Peer Revenue Returns subject to peer review

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

All of arXiv

Tuesday 1 September

Information Retrieval

Index-Free Dynamic Edge Retrieval with Energy-Tail-Aware Partial Scans

Dynamic maximum inner-product search (MIPS) returns the $K$ stored vectors with the largest dot products with a query while allowing the dataset to change through insertions, replacements, and deletions. For edge retrieval, the challenge is to achieve high recall and fast queries without making updates expensive. Full-vector scanning keeps updates simple but compares each query with every stored vector, while indexed methods reduce query cost at the expense of maintaining additional structures during updates. We propose ETAR, an index-free method that reduces query work while preserving simple updates. ETAR keeps the query coordinates with the largest squared values until they cover most of its total squared magnitude and treats the rest as a low-magnitude tail. It estimates similarity from the retained coordinates using a compact lower-precision representation, corrects for skipped coordinates, and reranks a fixed number of candidates using full-precision vectors. Across five runs on nine static datasets, ETAR averages 99.2% Recall@10, the fraction of exact top-10 results recovered, while running over 4$\times$ faster than exact scanning at a representative setting. This speedup also extends to an ARM-based mobile device, where ETAR is up to 6.9$\times$ faster across four synthetic distributions. Under five streaming workloads, it maintains 100% Recall@10 at every measured point without index rebuilds. Overall, ETAR offers a practical middle ground for dynamic MIPS by reducing query cost while retaining simple, index-free updates. Code is available at https://github.com/arasyi/etar-mips.
Read the full abstract

Metric Geometry

Asymptotic dimension of 3-dimensional CAT(0) manifolds

We prove that every $3$-manifold equipped with a proper metric admitting a convex geodesic bicombing has Assouad–Nagata dimension at most $3$. In particular, every proper CAT$(0)$ $3$-manifold has Assouad–Nagata dimension at most $3$. As a corollary, the isoperimetric results of Lang–Stadler–Urech \cite{LSU} and Peteranderl \cite{Pet}, previously applicable to $3$-dimensional Hadamard manifolds, also apply to proper CAT$(0)$ $3$-manifolds of asymptotic rank at most $2$.
Read the full abstract

Databases

Zeta-Lite: A Concurrent, Branchable In-Browser SQL Database for Agentic Memory

The browser has become a first-class database host: applications increasingly want to store, query, and reason over structured data entirely on the client - for privacy, offline operation, local-first collaboration, and, most recently, as durable memory for in-browser AI agents. One way to get SQL in the browser, compiling PostgreSQL to WebAssembly (PGlite), inherits PostgreSQL's process model: a single backend connection that executes one statement at a time and blocks. That model cannot express concurrent transactions, and it leaves richer capabilities - graph queries, database branching - to whatever the compiled server happens to include. We present zeta-lite, the browser form factor of the Zeta database engine: a WebAssembly build that compiles the same Zeta server down to a 2.87 MB gzipped artifact. Zeta-lite keeps the engine's log-centric asynchronous MVCC core, which yields two capabilities no other in-browser SQL engine provides. First, overlapping snapshot-isolated transactions on a single thread: multiple transactions hold distinct read/commit timestamps and interleave, with snapshot-isolation conflict detection between them. Second, copy-on-write database branching - whole-database fork, merge, and rebase - is unique in a browser SQL database and rare even in servers. On top of these, zeta-lite exposes a feature-complete PostgreSQL surface (joins, CTEs, window functions, JSONB with GIN indexes, full-text search, HNSW vector search, SQL/PGQ graph queries, multi-database) and snapshot-to-OPFS durability. Across Chrome, Firefox, and a native reference runtime, zeta-lite sustains 268k-315k point reads/s and holds a mixed read/write workload flat over millions of operations. This small, fully-featured, concurrent SQL database is an especially good fit for agentic memory - where cheap branchable state lets an agent explore, inspect, and commit or discard speculative work.
Read the full abstract

Quantum Physics

Bosonic quantum control with a weakly coupled fluxonium qubit

Echoed Conditional Displacement (ECD) gates constitute a fundamental building block for quantum control of harmonic oscillator modes. However, bit-flips of the auxiliary qubit remain a dominant error mechanism for this kind of bosonic control. In this work, we present a numerical case study of a bit-flip protected fluxonium operating as the control qubit and numerically implement ECD gates in a single-mode resonator-fluxonium device, demonstrating that fidelities exceeding 99.9% are possible. We systematically investigate the resonator dynamics using a combination of semiclassical trajectories and master equation simulations, numerically revealing asymptotic saturation of the dispersive shift in the strongly driven regime of the resonator. We develop an efficient technique to numerically simulate the strongly driven regime of the resonator using a semiclassical formulation that maps the full perturbation series in the dispersive expansion as order-by-order frequency shifts. This provides a compact polynomial description of the resonator which is intuitive and remains valid throughout the dispersive regime. Furthermore, we propose an improved ECD sequence that accounts for the effects of photon loss and spurious nonlinear terms on resonator trajectories.
Read the full abstract

All authors Anaida Ali, Shantanu R. Jha, Shoumik D. Chowdhury, Lev-Arcady Sellem, Max Hays, William D. Oliver, Baptiste Royer

Computer Vision

Video2Reaction: Training Foundation Video Models to Predict Audience Reaction

We introduce Video2Reaction, a multimodal dataset that maps short movie segments to the induced emotional reactions of viewers in the wild, as expressed through social media comments. Video2Reaction captures the natural diversity of emotional responses by aggregating reactions from online comments at scale, modeling labels as distributions over categorical emotions to better reflect the subjective and ambiguous nature of emotional perception. We benchmark two vision-language models (VLMs) finetuned with LoRA, showing that VLMs learn effectively from Video2Reaction and outperform specialized baselines on dominant reaction prediction. We further demonstrate that VLMs pre-finetuned on Video2Reaction transfer effectively to VCE, another induced emotion dataset with a different taxonomy and video domain. Notably, LLaVA-NeXT-Video-7B pre-finetuned on Video2Reaction and adapted on only 1% of VCE training data achieves a top-3 accuracy of 0.682, on par with the best reported VCE performance trained on the full dataset. The dataset is available at https://huggingface.co/datasets/infofusionlab/Video2Reaction
Read the full abstract

All authors Sidong Zhang, Trang Nguyen, Shiv Shankar, Gauri Jagatap, Deepak Chandran, Andrea Fanelli, Madalina Fiterau

Artificial Intelligence

Induction and Inquiry via Probabilistic Reasoning over Language and Code

How humans grow and maintain abstract knowledge from the sparse, streaming noisy data of experience is a longstanding challenge in cognitive science. Any computational account must satisfy at least three desiderata: It must be (1) data-efficient and compute-efficient, (2) capture gradations of uncertainty to support intelligent inquiry and information gathering, and (3) be flexible enough to mentally represent the endless range of concepts people can learn and think about. Here we introduce a computational model that captures these three properties, by encoding symbolic knowledge as mental programs that combine natural language with source code, and sequentially inferring mental programs using LLM-guided Bayesian learning algorithms. Across a range of behavioral studies this model successfully reproduces quantitative signatures of human inductive learning and active inquiry, such as anchoring, garden-pathing, and other effects. In contrast, pure LLMs and classic Bayesian models either fail at the underlying task, or do not reproduce human behavior, or succeed only at exorbitant computational cost. These results suggest that one way humans continually grow their knowledge is by mentally representing many hypotheses spanning language-like and program-like representations, then revising those hypotheses to approximate Bayesian updates, while a bottom-up neural mechanism (an LLM) makes inference both tractable and learnable.
Read the full abstract

All authors Wasu Top Piriyakulkij, Sam Acquaviva, Cassidy Langenfeld, Joshua Tenenbaum, Kevin Ellis

Artificial Intelligence

When Does Information Sharing Improve Decentralized Discovery? Aggregation, Independent Rescue, and Equilibrium Selection

Information sharing can improve a pooled estimate while eliminating independent rescue actions. This paper separates those effects in exact finite discovery models. A centralized action-budget profile shows that equal one-person accuracy can coexist with different portfolio values. Under a registered incremental-sharing protocol, a sharing step improves discovery exactly when pooled residual error contracts faster than an independent rescue attempt. Exact bounded registries exhibit compression, aggregation, neutral curves, and a bounded zero mixed class. In a two-agent Bayesian game with a hidden mixture of common and independent signal sources, the registered selected equilibrium yields a strict positive sharing interval at signal accuracy 3/5, while alternative equilibria show that the result is selection-dependent rather than universal. The models are synthetic and finite; no human or organizational data are used.
Read the full abstract

Human-Computer Interaction

Beyond Instruction-Driven Editing: Source-Grounded Problem Discovery with User-Governed Repair for Scientific Posters

Interactive editors usually assume that users already know what to change. Yet an important interaction state comes earlier: a user may recognize that an artifact is not working without knowing what intervention to request. We call this the articulation gap. We introduce PROS (Proactive Refinement Of Scientific Posters), which separates epistemic initiative from behavioral authority: the system can surface source-grounded candidate problems, while users decide which become repair goals and whether resulting changes are committed. Accepted issues hand off to native-object PPTX editing with validation and reversible preview. We also introduce PROS-Bench, a source-linked collection of 120 papers and 320 editable PPTX posters, including a 120-poster matched primary core and a separate conference representation challenge. On the primary core, PROS achieves a mean VLM-rated stage-balanced diagnosis quality score of 67.2 on a 0-100 scale and 87.6% operator-verified target resolution among accepted diagnoses. Temporally blinded automated scoring yields a +22.7-point paper-macro accepted-target uplift, yet 14.8% of assessable accepted targets decline. This divergence shows why problem discovery, local resolution, and realized outcome should be evaluated separately. More broadly, intelligent editors can support problem discovery before a concrete edit request exists without taking authority over consequential change.
Read the full abstract

All authors Xingda Lyu, Honglin Lu, Xinye Luo, Shiqi Yang

Mesoscale and Nanoscale Physics

Tuning Boron-Vacancy Qubit Coherence through Layer Number in hBN

The negatively charged boron vacancy ($\VB$) in hexagonal boron nitride (\hBN{}) was the first optically addressable spin qubit identified inside a van der Waals crystal, allowing atomically defined placement relative to a target. Its coherence in bulk \hBN{} is limited by the boron nuclei of the layers flanking the defect plane, which a thin flake removes. Here, we use a generalized cluster-correlation expansion with an extended central-spin block to calculate the Hahn-echo coherence time of $\VB$ in h$^{11}$B$^{15}$N as a function of layer number. Our results show that $T_2$ rises from $199\ns$ in the bulk limit to $653\ns$ in a monolayer, a factor of $3.3$ that is already saturated at three layers and that a sublattice decomposition attributes entirely to boron. The enhancement is confined to low field, is insensitive to stacking registry and twist angle, and requires alignment to within about two degrees, an onset we reproduce with no free parameters. Under dynamical decoupling, a $1/e$ threshold returns $206\ns$ for every thickness, but this is a zero of the first-shell modulation rather than a decay: measured without a threshold, the layer contrast survives and reaches a factor of $170$. Layer number thus emerges as a design parameter for $\VB$-based sensing.
Read the full abstract

Neural and Evolutionary Computing

Reinforcement learning to choose optimizers

No single optimization method is uniformly best for all problems, and the most suitable optimizer choice can change during a run. Existing approaches that change optimizer during execution typically predetermine part of the strategy: the portfolio is restricted to one algorithm class, the switch occurs once at a fixed time, or the frequency of decisions is treated as a hyperparameter rather than a learned one. We introduce "Reinforcement Learning to Choose Optimizers", which formulates the optimization algorithm choice as a sequential decision-making problem. At each decision, a recurrent policy reads the current run state and decides both which optimizer should be used next and for how long. The portfolio includes both gradient-based and derivative-free optimizers, and each switch passes on the current best solution and a representative step size. A context proxy conditions a gating network over expert heads, and training employs a decoupled actor-critic whose return is expressed in the same empirical runtime distribution metric used at evaluation. Training tasks and portfolio are designed jointly so that no optimizer dominates. On unseen problems, the learned policy outperforms every portfolio optimizer at all but the smallest budgets, and it remains robust under distribution shift.
Read the full abstract

Computation and Language

TalkFa: A Unified Benchmark for Farsi Dialogue Generation and Understanding

Farsi, spoken by more than 120 million people, lacks a comprehensive benchmark for dialogue generation and understanding. We introduce TALKFA, a unified benchmark comprising three complementary datasets: (1) WIKI-FADIAL, 4.2K Wikipedia-grounded dialogues for knowledge-grounded generation; (2) DAILYDIALOG-FA, 6.6K dialogues annotated for dialogue acts and emotions; and (3) PLAYDIAL-FA, 2.1K theatrical dialogues with sentiment labels. While LLMs assist data construction, every dialogue undergoes multi-stage review and revision by native Farsi speakers, and only the final human-approved dialogues are released. Experiments with six LLAMA and MISTRAL models show that LoRA substantially improves dialogue generation while requiring only 25-50% of the training data to recover over 90% of the final performance gains. Across classification tasks, FABERT achieves the best dialogue-act performance, LORA-MISTRAL-7B performs best on emotion recognition, and MISTRAL-24B achieves the highest sentiment score. Human evaluation and independent external validation demonstrate the reliability of the benchmark, while comparisons with GPT-4.1 as an LLM judge reveal that automatic metrics substantially overestimate dialogue quality. Zero-shot evaluation with frontier LLMs further shows that TalkFa remains a challenging benchmark. We will release all datasets, annotation guidelines, code, and checkpoints.
Read the full abstract

All authors Neda Jamshidi, Kamyar Zeinalipour, Fahimeh Akbari, Monica Bianchini, Marco Maggini, Marco Gori

Instrumentation and Detectors

Radiation Hardness of Commercially Available NUV-MT Silicon Photomultipliers

Silicon Photomultipliers (SiPMs) based on the near-ultraviolet, metal-filled trench (NUV-MT) technology offer improved photon detection efficiency and reduced correlated noise relative to earlier designs, making them attractive for a broad range of particle- and astroparticle-physics applications. As such devices may be deployed in high-radiation environments, quantifying their performance after irradiation is essential. Commercially available Broadcom AFBR-S4N series NUV-MT SiPMs were irradiated with 1 MeV neutrons to fluences between 2x10^{9} and 1.1x10^{10} n_{eq}/cm^2 at the TANDEM accelerator facility of NCSR "Demokritos" in Athens, and characterised before and after irradiation and over successive thermal-annealing stages. Irradiation increased the dark noise by up to three orders of magnitude and degraded the single-photon resolution by up to a factor of ten, with resolution lost entirely at the highest fluence, while no shift in breakdown voltage was observed. Thermal annealing partially recovered the performance, reducing the dark noise by up to a factor of two and restoring the single-photon resolution.
Read the full abstract

All authors T. Avgitas, M. Axiotis, Z. Balmforth, I. Manthos, K. Nikolopoulos, E. Taimpiri, C. Toukmenidis, A. Ziagkova

Computer Vision

Integrated Laser Scanning and Image-Based Topology Optimization Techniques for Detection and Quantification of Visible and Subsurface Structural Defects

Reliable characterization of structural defects requires methods capable of resolving both directly observable surface damage and damage that is not visible from the inspected surface. This study presents two complementary non-contact, vision-based approaches for the detection and quantitative characterization of defects in structural components. The first approach employs high-resolution laser scanning to generate three-dimensional (3D) point clouds of damaged steel specimens. Comparative processing of measured and reference point clouds is used to localize damaged regions, quantify geometric loss, and transfer the measured defect geometry to a finite element representation. The second approach combines full-field surface deformation measurements obtained using three-dimensional digital image correlation (3D-DIC) with finite element model updating and topology optimization. In this inverse framework, measured surface response is used to infer subsurface abnormalities through their influence on the spatial distribution of structural response. Experimental steel-beam specimens containing controlled smooth defects and randomly distributed defects are used to evaluate the approaches. Comparisons with milling-based ground-truth measurements demonstrate that both methods can identify and quantify defect geometry, while providing complementary information for visible and subsurface damage assessment. The combined framework establishes a pathway toward high-fidelity, non-contact structural condition assessment and model updating for components with complex and irregular damage.
Read the full abstract

Machine Learning

SPD: Single Pass Decoding for Generative Reranking

Large language models (LLMs) achieve state-of-the-art generative ranking quality, but the ranking they produce must be decoded, and autoregressive decoding spends one sequential forward pass per emitted token. We observe that the only tokens a ranker must emit are the $N$ ordinal values naming the items in ranked order, and that this narrow, permutation-structured output format admits decoding strategies which are much more efficient than left-to-right generation. We introduce SPD (Single Forward Pass), a format-specialized decoding strategy that decodes all $N$ ordinals in $O(1)$ forward passes. SPD reads an $N \times K$ item-position score matrix off the LLM's prefill hidden states with a lightweight self-attention head, then decodes the ordinals as the optimal bipartite assignment of that matrix via the Hungarian algorithm, yielding a valid permutation by construction rather than by repair. Through a systematic study of training signals and backbone adaptation, we show that LoRA-based fine-tuning combined with auto-regressive LLM ranking distillation reaches 28 ms end-to-end inference, a speed-up of 64x while maintaining ranking quality on par with the teacher. We provide a complete ablation decomposing the contributions of architecture, training signal, and backbone adaptation. Our framework connects generative ranking to combinatorial optimization, opening a path toward other $O(1)$-decode mechanisms for real-time ranking.
Read the full abstract

All authors Emil Laftchiev, Prachi Agrawal, Moe Kayali, Bixing Yan, Qi Xu, Zijie Lei, Chen Qiu, Zhi Hua, Ke Li, Luke Simon

Computer Vision

Consistency as Regularization for Unsupervised Shadow Removal

Shadow removal is an important preprocessing step for many vision tasks, yet existing supervised methods require paired shadow and shadow-free images, while unsupervised approaches often still rely on shadow masks or shadow-free references. We propose ShadowCLR, an unsupervised framework that learns shadow removal directly from shadow images. Our key observation is that shadows vary across observations while the underlying scene content remains largely consistent. We therefore use consistency across shadow observations as regularization, encouraging the model to recover scene-consistent appearance while suppressing shadow-specific variations. Global and local consistency further enable us to explore visually related images, learn from imperfectly aligned observations, and focus the representation on shared scene information. Experiments on multiple benchmarks show that ShadowCLR achieves competitive and often superior performance over state-of-the-art unsupervised methods, demonstrating that consistency can provide regularization for shadow removal without shadow masks or shadow-free images.
Read the full abstract

Strongly Correlated Electrons

Evolution from Kitaev to XXZ spin chains via distortion: Application to BaCo$_2$V$_2$O$_8$

BaCo$_2$V$_2$O$_8$ is a prototypical spin-orbit-coupled Ising-chain antiferromagnet that provides a unique platform for studying field-induced quantum magnetism. Under a transverse magnetic field, it exhibits unusual magnetic properties, including an anomalous staggered magnetization and a pronounced anisotropy of the critical field with respect to the in-plane field direction. While these phenomena have been attributed phenomenologically to a site-dependent anisotropic $g$-tensor, a recent microscopic theory has shown that spin-orbit coupling naturally generates bond-dependent Heisenberg, Kitaev, and $\Gamma$ exchange interactions. Here, we unify these two pictures by extending the microscopic theory to incorporate distortions of the CoO$_6$ octahedra. We show that the distortions not only generate the site-dependent anisotropic $g$-tensor but also renormalize the staggered exchange interactions through a distortion-induced contribution that partially compensates the Kitaev-derived staggered term. Using point-charge calculations to estimate the $g$-tensor of BaCo$_2$V$_2$O$_8$, we demonstrate that the strong anisotropy of the critical field originates from the combined effects of the modified exchange interactions and the anisotropic $g$-tensor. Our work provides a unified microscopic framework for understanding the magnetic anisotropy of spin-orbit-coupled Ising-chain materials.
Read the full abstract

Econometrics

Robust Variance Estimation in Linear Regression: A Projection-Geometry Perspective

Inference in linear regression commonly treats OLS residuals as proxies for unobserved errors. This approximation can fail when the regression projection is nonlocal relative to the error-dependence structure. Residualization then shifts covariance information across observations and clusters, while conventional heteroskedasticity-consistent (HC) and cluster-robust variance estimators (CRVE) retain only diagonal or within-cluster residual moments and may therefore understate sampling uncertainty. This paper develops a projection-geometry framework for robust variance estimation. The variance of the OLS estimator is represented exactly as a Riesz functional of latent covariance blocks, and observable residual moments are linked to the target through a linear operator determined by the full regression projection. This formulation reduces variance estimation to a linear inverse problem. I propose a Riesz variance estimator that combines within- and cross-cluster residual moments. Conventional HC and CRVE emerge as restricted approximations whose validity depends on negligible projection spillovers. The estimator remains well defined when cluster-specific leverage matrices are singular and is computed by an iterative algorithm that avoids explicit matrix inversion. Simulations show substantial undercoverage by conventional methods under projection spillovers, whereas the proposed estimator restores near-nominal coverage. In an application to colonial governor promotions, the correction changes the significance of four of five reported coefficients.
Read the full abstract

General Relativity and Quantum Cosmology

Exploration of Zero-Complexity Compact Stars in Higher Dimensions under the Finch-Skea Background

Motivated by the recent extension of Herrera's gravitational complexity to arbitrary higher-dimensional spacetimes, we investigate exact compact-star models that satisfy the vanishing-complexity condition within the Finch–Skea geometry in $(n+2)$-dimensional Einstein gravity. By combining the higher-dimensional Einstein field equations with the generalized complexity formalism, we obtain a new class of exact interior solutions describing anisotropic fluid spheres with zero complexity. The physical properties of these models are evaluated by analyzing the behavior of the matter variables, pressure anisotropy, and equilibrium conditions, alongside the standard requirements of regularity, energy conditions, causality, and stability. We further explore the influence of spacetime dimensionality on the structural characteristics of the stellar configurations, demonstrating that the presence of extra dimensions significantly modifies the internal matter distribution while preserving physical viability. The solutions presented here represent the first exact higher-dimensional Finch–Skea compact-star models constructed within this recently developed generalized complexity framework \cite{choudas}. These results provide a natural extension of zero-complexity stellar configurations beyond four-dimensional General Relativity and offer a robust framework for investigating self-gravitating systems in higher- dimensional gravity.
Read the full abstract

All authors Shyam Das, Megandhren Govender, Kevin Reddy, Bikram Keshari Parida

Machine Learning

D-FROST: Decentralized Federated pRompt-tuning via Optimal tranSporT for Non-IID and Imbalanced Data

Prompt tuning provides a parameter-efficient way to adapt foundation models (FMs) by freezing the pretrained backbone and updating only a small set of learnable prompts. This property makes prompt tuning especially suitable for decentralized federated learning (DFL), where exchanging full-model updates can be prohibitively expensive. However, prompt tuning in DFL introduces new challenges. Prompt sets learned from heterogeneous local data may not be index-wise aligned, making standard decentralized averaging unsuitable. In addition, the algorithm should be theoretically guaranteed to achieve consensus and make progress toward the shared objective. In this work, we provide the first study of prompt tuning in DFL. We formulate decentralized prompt tuning as a Wasserstein-based optimization problem over prompt measures, which captures the set-valued structure of prompts. We then propose D-FROST, an optimal-transport-based (OT-based) decentralized prompt-tuning algorithm that merges neighborhood prompts into compact representative prompt sets through transportation-based matching. We further analyze D-FROST by bounding the Wasserstein consensus error across clients, and establishing convergence of the network-level prompt barycenter to a neighborhood of stationarity. Experiments under heterogeneous client data demonstrate the effectiveness of D-FROST for decentralized prompt tuning.
Read the full abstract

All authors Quan Minh Nguyen, Hoang M. Ngo, Trong Nghia Hoang, My T. Thai

Solar and Stellar Astrophysics

Chemical evolution of Na, Mg, and Al in the Galactic bulge from UVES data

The formation of the Galactic bulge remains incompletely understood, with evidence pointing to different stellar populations, including a bar-driven component, an inner-disk population, and an older spheroidal component. Chemical abundances provide critical constraints on the origin of these populations, particularly for odd-Z elements such as Na and Al, as well as Mg, whose behaviour at high metallicity is still a matter of debate. We aim to investigate the presence of overabundances of Na, Mg, and Al in Galactic bulge stars, with particular emphasis on the metal-rich regime, and to evaluate their consistency with predictions from chemical evolution models. We re-derived the abundances of Na, Mg, and Al for a sample of 55 bulge red giants previously analysed in the literature. Our study is based on high-resolution UVES spectra obtained with the ESO Very Large Telescope and employs spectrum synthesis using the Turbospectrum code, with updated atomic and molecular line lists. We find somewhat lower abundances of Mg and Al at the metal-rich end than previous studies, while a fraction of the metal-rich stars still exhibit significant Na enhancements. These enhancements persist when different sets of stellar parameters are adopted, indicating that they are robust. The presence of Na-enhanced stars at high metallicity is difficult to reconcile with standard chemical evolution models and suggests additional enrichment processes in the bulge, or a particular behaviour of stellar yields with metallicity. The Na enhancement could be due to metallicity-dependent yields from massive stars, not taken into account in available models, and/or enrichment by asymptotic giant branch stars, or due to second-generation stars evaporated from globular clusters, the latter option arising because for the metal-rich ([Fe/H]>0) stars a Na-O anti-correlation appears to occur.
Read the full abstract

All authors R. P. Nunes, B. Barbuy, C. Chiappini, A. Friaça, M. C. Jelin, V. Hill, M. Zoccali, A. Renzini, A. Pérez-Villegas, H. Ernandes, S. O. Souza, M. J. Castro

Computer Vision

Improved Automatic Target Recognition in Synthetic Aperture Sonar Imagery Using Large Deep Neural Networks

Automatic Target Recognition (ATR) in Synthetic Aperture Sonar (SAS) is often underpinned by deep neural networks (DNNs). Recent research has focused on relatively low-capacity DNNs (i.e., relatively fewer trainable parameters) to avoid overfitting on the small training datasets that are typically available for SAS. However, recent research has found that increasing DNN capacity often improves performance, even when training data is scarce. Motivated by this emerging finding - termed the double descent phenomenon - this work investigates the use of especially high-capacity DNNs for SAS-ATR. We evaluate the performance of several state-of-the-art DNN architectures as we increase their model capacity. We examine performance in a realistic leave-one-geo-location-out cross-validation scheme. We find that increasing model capacity often increases performance substantially, and is never harmful. All of our large models significantly outperform two baseline models from recent publications. Training high-capacity models on small dataset however requires careful regularization, and we also perform systematic experiments showing the impact of several regularization strategies: pretraining on other imaging modalities, data augmentation (including some new to SAS), DropPath, and weight decay. We show that each of these strategies independently improves the performance of the large models.
Read the full abstract

Robotics

Designing Versatile Samples for Learned Trajectory Scoring

Many current end-to-end driving policies emit a pool of candidate trajectories and select one, which makes selection a separable component: a scorer can be retrained while the planner, its backbone, and its trajectory generator all stay frozen. However, many strong planners concentrate their proposals around safe mode, providing limited supervision near decision boundaries. In this work, we design a training dataset that provides more informative supervision for the scorer. In particular, we construct two generators that perturb the logged human trajectory along the two axes a vehicle can be displaced: laterally toward the drivable boundary and longitudinally toward a leading vehicle. The designed dataset produces more informative positive and negative samples than the base planner's proposal pool. We attach a transformer-based scorer to two frozen generative planners, DiffusionDrive and MeanFuser, and train it on the NAVSIM navtrain dataset. The results of the experiments show that we achieve 90.1 EPDMS on DiffusionDrive and 90.4 EPDMS on MeanFuser when using ResNet-34, with 0.4 and 0.3 EPDMS respectively, from the designed training dataset. The code is publicly available at https://github.com/PurdueDigitalTwin/Versatile-Scoring.
Read the full abstract

All authors Yaguang Li, Jiaru Zhang, Chuheng Wei, Can Cui, Ziran Wang

Computation and Language

How Do Prompt Variations Affect Energy Consumption in On-Device LLMs?

Large language models (LLMs) are increasingly deployed on mobile devices, making energy efficiency a key deployment constraint, yet the energy impact of prompt design remains underexplored. This paper aims to understand how two prompt properties, cognitive load and phrasing pattern, shape the energy behavior of on-device LLM inference. We conduct a broad empirical study covering prompt properties, datasets, models, and devices, with phase-level profiling that separates prefill and decode energy. We find that cognitive load primarily affects the energy cost per token, while phrasing pattern affects energy largely through token usage. Our energy-quality analysis further shows that prompt design reshapes the attainable frontier differently across models, highlighting the need for model-aware prompt design in energy-efficient on-device LLM inference. Code, datasets, and scripts are available at https://amai-gsu.github.io/PromptProperty/.
Read the full abstract

All authors Wei Hu, Xiaolong Tu, Dawei Chen, Yitao Chen, Kyungtae Han, Haoxin Wang

Quantum Physics

Quantum Representation Learning Beyond Pairwise Fidelity

Quantum contrastive, metric, and self-supervised learning often expose encoded quantum states to the learner through transition probabilities, especially fidelity. Quantum states are known to possess higher-order relational invariants, but their consequences for learned representations remain unclear. Here we show that a transition-probability-only learning interface can possess exact continuous blind directions in certain quantum-state families. We recover this missing information with a batch operator, built from coherent overlap amplitudes and negative masking, where its second moment $q_-$ retains four-state interference. Moreover, $q_-$ is directly measurable through two-copy interference and can enter variational learning via methods like parameter shift. In relational quartets derived from toric-code and double-semion states, this fidelity-blind signal encodes inequivalent modular data despite identical pairwise fidelities. Finally, in a four-photon benchmark with preparation drift, augmenting all six pairwise fidelities at two orthogonal probes with the corresponding normalized $q_-$ reduces the mean out-of-distribution phase error by $86\%$ at equal total shot budget. These results establish multistate relational observables as measurable, trainable, and physically consequential signals for quantum representation learning beyond pairwise fidelity.
Read the full abstract

Astrophysics of Galaxies

Dark Matter Halo Tumbling Induced by Torques from Massive Mergers

Dark Matter halos exhibit slow tumbling known as `figure rotation'. Figure rotation occurs in roughly $80\%$ of isolated $\Lambda$CDM halos and is relatively unaffected by baryonic feedback, making it an attractive and relatively unexplored test for $\Lambda$CDM. However, galaxy mergers are capable of exerting strong torques on a host halo's quadrupole, warranting an investigation of whether such interactions are capable of influencing figure rotation. Using isolated $N-$body simulations and 6 TNG50 halos containing MW analog galaxies, we investigate figure rotation in the context of massive mergers. For isolated galaxies we find that massive mergers elicit rotation of the host halo in both a rapid transient phase during the satellite infall with pattern speeds between $10^\circ-60^\circ$ Gyr$^{-1}$, and post-merger as a steady tumbling with pattern speeds between $2^\circ-40^\circ$ Gyr$^{-1}$ lasting for at least $6-8$ Gyr. Both transient and long-lived tumbling depend on the trajectory of the infalling satellite. We find that figure rotation can be induced by radial mergers with no orbital angular momentum, suggesting that tidal torques from the satellite are a driving mechanism. Meanwhile, TNG50 halos show quadrupole tumbling modified during major mergers in both pattern speed (3/6 halos) and rotation axis orientation (6/6 halos). Quadrupole tumbling remains stable after the last major merger in 5 of our 6 halos. Our results suggest that, rather than tracing the early tidal shear, figure rotation at present day may instead retain a dynamical memory of a halo's last major merger.
Read the full abstract

Computer Vision

DESA-TTA: Dynamic EMA and Source Anchoring for Test-Time Adaptation

Vision-language object detectors (VLODs) achieve strong zero-shot performance but remain vulnerable to distribution shifts during deployment. Mean-teacher methods for test-time adaptation (TTA) can improve robustness by updating a student model using teacher-generated pseudo-labels. However, mean-teacher TTA is highly sensitive to the choice of a fixed exponential moving average (EMA) coefficient for teacher updates, and repeated optimization with noisy pseudo-labels can cause cumulative student drift. We propose Dynamic EMA and Source Anchoring for TTA (DESA-TTA), a low-overhead method that jointly regulates teacher updates and student drift through dynamic temporal averaging and source anchoring. Dynamic temporal averaging estimates teacher uncertainty from pseudo-label confidence and box density and uses it to select a sample-wise EMA coefficient within bounds determined by teacher parameter drift. Source anchoring partially restores the updated student parameters toward their pretrained values, with the anchoring strength increasing according to student drift. Experiments across diverse distribution shifts and two VLOD architectures show consistent improvements over existing TTA methods. On VOC-C, DESA-TTA improves AP$_{50}$ by 14.5 points over zero-shot inference while achieving 55% higher inference throughput than the previous state-of-the-art TTA method for YOLO-World. Our code: https://github.com/imatif17/DESA-TTA
Read the full abstract

All authors Atif Belal, Lilian Hollard, Marco Pedersoli, Eric Granger

Computation and Language

Disentangling Statistical Preemption from Entrenchment in Language Models' Avoidance of Overgeneralization

How do learners avoid overgeneralizations such as Tom laughed me without explicit negative evidence? Constructionists have posited two proposals that describe indirect negative evidence against overgeneralizations: preemption (which privileges exposure to near-synonymous construction—e.g., she made him laugh) vs. entrenchment (all exposures to a verb's grammatical usages, including cases like He laughed). We disentangle these hypotheses by running controlled rearing experiments on LMs trained on child-caregiver conversations, where we systematically remove preemptive vs. non-preemptive evidence. We find that while LMs avoid overgeneralizations, they do not show preemption at a verb-specific level, instead showing weak but non-zero evidence of abstract preemption. Combined with results from analyzing the LMs' training dynamics, we find that LMs treat competing structures as indirect positive—as opposed to negative—evidence in the verb-specific condition. Insofar as preemption is the more plausible route to avoiding overgeneralizations in humans, our results point the need for there to be sensitivities to indirect negative evidence in neural network learners, and suggest new human experiments to test abstract preemption.
Read the full abstract

Plasma Physics

Revisiting the MRX Electron Current Sheet Width with Semi-Collisional Kinetic Simulations at Hydrogen Mass Ratio

For over eighteen years, the electron current sheet measured in the Magnetic Reconnection Experiment (MRX) has stood a factor of 2–5 wider than predicted by MRX-like kinetic simulations, and the measured electron force balance has not closed with the classical terms alone. As a consequence, the dominant nonideal terms responsible for breaking the frozen-in condition have remained unexplained. Here, two-dimensional kinetic simulations with binary Coulomb collisions are performed in a cylindrical geometry representative of MRX, at the realistic hydrogen mass ratio $m_i/m_e = 1836$ and at MRX-relevant collisionality. For the first time, these simulations reproduce the measured electron current sheet half-width. The simulated value, $\delta_{BT} = 0.744 \pm 0.054$~cm or $6.26 \pm 0.45$ electron skin depths ($d_e$), lies within the experimental range of 5.5–7.5~$d_e$. The electron force balance closes through the classical channels alone: the pressure-tensor divergence supports 76% of the nonideal electric field and collisional friction the remainder. The historical force-balance deficit reappears only when the simulated layer is sampled at the experimental 3~cm outflow resolution, suggesting that the deficit reflects probe resolution rather than anomalous dissipation. Beyond this reproduction, an analytic model of the layer width is developed that orders the meandering electrons by the coherence of their orbits against collisions. In this model, the Dreicer ratio $E_D/|E_y|$ selects the electrons whose current-carrying motion survives, and the resulting width prediction brackets the measured values across a wide collisionality scan. A discrepancy remains in the width normalized to the local electron gyroradius ($\rho_e$), whose measured value lies a factor of 3–6 above both the model and the simulations.
Read the full abstract

All authors Sung Hyun Son, Adam Stanier, William Daughton, Jongsoo Yoo, Tongnyeol Rhee, Hantao Ji

Sound

Efficient Passive Acoustic Monitoring of Killer Whales Using a Two-Stage Detection and Ecotype Classification Cascade

Passive acoustic monitoring of killer whales is particularly important for conservation of the endangered Southern Resident killer whale population, but requires accurate models that can operate in real time under severe class imbalance and deployment shift. We propose a lightweight ResNet-based two-stage cascade that first detects killer whale vocalizations and then classifies confident detections into five eastern North Pacific ecotypes, abstaining on ambiguous calls. We train and evaluate the pipeline on the DCLDE 2027 dataset, where the detector achieves 0.960 macro-F1 and the classifier 0.958, outperforming frozen Perch 2.0 embeddings on the five-ecotype benchmark. By separating detection from ecotype recognition, the end-to-end cascade improves seven-class macro-F1 from 0.919 for a single-stage model to 0.933, with the largest gain on the rare OKW ecotype. To assess transfer beyond the benchmark, we use active learning to adapt the Stage 1 to the acoustic environment of Puget Sound, WA, increasing killer whale detection F1 from 0.405 to 0.755 on manually verified detection windows. Finally, each stage processes a 3 s window in approximately 1.4 ms on an NVIDIA H100, enabling faster than real time inference. These results demonstrate that the proposed two-stage cascade pipeline enables reliable killer whale detection and classification, adaptation to new acoustic domains, and real-time monitoring for conservation applications.
Read the full abstract

All authors Daniela Ruiz, Manuel Castellote, Zhongqi Miao, Carl Chalmers, Bruno Demuro, Rahul Dodhia, Pablo Arbelaez, Juan M. Lavista

Geometric Topology

Minimum $b_2$ of symplectic fillings of lens spaces

We classify lens spaces for which the canonical negative-definite plumbing has minimal $b_2$ among all symplectic fillings of the standard contact structure. The classification is given by 4 infinite families of "forbidden subgraphs" in a manner similar to Aceto-McCoy-Park's work on smooth negative-definite fillings of lens spaces. In particular, no classification of this form can be given using a finite list of forbidden configurations, highlighting a difference between symplectic fillings and smooth negative-definite fillings of lens spaces.
Read the full abstract