Papers

When a tool-using agent's write times out or returns a server error, the action may already have taken effect. Retrying blindly duplicates it – a second charge, a second announcement, a second deployment – while giving up skips required work. We ask where exactly-once behaviour should be enforced: in the model, in the agent harness, or in the tool contract. We introduce LIMBO, a deterministic sandbox of six services with realistic contracts (optional idempotency keys, eventually consistent and missing read paths) and twelve fault modes injected at the service boundary, including late commits, redelivery and partial batches; every episode is graded against a ledger of committed effects. Across 25,930 episodes spanning nine recent models, three production agent harnesses, two contract variants and fifteen recovery conditions, the answer depends on the fault. When an immediate read-back can reveal what happened, the model decides: frontier models instructed to act exactly once almost never duplicate a write whose acknowledgement was lost (0.5%), weaker models often do, and the model explains 53% of the explained variance. When it cannot – the request is still in flight, or the transport delivered it twice – the same frontier models duplicate in 56% and 74% of episodes, and the contract explains 81%. We prove that no verification-only policy is exactly-once under late commits without a bound on in-flight time. Waiting works when such a bound is short and known, but with heavy-tailed in-flight delays even an hour of waiting per episode falls short of offering an idempotency key on every write, which lowers the duplicate rate from 28% to 4% because agents use keys when they exist. The harness barely matters, a guard that attaches keys transfers across harnesses unchanged, and agents reported success in 90% of the episodes in which they had duplicated an effect.

The D-region ionosphere is sluggish in nature while responding to the incoming solar radiation. We study the altitude (h) profile of mid-latitude D-region response time delay (Delta t) during three chosen solar flares, namely, C5.2, M5.2, and X2.2 classes. We solve "electron continuity equation" numerically at each and every h to obtain Delta t-h profile. We investigate the 1) latitudinal variation (over both Northern and Southern hemispheres) and 2) seasonal variation (throughout the year) of Delta t-h profiles of each of these solar flares separately. We go over noteworthy variations of Delta t-h profile for both 1) and 2) with reasonably different results over different hemispheres. Also, we study some contrasting behaviours of Delta t-h profile of X2.2 class for both 1) and 2) in comparison to C5.2 or M5.2 classes. We conclude the outcome with possible explanations.

Continuum manipulators provide dexterous motion in confined spaces, but structural compliance, hysteresis, and load-dependent deformation leave residual position and orientation errors that can undermine reliable contact with rigid grippers. To address this limitation, this paper presents a lightweight support-enhanced granular-jamming gripper tailored to a continuum manipulator. The gripper maintains compliance before jamming while establishing a direct load path to the continuum manipulator tip after jamming. To improve its grasping performance, we systematically designed membrane materials, particles, filling ratios, and the internal support structure, and further identify geometry-dependent grasp boundaries with respect to contact offset and object shape. Building on these results, we construct a physical manipulation system integrating the continuum manipulator, granular-jamming gripper, visual feedback, tendon actuation, and pneumatic control. We then train a reinforcement-learning-based reaching controller in a randomized simulation and deploy it on the physical system, demonstrating how positioning control and contact level mechanical adaptation can complement each other in a modular grasp-and-release task. By introducing an adaptive structure that relaxes the need for highly accurate modeling and positioning control, this work explores a design paradigm that integrates physical and embodied intelligence.

Vision-based legged locomotion methods assume clean depth at training time and rely on hand-tuned post-processing filters at deployment. However, filter parameters are rarely disclosed, hindering reproducibility, and performance degrades substantially when depth noise is left unaddressed. Building noise robustness directly into the learning pipeline would eliminate this dependency. While such robustness has been explored for proprioceptive inputs, analogous approaches for depth perception remain largely absent in legged locomotion. We propose DAWN (Denoising and Alignment in World models for Noise-robustness), a noise-robust perception framework for legged locomotion, which builds noise robustness directly into a world model via two modifications: (1) feeding noisy depth to the encoder while keeping clean depth as the reconstruction target, forcing the model to implicitly denoise its input; and (2) applying contrastive learning to align the latent states of noisy and clean depth. Importantly, DAWN is not tied to a specific noise model, requiring no manual tuning to the noise distribution at deployment. Furthermore, it incurs no additional inference cost over existing world model-based methods. Without any manual filter calibration – relying solely on the learned noise-robust representation – DAWN achieves zero-shot quadruped parkour on a Unitree Go1: traversing stairs up to 18 cm, clearing gaps up to 70 cm, and mounting steps up to 45 cm from raw depth observations. Ablation studies show that denoising and contrastive alignment contribute at complementary levels – reconstruction and representation, respectively – and yield additive gains when combined. Videos and code are available at: https://dawn-parkour.github.io/

Recent advances in agentic systems have substantially enhanced the long-horizon capability of embodied manipulation. However, many existing frameworks still follow a passive execution paradigm, which limits their applicability to real-world scenarios involving textual semantic cues, distractors, and initially invisible targets. To bridge this gap, we propose an agent-based active exploration framework that enables robots to dynamically interact with the environment rather than merely execute predefined instructions. Specifically, our framework consists of three collaborative modules: a planning module for high-level task reasoning, a perception module for visual scene understanding, and an execution module for low-level manipulation. This design allows the robot to actively acquire task-relevant information, adapt its behavior based on environmental feedback, and complete manipulation tasks under partial observability. Furthermore, we introduce a fine-grained perception-execution interleaving strategy, which tightly couples visual feedback with skill execution to improve exploration robustness. We evaluate our method on a realistic Find-and-Place task, demonstrating its effectiveness in challenging environments where target objects must be actively discovered before manipulation.

We replicate the distributional-semantics extractive summarisation method of Mohd, Jan and Shah (2020) and adapt it to Hindi, substituting a Devanagari-appropriate component at every language-specific step. The system is evaluated on two independent corpora — the Hindi portion of XL-Sum and FIRE ILSUM 2.0 Hindi — under a Devanagari-aware ROUGE implementation validated against the XL-Sum authors' own multilingual scorer, with all comparisons drawn as 1000-resample paired bootstraps. In its published equal-weight configuration the replicated system is significantly worse than a three-sentence lead baseline on both corpora, trailing Lead-3 by 0.042 ROUGE-1 Fon XL-Sum and by 0.265 on ILSUM. A feature ablation shows that sentenceposition is the only feature that contributes: position alone reproduces the lead baseline exactly, removing position gives the weakest configuration,and a validation-tuned weighting can at best equal Lead-3 and never exceed it. TextRank fails identically, making this a class-level rather than an implementation-level result. A selection analysis shows the remaining features steer extraction towards long, entity-dense body sentences while the references reuse the article lead.Current Hindi benchmarks therefore cannot reward non-lead content selection, motivating purpose-built evaluation resources.

Circularly polarized light can rectify orbital motion into a static magnetization through the inverse Faraday effect (IFE), but in electron-hole bilayers the electron and hole contributions cancel exactly when their properties are equivalent. We show that electron-hole bilayers in transition-metal dichalcogenide platforms avoid this cancellation through effective-mass asymmetry alone, and that the surviving orbital IFE is sensitive to interlayer coupling. In the weak-coupling regime, a two-component Drude description yields an induced magnetization of order one Bohr magneton per carrier for representative terahertz driving. In the strong-coupling regime, where the carriers bind into interlayer excitons, we treat the relative motion using a hydrogenic model with a Rytova-Keldysh interaction and obtain a reduced but finite response. We give closed-form expressions that allow estimates across a broad parameter range and identify where the response is largest.

The winding number is a simple topological invariant. In the case of chiral symmetry it characterises gapped phases of Fermions. We study statistical properties of this topological index or invariant in a chiral symplectic setting using Random Matrix Theory. We consider ensembles of Hamilton matrices in the symmetry class CII (quaternionic matrices) according to the classification in the tenfold way. We set up a parametric random matrix model and derive expressions for parametric correlations of the winding number density as well as for the discrete winding number distribution. We found a super-universality in the limit of large matrix dimensions for the one- and two-point correlators for the bulk of parameters, meaning that the results agree up to rescaling with those of the class AIII (complex matrices). In this context we employ a new method of unfolding and discover the Gaussian behaviour of the winding number distribution.

While deep learning accelerates expensive partial differential equation solving in computational fluid dynamics (CFD), existing methods like PINNs and FNOs often struggle with generalization, noise robustness, and physical consistency. We introduce the Transformer-Mamba for Flow Field (TM4FF) framework, a physics-constrained operator learning model with three key innovations: a Residual Wavelet Mamba (RWM) layer for feature denoising, a Transformer-based attention mechanism for enhanced feature fusion, and a physics-informed loss using Fourier derivatives to enforce the Navier-Stokes equations. Experiments on four CFD datasets show TM4FF achieves high accuracy and robust generalization across varying flow conditions.

Low-pass (LP) filtering is a fundamental operation in graph signal processing (GSP). Among finite-order nodal-domain methods, Lanczos-based filtering provides more accurate approximations of ideal LP filters than Chebyshev polynomial methods. We show that the approximation of Lanczos filtering can be further improved through algorithm unrolling and data-driven parameter learning. The key insight is that, because ideal LP filtering is a projection operation into the low-frequency eigen-subspace $\cS_K$, instead of approximating individual eigen-pairs of a graph Laplacian $\L$ as done in classical Lanczos, an unrolled Lanczos network can directly approximate $\cS_K$. Specifically, we first establish a theorem identifying properties of the Lanczos tridiagonal matrix $\T_m$ that promote accurate approximation of the low-frequency eigen-subspace $\cS_K$. Guided by this theory, we relax the orthogonality constraint on Lanczos vectors, resulting in Ritz vectors that better span $\cS_K$. To ensure numerical stability, we constrain $\T_m$ to be similar to a symmetric matrix, thereby guaranteeing real-valued eigenvalues. Experimental results on random graphs and learned graphs in two natural language processing (NLP) tasks show that our unrolled Lanczos network achieves superior ideal LP filter approximation compared to the classical Lanczos method.

The main objective of this paper is to investigate orthonormal Strichartz estimates for Schrödinger equation on the Wiener amalgam space $\mathcal{W}(\mathcal{F} L^p, L^q)$. More precisely, we first examine improvements in the time-integrability exponent of the existing Strichartz estimates established by Cordero and Nicola in \cite{NFC} for Schrödinger equations associated with $\mathcal{H}^{+}=-\frac{1}{4\pi}\Delta+|x|^2$. We then extend these improved Strichartz estimates from a single initial datum to systems of orthonormal families, providing, to the best of our knowledge, the first results in this direction in the setting of Wiener amalgam spaces. We further extend the classical Strichartz estimates in Wiener amalgam spaces from a single initial datum to systems of orthonormal families of initial data for the Schrödinger equation associated with the operator $\mathcal{H}^{-}=-\frac{1}{4\pi}\Delta-|x|^2$ and Hamiltonian operator of the form $\mathcal{H}_{\mathcal{A}} = -\frac{1}{4\pi} B \nabla \cdot \nabla$, where $\mathcal{A} = \begin{pmatrix} 0 & B \\ 0 & 0 \end{pmatrix} \in \mathrm{Sp}(d,\mathbb{R})$ with $B = B^*$ and $\det B \neq0$. A key ingredient of our approach is Stein's complex interpolation theory for Wiener amalgam spaces, combined with a duality argument inspired by the work of Frank and Sabin. As an application of these orthonormal estimates associated with $\mathcal{H}^{+}, \mathcal{H}^{-}$, and $H_{\mathcal{A}},$ we establish local and small-data global well-posedness for the Hartree equation with infinitely many particles, for non-trace-class initial data.

Developing optimized machine-learning algorithms for edge devices with limited computational and memory resources is challenging, time-consuming, and highly dependent on device-specific constraints. In this work, we streamline an edge ML workflow to enable rapid development, optimization, and deployment of machine-learning (ML) models directly on the WeBe Band, a wrist-worn wearable device designed for multimodal physiological data monitoring. The proposed system automatically generates hardware-efficient ML models that can be easily integrated into the WeBe core firmware, supporting AutoML, hardware-aware quantization, and performance profiling to build models that meet desired latency targets while remaining compatible with device memory and power limitations. The proposed framework tightly integrates the open-source Piccolo AI ecosystem with an automated pipeline that generates deployable firmware artifacts, performs hardware-aware model compilation, and supports over-the-air (OTA) deployment. The system supports multiple lightweight model classes, including classical machine-learning algorithms and neural networks, and provides built-in on-device profiling tools to evaluate inference latency and memory footprint under realistic execution conditions. Experimental results demonstrate clear trade-offs between model complexity and deployability on a microcontroller, showing that classical models offer strong real-time performance while lightweight neural networks require careful resource management. Rather than proposing new learning architectures, the current work mainly focuses on system-level automation, deployability, and enabling researchers and developers to rapidly iterate on models and evaluate them directly on target hardware. Although demonstrated on the WeBe Band platform, the workflow is designed to be extensible to other ML-powered edge devices.

Physical Internet (PI) is an emerging concept that applies the digital internet as a design metaphor for the development of sustainable, interoperable, and collaborative freight transportation. It is considered a way to bring logistics into the next generation of transformation. Effective tools for organizing and integrating knowledge are essential for navigating the emerging research in this area. In this study, we explore the application of Graph Retrieval Augmented Generation (GraphRAG) in the context of PI, using GPT-4o mini and Neo4j to construct a knowledge graph and systematically analyze existing PI-related literature. Our approach synthesizes scattered research findings, highlights emerging trends, and identifies knowledge gaps. Furthermore, we demonstrate that GraphRAG improves accessibility by structuring complex information into interconnected graphs and provides a deeper understanding of underlying research dynamics. This research will contribute to future research and innovation by providing a new method of information analysis in the PI domain.

We explore a nonequilibrium Bose-Einstein condensate (BEC) in a bosonic Kitaev chain. Within a mean-field framework, we show that bosonic superfluid order develops once the pairing strength exceeds a critical threshold. The condensate order parameter adopts an alternating phase between even and odd sites, giving rise to a spontaneous sublattice ordering. By mapping the emergent sublattice degrees of freedom onto a pseudospin, the elementary Bogoliubov excitation is described by a non-Hermitian Dirac Hamiltonian, displaying two prominent features: a re-entrant bosonic Andreev bound state localized around boundaries, and a macroscopic degeneracy of exceptional Majorana bosons at the critical point.

This paper develops a compact fourth-order positivity-preserving active flux (AF) method for the one- and two-dimensional compressible Navier–Stokes equations on Cartesian meshes. The method retains the cell averages and shared point values of the standard third-order AF method as its degrees of freedom. To avoid the order reduction that can arise when diffusion is discretized using operators from the standard third-order AF method, while maintaining compactness, the divergence of the viscous flux is discretized directly using compact fourth-order operators. For the inviscid part, incorporating a downwind point value into the biased stencil yields fourth-order accuracy. A monolithic flux limiting blends high-order total numerical fluxes with low-order positivity-preserving counterparts, treating the inviscid and viscous fluxes jointly while maintaining local conservation. Together with a scaling limiter for point values, this procedure preserves density and pressure positivity for both cell averages and point values. Numerical experiments demonstrate fourth-order convergence, positivity preservation, and accurate resolution of shocks and viscous flow structures. For the two-dimensional viscous shock tube, a comparison with a discontinuous Galerkin method shows improved computational efficiency in resolving complex interaction of shock waves and boundary layers.

In the last two decades, last-mile delivery (LMD) firms have seen immense growth fueled by the success of e-commerce, leading to faster and cheaper deliveries. Operating on thin margins, LMD firms strive for successful first-time deliveries to avoid the financial and reputational costs of reattempts. Delivery Agents (DAs) are integral to LMD efficiency, influencing customer experience, delivery success, and productivity. However, most LMD performance enhancement research focuses on process, technology, and incentives, which presume workers will conform to procedures and monitoring tools will function flawlessly. Nevertheless, in practice, DAs deviate from expected behaviors, i.e., indulge in misconduct, negatively affecting delivery efficiency, often resulting in returned parcels. One of the major misconducts is fake remarked deliveries, wherein DAs intentionally do not deliver the parcels and provide a fake reason for it. For instance, even without reaching a delivery address, a DA remarks 'customer unavailable' and records a delivery failure. In this study, we collaborated with a leading Indian LMD firm and, using instrumental variable regression, find that such misconduct leads to a spillover productivity loss. This effect reduces the next day's successful deliveries by 1.60% and first-time-right deliveries by 1.86%. We discuss misconduct's correlation with factors such as task complexity and offer novel insights into how opportunistic circumstances can influence worker behavior.

Polycrystalline SrTiO3 is employed as a model system to investigate microstructural evolution under applied electric currents. Under a substantial current density, well-aligned, elongated abnormal grains develop near the anode following a flash event, in contrast to previously reported cathode-side enhanced grain growth under negligible currents. The equivalent diameter of the abnormal grains increases linearly with time, deviating from classical parabolic grain growth kinetics. The applied current drives elemental redistribution near the anode, producing a Ti-rich region adjacent to a Sr-rich belt that migrates toward the cathode, from which the abnormal grains nucleate. Aberration-corrected scanning transmission electron microscopy and electron energy-loss spectroscopy reveal that the fast-moving grain boundaries (GBs) within the Ti-rich bulk region are Sr-enriched, O-depleted, and Ti-reduced. An analysis based on the Brouwer diagram suggests the formation of p-i-n regions under the applied electric field. Conversion between electronic and ionic currents at the p-i and i-n junctions, field-driven precipitation and dissolution of the Sr-rich Ruddlesden-Popper phase, and field-driven migration of Sr and O vacancies collectively explain the elemental redistribution and redox-modulated migration of the Sr-rich belt. Incomplete redox reactions at the moving junctions create the moving Sr-rich belt and generate a locally reducing environment, consequently producing fast-moving, Sr-rich, reduced GBs. These findings reveal new mechanisms of electric current-driven defect-mediated microstructural evolution.

Pulsar timing array (PTA) collaborations have reported evidence for a stochastic gravitational wave (GW) background in the nHz band. Should scalar-induced GWs, sourced at second order by enhanced primordial curvature perturbations, contribute to this signal, a coincidence of scales makes them directly sensitive to the softening of the equation of state around the QCD crossover. We solve the tensor and scalar equations of motion across the Standard Model (SM) thermal history and tabulate the transfer functions for direct use in present and future PTA analyses. We show that this SM effect modifies the height of the induced spectrum by up to $\approx55\%$ across the PTA band relative to the radiation-domination expectation, with either sign depending on whether the source modes cross the horizon before or after the crossover. Fitting the NANOGrav 15-year data with a broken-power-law curvature power spectrum, we find that including the crossover shifts the inferred peak amplitude and scale by an amount that could already be relevant for the comparison with primordial black hole overproduction bounds. The importance of this SM effect will grow as the statistical uncertainty on the amplitude and scale shrinks with future, more sensitive PTA datasets, at which point neglecting it could significantly bias the inference.

The first two authors derived the Weitzenböck-type Formula for a CR Dirac operator. In this paper, we apply the above-mentioned formula and follow the process as Riemannian case to derive a Reilly-type inequality. We give two applications for the CR Reilly inequality. One is a lower bound estimate of the p-mass in terms of the first eigenvalue of the modified hypersurface CR Dirac operator. Another one is to solve the CR Dirac equation on a bounded domain with an $S^{1}$-action and an APS-type boundary condition.

Predicting blood-brain barrier (BBB) permeability is critical for central nervous system drug discovery. Using the MoleculeNet BBBP dataset (n = 2039), this study systematically ablates molecular feature spaces to isolate featurisation from model architecture. We evaluate three feature families (Morgan fingerprints, RDKit physicochemical descriptors, SMILES bigrams) across four learning algorithms. Results demonstrate that predictive performance depends jointly on feature representation and algorithm. Dynamic Random Forest using combined features achieved the highest mean AUC (0.970, 95% CI: 0.963-0.977). Second, this optimal representation enables exploratory estimation of heterogeneous associations between molecular structure and BBB permeability using Generalized Random Forests. Constructing a pseudo-treatment from a LogP median split, we applied double/debiased machine learning to account for confounding. Orthogonalization substantially attenuates the heterogeneity detected by naive causal forests; no conditional effects remained significant after false discovery rate correction (smallest adjusted p = 0.082). Furthermore, orthogonalized feature importance shifted toward residual structural information in SMILES bigrams. Ultimately, once observed confounding is properly accounted for, evidence that LogP-BBB associations vary systematically across chemical space is insufficient. This underscores that feature representation and model architecture are coupled design choices, and that unorthogonalized causal forests risk overstating genuine treatment effect heterogeneity.

Cancer registrars, including Oncology Data Specialists (ODSs), must interpret complex and frequently updated coding and staging standards. We developed CRISS (Cancer Registry Intelligent Support System), a retrieval-augmented generation (RAG) conversational assistant that provides rapid, citation-supported access to registry guidance. This study evaluated whether CRISS could (1) support accurate and citation-supported responses, (2) improve access to and interpretation of relevant guidance, and (3) support training/helpdesk use while preserving human oversight of final abstraction decisions. We built a domain-specific knowledge base from national cancer registry standards, segmented into metadata-tagged passages and indexed as dense embeddings. Retrieved passages were used to generate citation-grounded responses through a large language model (LLM). Open-weight, proprietary, and non-RAG baseline models across Gemini and GPT families were evaluated on easy, medium, and hard registry questions using an LLM-as-a-Judge protocols. RAG configurations consistently outperformed non-RAG approaches, especially as question difficulty increased. Mean grounding scores for RAG were 0.62/0.56/0.59 across easy/medium/hard tiers versus 0.29/0.26/0.29 for non-RAG. RAG models also achieved higher semantic-similarity scores overall. Proprietary RAG models performed strongest on easy and medium questions, while local RAG models ranked highest on hard questions and proprietary models were generally more cautious. Domain-specific RAG improved evidence grounding and response quality for cancer registry questions while enabling citation-supported assistance across complexity levels. CRISS demonstrates the potential of human-centered, citation-grounded AI to support cancer registrars while preserving human oversight for final coding decisions.

We show that the angular structure of the leading radiation around a bound state in the continuum (BIC) directly encodes its topology. This angular information can be determined from the BIC itself and used to design prescribed topological polarization textures. As examples, we identify BICs carrying high topological charges $q=4$ and $q=-5$ in $C_{6v}$ photonic crystal slabs, addressing the question of whether high-charge BICs exist and how they can be systematically realized. We further show that zeros of the leading radiation can drive both topological transitions and the bifurcation of new BIC families. These findings may enable new approaches to polarization engineering in structured light and resonant photonics.

Vision-Language Models (VLMs) describe scenes well but reason poorly about metric 3D structure such as absolute distances, physical sizes, or egocentric directions. We present a modular, predictor agnostic, tool-augmented framework that equips a small VLM (Qwen3.5-4B) with geometric tools: 3D object detection, metric depth estimation, and deterministic solvers for distance, size and bearing. Each object is detected in the camera frame of its own best view, and the tools use that frame's pose to lift every detection into one shared world frame. Moving metric computation out of the model's weights and into explicit solvers yields large gains on three of four ReVSI-Bench tasks: with a strong monocular detector (WildDet3D), absolute distance rises from 0.46 to 0.74 Mean Relative Accuracy (MRA), relative distance from 39.1% to 67.4%, and relative direction from a below-chance 25.9% to 73.4%. Because any detector can be swapped in behind the tool interface, comparing real detectors against ground-truth boxes separates perception error from reasoning error: orchestration costs only 0.03 MRA. Object size is bounded by the detector: the tools are near-exact on groundtruth boxes (0.97) yet the best real detector barely beats the no-tool baseline (0.61 vs. 0.58), because size reads straight off a box extent monocular detectors get wrong. Without a predefined recipe, the model already sequences the tools correctly on its own, matching a scripted pipeline on three of four tasks.

The exploration of the early Universe is being transformed by the James Webb Space Telescope (JWST), which delivers unprecedented angular resolution at infrared wavelengths and opens a unique window redward of the K band (2um). Thanks to NIRCam, NIRISS, NIRSpec, and MIRI instruments, which provide both imaging and spectroscopy with exquisite efficiency, new classes of sources have emerged within the first years of operations. The first half-Gyr of cosmic time is now routinely probed, revealing massive blue/red galaxies and a population of Active Galactic Nuclei (AGN) appearing as "little red dots" together with a rest-frame near-infrared/optical view of sources across the reionization and post-reionization epochs. Angular resolution will remain pivotal in the 2030s - 2040s, when extreme adaptive optics (AO) facilities will be deployed on both (8 - 10)m (e.g., the VLT Multi-Conjugate-Adaptive-Optics (MCAO) - Assisted Visible Imager and Spectrograph, MAVIS) and on extremely large telescopes, like the 39m ELT (e.g., Multi-conjugate adaptive Optics Relay For ELT Observation, MORFEO). Operating at the diffraction limit, these facilities will improve JWST's resolution, with ELT achieving a factor of ~ 6 smaller Point-Spread-Function (PSF). An ELT diffraction-limited PSF (with a Full Width Half Maximum, FWHM ~ 8-12 mas) in the near infrared will resolve spatial scales <100 pc at any redshift (z ~< 18), revealing abundant star-forming clumps with sufficient sensitivity. Leveraging gravitational lensing as a cosmic telescope, even with moderate magnification factors (mu ~ 4-8), diffraction-limited 8m and 39m telescopes will probe physical scales ~< 25 pc, enabling systematic studies of star formation down to star-cluster scale at cosmological distances. Such observations are poised to become routine in the 2030s - 2040s.

Score-informed note separation seeks to extract the performed waveform of all individual notes, often from a polyphonic recording. Existing deep learning systems generally only target instrument-level stems. We present, to our knowledge, the first deep learning approach to score-informed note separation, NoteSep. NoteSep extracts the queried notes by applying an extraction stage model, NoteGrab, once per note. Conditioned on pitch, onset, and offset, NoteGrab separates harmonic and percussive components in two U-Nets linked by bidirectional cross-attention; selective harmonic gating suppresses lower-octave interference while preserving percussive attacks. Finally, a joint separation stage applies Adaptive Set Ownership (ASO) to compare concurrent NoteGrab estimates and reallocate mixture energy. We curate SCNS-Train (25,729 mixtures and 743,920 targets) for training and SCNS-Eval (16 instruments, disjoint scores and libraries) for evaluation. On SCNS-Eval, NoteSep reaches a median SI-SDR of 7.39~dB, compared with 2.49~dB for our strongest baseline. See the demo page at https://benschou.com/notesep.

In this paper, we investigate the existence of pairs of orthogonal quantum Latin squares, each with maximal cardinality. Using characters and full-rank factorizations of finite abelian groups, together with a product construction, we establish two infinite families of orders for which such pairs exist.

We introduce BranchShine-CR, a 25M-parameter model for multilingual transcription into the International Phonetic Alphabet (IPA). It combines log-mel features, a rotary-position E-Branchformer encoder, intermediate self-conditioned connectionist temporal classification (CTC), and consistency regularization across augmented views. On 16,646 shared IPApack++ test utterances, it achieves 4.47% IPA character error rate, a 22.3% relative reduction from ZIPA-CTC-NS, with approximately one-twelfth as many parameters while being trained from scratch. BranchShine-CR also outperforms a similarly sized NeMo Conformer baseline across all 41 dataset language labels. Ablation studies indicate the individual components synergetically acting in model performance contribution. These findings support compact IPA recognition capabilities under limited compute budget, for applications in low-resource on-device pronunciation assessment.

The purpose of this research is to design, develop, implement, and provide preliminary validity and reliability evidence of the Software Engineering Self-Efficacy Scale (SESES). Framed by a conceptual framework using guidance in software engineering curriculum and concepts along with the notion of self-efficacy, we generated an initial item pool of n = 87 items to operationalize and measure software engineering self-efficacy among undergraduate computing students. The conceptual framework traces five dimensions: 1) Requirements Engineering, 2) Teamwork and Collaboration, 3) Software Quality Management, 4) Software Design and Architecture, and 5) Software Agile Methodologies. We pilot tested the SESES with n = 527 undergraduate computing students who had completed a software engineering course in the current semester or a previous academic semester. We employed Exploratory Factor Analysis (EFA) with the Principal Axis Factoring method and an oblique (Promax) rotation to examine the underlying structure of the SESES, resulting in the same five internally consistent latent constructs in the conceptual framework with minimal cross-loading and a simple structure in the pattern matrix, explaining approximately 57% of the variability in these data. Our findings suggest that software engineering self-efficacy is a multidimensional construct of five theorized and correlated, yet distinct latent factors. We unpack the limitations and delimitations of the research while exploring undergraduate computing students' software engineering self-efficacy using necessary domain-specific measurements.

Tile-based programming models, such as Triton and cuTile, aim to simplify high-performance kernel development, but their practical performance, tuning behavior, and usability remain difficult to compare systematically. We present TileBench, a controlled benchmark for evaluating Triton and cuTile on NVIDIA B200 GPUs under matched operator semantics and comparable implementation structures. TileBench contains 45 operators covering diverse AI-kernel patterns and memory/computation behaviors. Each task provides a PyTorch reference, verified Triton and cuTile implementations, standardized data-types (dtype) and input-size sweeps, default and autotuned configurations, roofline-based metrics, and profiling-guided diagnosis. Our evaluation shows that performance gaps are workload-dependent: cuTile excels on a small cluster of Tensor-Core/TMA-friendly kernels, while Triton is stronger on many irregular, streaming, and bandwidth-bound operators. We further evaluate LLM-generated cuTile and Triton kernels and find that Triton is consistently more token-efficient than cuTile under the same iterative refinement protocol. TileBench is publicly available at https://github.com/Deep-Learning-Profiling-Tools/Tilebench.

A decision maker may have several information sources available and choose which one to consult only after learning the decision problem she faces. When is one such set of sources uniformly more valuable than another? For unrestricted Bayesian decision problems, we show that the answer can be stated entirely in terms of Blackwell comparisons. Form a tagged mixture by drawing a source independently of the state and revealing both its identity and its signal. One source set is more valuable in every decision problem if and only if each tagged mixture of the second source set is Blackwell dominated by some tagged mixture of the first. The result applies to compact, possibly infinite source sets and general signal spaces. It also has an exact quantitative counterpart: the largest normalized value shortfall is the directed Le Cam deficiency between the source sets' tagged hulls. The analogous program for monotone decision problems reveals a boundary. We call the passage from problem-by-problem source-set superiority to a problem-independent pairwise dominance a lifting. The Blackwell lifting does not extend directly to the Lehmann order: mixing sources that individually satisfy the monotone likelihood ratio property (MLRP) need not preserve MLRP, and even when it does, no fixed Lehmann-dominating mixture need exist. Requiring one source to serve a finite bundle of monotone decision problems restores the equivalence.