the zero-mode integral, we determine its Lefschetz-thimble decomposition depending on $\beta$. For $0<\beta<1$, it precisely reproduces the saddle structure encoded in the analytically continued DOZZ result. We identify the integration contour as the source of the difference between the zero-mode saddle descriptions. As $\beta$ approaches $1$, the zero-mode saddle moves to the boundary of field space, while for $\beta>1$ the same contour is represented by a single contributing thimble. We further relate this Stokes reorganisation to recent probabilistic constructions of timelike Liouville theory.

Finite-range hopping constrains both the number of Dirac points and the masses that gap them. We determine the maximal absolute Chern numbers of an isotropic finite-range extension of the Qi–Wu–Zhang model on a square lattice through thirteenth-neighbor hopping. Algebraic zero counting and mass-weighted sum rules yield global bounds, and we construct explicit Hamiltonians that saturate them. We find $|C|_{\max}=36,40,51$ for eleventh-, twelfth-, and thirteenth-neighbor hopping. Equal numbers of Dirac points can produce different maxima because finite-range mass harmonics distinguish opposite vortex charges differently. Twice the squared hopping radius bounds every maximum through twelfth-neighbor hopping and is saturated at the eleventh and twelfth, but the thirteenth-neighbor maximum exceeds it. Thus finite-range hopping constrains the maximal Chern number in two complementary ways: through the number of Dirac vortices it can generate and through the mass-sign patterns it can assign to them. This geometric viewpoint provides a design principle for high-Chern bands and a route to analogous bounds in moiré and other lattice systems.

The time-resolution function is one of the key factors determining the accuracy of neutron-resonance parameter reconstruction in time-of-flight experiments. A method is proposed for determining the time-resolution function of the RADEX neutron source associated with neutron slowing down in the target, based on detailed Geant4 modeling. The neutron time spectra obtained in the Geant4 simulations are approximated by a sum of three Gamma functions with dimensionless parameters. This model is shown to describe the shape of the time distributions well over the neutron-energy range 50–300 eV. To test the applicability of the proposed approach to a source of a different design, a comparison is made with the calculated n\_TOF time-resolution function. Its shape is also shown to be well described by a three-component Gamma model.

With the advent of snapshot cameras, hyperspectral video is becoming more readily available. In recent years, new applications have emerged which have led to increasingly larger datasets. However, hyperspectral video compression remains in the early stages. In this study, we explore the use of implicit neural representation as a candidate solution. We propose a novel extension of an existing RGB video compression model, achieving Bjøntegaard Delta PSNR gains of +4.99 dB and Bjøntegaard Delta rate of -88.88% compared to traditional hyperspectral image compression methods applied frame-by-frame. In addition to reconstruction quality, the effects on downstream task performance are measured in the form of object tracking success. Compared to video compressed with methods based on principal component analysis and JPEG2000 in low data regimes, our proposed method improves tracking area under the curve by up to 23.42% and distance precision by up to 35.56% on examples from the HOT2026 dataset.

Task-agnostic assistive exoskeleton control based on human intention offers greater flexibility than conventional approaches that rely on predefined tasks or motion patterns. Human joint torque estimation enables task-agnostic assistance by characterizing user actions. Physics-consistent methods such as Deep Lagrangian Networks (DeLaN) have been applied to estimate the human torques in multi-user settings, but existing approaches cannot adapt to a specific user without retraining, and do not account for intermittent contacts during locomotion. We propose ExoLaN, a Context-Aware DeLaN for human-exoskeleton interaction that learns the full coupled system dynamics while adapting to changes in interaction context. ExoLaN combines temporal context with partial contact-force measurements from force-sensitive insoles to infer latent dynamics embeddings and estimate generalized contact torques. On seven unseen users performing 21 unseen tasks, ExoLaN reduces torque estimation MSE by 7% compared to a black-box baseline. Beyond inverse dynamics, ExoLaN serves as a unified model that also enables accurate forward prediction: training with a multi-step prediction loss reduces acceleration MSE by 59% and long-horizon position and velocity errors by 60% and 93%, respectively, compared with a single-step loss. Moreover, the learned latent context captures task information without explicit task labels, making it a promising signal for task-aware assistive control.

Two-atom superradiance is studied for atoms having a J=0 ground state and a J=1 excited state. The atoms lie on the x-axis and are excited by a \sigma_{+}-polarized laser pulse propagating in the z-direction. We use two, complementary approaches to obtain the final angular momentum in the field, a source-field approach and a Schrodinger picture approach. Both approaches give identical answers, but it is necessary to include terms in both calculations that, at first glance, might not be expected to contribute. We show that, owing to an exchange between a state in which atom 1 is in its m=1 excited state and atom 2 is in its ground state to one in which atom 2 is in its m=-1 excited state and atom 1 in its ground state, the final angular momentum in the field is not necessarily equal to the initial angular momentum in the two-atom system. The missing angular momentum appears as orbital angular momentum of the two atoms.

Recently, the issue of sound quality inside vehicles has attracted interest from both researchers and industry alike due to health concerns and also to increase the appeal of vehicles to consumers. This work extends the analysis of interior acoustic noise inside a vehicle under several conditions by comparing measured power levels and two different models for acoustic noise, namely the Gaussian and the alpha-stable distributions. Noise samples were collected in a scenario with real traffic patterns using a measurement setup composed of a Raspberry Pi Board and a microphone strategically positioned. The analysis of the acquired data shows that the observed noise levels are higher when traffic conditions are good. Additionally, the interior noise presented considerable impulsiveness, which tends to be more severe when traffic is slower. Finally, our results suggest that noise sources related to the vehicle itself and its movement are the most relevant ones in the composition of the interior acoustic noise.

Accurate segmentation of axons in 3D microscopy data is important for analyzing white-matter organization, but dense ground truth labels are expensive to obtain. Existing supervised axon segmentation methods rely on target-domain annotations and can be brittle when tissue type, species, modality, or acquisition conditions change. We present AxonSynth, a domain-randomized synthetic-data framework for training 3D axon segmentation models without manually annotated real training volumes. AxonSynth generates dense synthetic axon labels with orientation priors that reflect realistic fiber configurations and renders them with randomized density, contrast, bias fields, blur, and noise. A three-class 3D U-Net is trained to predict background, axon sheath and intra-axonal space. We evaluate zero-shot transfer on 10 held-out light-sheet microscopy (LSM) patches from macaque and human brain samples labeled with one of three axonal markers, comparing against calibrated thresholding and Frangi filtering using overlap, corrected detection, false-positive, and topology metrics. On macaque samples, AxonSynth achieved the best corrected Dice and corrected precision (0.826 and 0.851), compared with 0.765 and 0.754 for thresholding and 0.685 and 0.762 for Frangi. On human samples, corrected Dice was comparable to thresholding (0.857 vs. 0.868), while component-count error decreased from 22,504 to 3,377. Across all held-out patches, AxonSynth reduced component-count error in 10/10 patches and Euler-characteristic error in 8/10. These results show that synthetic-label domain randomization can reduce dependence on manual axon annotation while supporting synthetic-to-real 3D segmentation.