Systems and Control
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Friday 6 November
This paper presents an equivalent circuit model (ECM) for dynamic traffic modeling by exploiting analogies between electrical and transportation variables. Vehicle concentration is represented by capacitor voltage, while traffic flow is modeled using controllable current sources. The proposed formulation is shown to preserve the conservation and sending/receiving relationships of the spatially discretized Lighthill Whitham Richards traffic model. The framework is extended to restricted inflow, traffic signals, and two road merging and diverging scenarios. Validation against microscopic Simulation of Urban MObility (SUMO) simulations, involving traffic signals, two-road merging, and single-road diverging scenarios, demonstrates that the proposed ECM successfully reproduces the principal macroscopic traffic dynamics, including queue formation, backward congestion propagation, vehicle accumulation, saturation, and subsequent traffic recovery.
Computer Vision
From Weapons to Events: Evaluating Evidential Support in Multimodal AI for Policing
Computer vision (CV) for policing has centred on recognising objects such as seized weapons, while body-worn video (BWV) is still reviewed by people, with existing AI tools using only its audio. Multimodal video AI could make BWV review faster and cheaper. At the object level, we present the KnifeHunter dataset, developed with UK law enforcement, with over 25,000 images of 543 knife classes and a protocol reflecting investigative conditions, and KHNet, a retrieval architecture fusing local and global features, which with about 125,000 distractor images attains 85.7% and 63.4% mean Average Precision (mAP) on the medium and hard protocols. At the event level, we present an evaluation protocol for video understanding pipelines, which describe a video as events and a transcript, that scores transcription, speaker attribution, event detection, temporal localisation and unsupported events. Whether each event is supported by the recording is labelled by a judge validated against blinded human labels. We apply it to complete pipelines, three 2026 omni-modal models and our hybrid NarrativeBridge-PAL-GPT-5 pipeline, on UKBWV-27, a pilot stress-test set of 27 publicly released UK BWV recordings (118 minutes). On this set the hybrid pipeline leads on five of six measures, yet it leaves 13.6% of predicted events unsupported and misattributes 27.7% of utterances, and every pipeline degrades on recordings from edited releases, long recordings and noisy audio. We propose readiness levels separating assistive tasks from those requiring verification of every output.
Read the full abstract
Optimization and Control
The microstructure of information through mean field learning interaction
We study how knowledge about partially observable model parameters is being created and propagated in large scale systems. Information is generated from observational signals whose dynamics depend endogenously on the learning history, the learning duration, the decisions and the related costs encountered by a continuum of decision makers. Such characteristics give rise to a new class of sequential testing problems in which there is continuous endogenous feedback. Adopting the classical approach of Shiryaev, we relate the problem to an optimal stopping one for the representative belief process, whose generator now depends on the evolving interactions. Considering a fixed point for the latter, we introduce a notion of mean field game and prove the existence of an equilibrium. Our arguments, which are based on a combination of pde and probabilistic techniques, involve a detailed analysis of the underlying variational inequality and the geometry, stability and continuity of the free boundaries.
Read the full abstract
Algebraic Topology
A tale of two periodicities
We provide a common explanation for two 4-periodic phenomena arising in the study of connective real K-theory: first, the family of twists of ko over $B\mathbb Z/2$, and second, the torsion in the Picard group of $Q_1$-local $\mathcal A(1)$-modules. These phenomena have been previously studied, as they lead to 4-periodic patterns in Smith homomorphisms, resp. in the ko-homology of $(B\mathbb Z/2)^n$. In this paper, we exhibit both of these 4-periodicities as consequences of an isomorphism of two extraspecial groups of order 32.
Read the full abstract
Databases
LoStBench: Benchmarking Longitudinal Studies Data Integration
A longitudinal study observes the same subjects or variables, repeatedly over an extended period of time, allowing researchers to examine changes, trends, and relationships as they develop. Such studies typically evolve over their life cycles, as improved understanding of the study's domain, or external factors trigger the need for revisions. The resulting heterogeneity among collected data must be resolved before the survey's results may be analyzed. Although this problem is a form of schema matching, it is a form of schema matching that differs significantly from prior workloads explored by the database community. For instance, rather than the short attribute names and data instances that many schema matching tools rely on, a stronger signal in longitudinal studies arises from longer prose questions that these tools are less equipped to handle. This paper documents the joint efforts of a team of computer scientists and domain experts to adapt existing schema matching algorithms to our new use case of longitudinal data integration. We present a case study in the context of a representative workload: The American National Election Studies (ANES) dataset, a nearly eight-decade survey with publicly available data and a canonical domain-expert curated schema matching. Using ANES, we develop LoStBench (Longitudinal Studies Benchmark for data integration), a medium-scale, subject-matter-expert-annotated benchmark. Using LoStBench, we conduct a comprehensive empirical study evaluating state-of-the-art schema matching algorithms, and expose limitations of these algorithms when applied in the context of longitudinal studies. We propose improvements that address some of these limitations, and identify key challenges to motivate future data integration work to address the others.
Read the full abstract
Robotics
OfficeTwin: Realizing 3D Reconstruction of Redundant Scenes through Partial Viewpoint Perception
Feed-forward 3D reconstruction turns an uncalibrated monocular stream into a dense point cloud in a single inference pass, orders of magnitude faster than classical structure-from-motion (e.g., COLMAP), and reliable on the weakly textured, repetitive surfaces where keypoint matching fails outright. However, raw point clouds lack the object identity, physical properties, and watertight volume required for robot learning. This gap widens under robot-native capture. A camera mounted on a robot constrained to a few degrees of freedom sees each object only partially, and the repetitive, weakly textured structure of an office makes distinct workstations nearly indistinguishable, leaving redundant and partial observations that downstream reconstruction cannot resolve. Existing image-to-3D generation methods struggle with these conditions, hallucinating geometry under incomplete views and confusing repeated instances. We present OfficeTwin, a robot-native framework that converts these incomplete observations into simulation-ready digital twins. OfficeTwin first imposes metric and structural constraints, then exploits the office's structural redundancy by registering repeated cubicles into a shared canonical frame, where complementary observations recover geometry occluded in any single view without hallucination. On a real office scan, OfficeTwin recovers every cubicle as an individually watertight, metric solid, outperforming feed-forward reconstruction and generative baselines that fail to produce simulation-ready assets.
Read the full abstract
Machine Learning
PACE: Keeping the Right Tokens in the KV Cache Is Not Enough for Long-Horizon LLM Agents
Long-horizon LLM agents accumulate sessions of up to millions of tokens, far more key-value (KV) cache than one GPU holds, so serving systems keep a small budget of tokens and evict the rest. Eviction methods differ in which tokens they keep but share one goal: that the tokens a later step needs are still cached. We show this is not enough. In 100K-token coding sessions on Qwen3-8B with a 2,048-token budget, a planner-guided policy that keeps over 99% of the needed tokens resident still fails late steps increasingly often: its accuracy falls from 93% to 64% as the gap between a definition and its use grows from 2K to 90K tokens, at residency 0.996. The needed tokens are resident but unretrievable: what grows is their rotary-position distance to the query. We present PACE (Planner-Aware Context Engine), a KV-cache engine that couples plan-aware retention, in which the agent's planner tells the cache which context future steps will read, with bounded positions, which renumber kept tokens so their rotary distance to the query stays below 8K at any session length. In a pre-registered evaluation on fresh tasks, PACE passes 97% of hidden unit tests at 100K tokens, against 42% for H2O and 3% for SnapKV under the same budget, and needs both components: either alone reaches 63% (the combination beats each, p < 10^-4). The result holds with length: in 1M- and 2M-token sessions on one 24 GB GPU, with definitions written 0.9M and 1.8M tokens before use, PACE passes 99% and 100% of checks and recalls every session fact, while bounded positions without foresight falls from 33% to 13%; PACE holds under 0.7 GB of KV state where the full cache would need 147-295 GB. Renumbering is exact, costs no measurable perplexity, and adds 9% wall time. On Mistral-7B, whose coding ability caps it at 40% even with nothing evicted, PACE stays at that ceiling while retention alone falls below it.
Read the full abstract
Robotics
DASH-SLAM: Calibration-Free Metric Monocular SLAM via Depth Scale Propagation
Geometric foundation models estimate dense geometry and camera parameters from monocular images, but reliable SLAM also requires consistent scale and accurate data association across successive reconstructions. We present DASH-SLAM, a monocular SLAM system that integrates metric depth predictions from pretrained Depth Anything 3 into a shared-keyframe pose graph without task-specific fine-tuning. DASH-SLAM derives scale corrections from depth ratios at shared anchor images and accumulates them across reconstructions, with a damping factor controlling their influence. These corrections rescale odometry translations, loop constraints, and map geometry. Experiments on TUM, Replica, and 7-Scenes show that DASH-SLAM outperforms state-of-the-art methods. We evaluate performance using both rigid and similarity alignment to distinguish metric-scale fidelity from accuracy up to scale. We further evaluate DASH-SLAM on four challenging sequences in perceptually-degraded environments from the Unified Autonomy Stack field dataset, covering tunnels, fog, and a frozen lake. On these sequences, DASH-SLAM outperforms the foundation-model SLAM state-of-the-art and maintains tracking where classical SLAM methods fail.
Read the full abstract
Artificial Intelligence
CoRA-Bench: Evaluating Evidence Preservation by Context Selectors Under Token Budgets
Context selection can reduce input length during long-input inference but can remove evidence required to answer a query. We introduce CoRA-Bench, a benchmark for evaluating evidence preservation under shared token budgets. We compare head and tail truncation, BM25, dense retrieval, dense retrieval with reranking, LLMLingua-2, and Provence across BABILong, LOONG Financial Spotlight, a custom NoLiMa diagnostic, and LongMemEval. Evidence-preservation results reveal task-dependent tradeoffs: BM25 achieves 0.921 supporting-fact recall on BABILong at 128K, dense retrieval and dense reranking each achieve 0.908 needle recall on NoLiMa at 64K, and dense retrieval achieves 0.993 answer-bearing-turn recall on LongMemEval at 64K. Evidence, token retention, and complementary reader evaluations use the same selected contexts. Reader evaluations use dataset-native answer metrics and three runs on the same questions, with paired bootstrap confidence intervals. At a 32K budget, BM25 achieves 46.20% BABILong answer accuracy and dense retrieval achieves 71.13% LongMemEval accuracy. The results show that method rankings vary across evidence types and between evidence recall and downstream accuracy.
Read the full abstract
Machine Learning
Implicit Bias in State Space Models and Linear Autoregressive Training
Linear State Space Models (SSMs) and linear autoregressive sequence models reuse the same parameters across a sample-dependent number of steps. As a result, different examples impose constraints through different predictors, even though all predictors share the same weights. We study this phenomenon through a general model of multiple non-homogeneous polynomial predictors with shared parameters, in a classification setting. Under standard interpolation and directional-convergence assumptions, together with a trajectory-dependent effective-degree condition, we show that gradient descent on the exponential loss converges in direction to a KKT point of an effective-degree max-margin problem. In this problem, only examples with minimal effective degree impose hard margin constraints; examples with faster-growing margins are asymptotically inactive except for feasibility. Our result provides a first-order characterization of the implicit bias in linear SSMs and linear autoregressive models. It exposes a mechanism of length bias: a longer input sequence or additional autoregressive steps may increase the example's effective degree, and the examples with the smallest effective degree dominate the limiting classifier. Our result highlights how non-homogeneity and parameter sharing alter classical homogeneous implicit-bias results.
Read the full abstract
Computer Vision
Anatomy-Preserving Unpaired Medical Image Translation via Shared Latent Anchoring
Unpaired medical image-to-image translation requires synthesizing target-modality images from source-modality images without aligned supervision. Existing methods learn a direct source-to-target mapping that entangles structure preservation and appearance generation within a single model, often distorting clinically meaningful anatomy in the translated output. We argue that these two objectives should be decoupled entirely: anatomical structure should be captured independently of generation, and the generator should operate only on a frozen structural representation. To this end, we propose a two-stage framework for unpaired medical image translation. In the first stage, we learn a shared structure space by training a structure encoder-decoder on source images, target images, and structural masks jointly, using mask reconstruction to anchor the latent space to modality-agnostic anatomy. For source images without mask annotations, a target-domain memory bank provides soft structural supervision via prototype retrieval. In the second stage, a conditional flow model learns to render target-domain appearance under fixed structural conditions. This ensures the generator inherits target appearance statistics and can focus entirely on structural conditioning rather than learning appearance from scratch. We validate our framework on OCT-to-OCTA synthesis and brain MR-to-CT translation. Our method outperforms all unpaired baselines on both tasks and, despite requiring no paired supervision, also surpasses paired baselines, improving PSNR by up to 11.2% and SSIM by up to 26.4% on OCT-to-OCTA, and achieving consistent gains across whole-image, soft-tissue, and bone regions on MR-to-CT. The project repository is available on GitHub.
Read the full abstract
Robotics
Agibot: An Open-Source Modular Swerve-Drive Robot Platform for Agricultural Robotics Research and Education
Agricultural production faces persistent labor shortages and increasing demand for automation, yet progress in agricultural robotics remains constrained by the limited availability of affordable, flexible, and accessible research platforms. This paper presents Agibot, an open-source, modular, low-cost agricultural robotics platform designed for autonomy research and experimentation, with potential applications in hands-on robotics education. The platform integrates four-wheel independent swerve drive, swappable computing, distributed embedded control, multimodal sensing, and configurable manipulation within a unified sense-plan-act-verify framework supporting perception, control, navigation, manipulation, and autonomous task verification. The platform was evaluated through representative row-crop navigation and manipulation tasks under varying lighting and soil conditions. Experiments demonstrated reliable lane following, autonomous row-to-row navigation, target identification, seed dispensing, plant manipulation, and closed-loop verification of task success. Its modular architecture supports independent testing, modification, and validation of autonomy components for rapid prototyping and experimentation. These proof-of-concept tasks, controller-tuning studies, and integrated trials demonstrate the platform's research capabilities, while its accessible and reconfigurable design supports its use for hands-on robotics education. By combining affordability, modularity, open-source accessibility, configurable manipulation, and end-to-end sense-plan-act-verify autonomy, Agibot provides a practical platform for agricultural robotics research, technology development, and workforce training.
Read the full abstract
Artificial Intelligence
ScreenProof: Where Is This Screen Wrong? Finding UI Faults Without a Reference
A phone screen can be wrong in ways that need no reference: a label that runs off the screen, letters that touch, a line that breaks one character before its end, one button with sharp corners among round ones. We ask whether a model can box every such fault on a screenshot, given no target design and no reference image. Such a critic could serve as a cheap first pass over generated pages before a designer reviews them, but only if it stays silent on clean pages. ScreenProof poses this target-free critique over three families of faults: general faults, local style faults (one element breaks the style its page sets) and text layout faults (letter spacing, line height, line breaks). A data engine makes each fault in page code and records the changed element's box, and hard negatives change every sibling alike. No frontier judge is reliable: frontier models flag 18-62% of clean real screenshots and 18-82% of clean generated pages, find 47-72% of text layout faults, and rarely find a foreign icon (5-19%). Our small critic ProofJudge (Qwen3.5-9B with LoRA), trained on code-made pages with one to four faults each and keeping a finding only when it is confident, reaches F1 0.48, the highest of all judges (Gemini 4 Argon 0.37; open models at most 0.15; 0.46 even without the confidence cut), and is the only fault-finding judge within 15% false alarms on clean pages (2% real, 9% generated), the property such a first pass needs, at about a hundredth of Argon's cost. It finds the most general faults (59%) but fewer local style faults than Argon (25% against 41%) and fewer text layout faults than the API judges (43%; Gemini 4 Argon 47%). Against a named brand the task is harder still: models know a design language about as well as its public documentation states it.
Read the full abstract
Artificial Intelligence
ProofDiff: Auditable Argument-Level Comparison of AI-Generated Mathematical Proofs
Mathematical understanding depends on the route from hypotheses to conclusion: how the parts of a proof function and which dependencies carry the argument forward. This structure can persist when wording or notation changes, while a small alteration to a hypothesis or inference can produce a different argument beneath similar prose. Formal verification can establish that a rewrite proves the same theorem, but not whether it follows the same mathematical route. Comparing informal proofs, especially those produced or revised by generative models, therefore requires reconstructing which claims, dependencies, and reasoning steps are preserved, reorganized, or replaced. We present ProofDiff, a system for comparing informal mathematical documents. It is designed to support human review, not to certify mathematical correctness. Given two documents, ProofDiff independently constructs a source-grounded argument map for each and proposes correspondences between the resulting maps. Each map captures the document's claims and reasoning steps and records the typed relations that connect them. An integrated human-review layer allows reviewers to revise the maps and correspondences while preserving an auditable history. Comparing the two argument maps provides a basis for inspecting whether the documents follow the same mathematical route and where their claims or dependencies differ. We demonstrate ProofDiff on four illustrative pairs of AI-generated proofs, ranging from a detailed proof and its concise revision to proofs using different mathematical strategies.
Read the full abstract
Materials Science
A Transport-Constrained Gradient-Flow Description of Continuum Dislocation Dynamics
Gradient flow frameworks are fundamental to numerous models of defect and microstructure evolution. This work develops a gradient flow formulation for continuum dislocation dynamics, with the key distinction that the admissible rates are transport-constrained. Using the dislocation density and compatible slip fields as the primary state variables, their admissible rates are restricted by geometric conservation laws. The corresponding transport velocities are then selected variationally through a constrained Onsager principle. The resulting model has both a variational gradient flow structure and consistency with the geometric dislocation conservation statement; existing models of dislocation evolution typically have either one or the other feature. This thermodynamic structure provides a dislocation driving force that includes elastic stress contributions – corresponding to the classical Peach–Koehler force – as well as contributions from the core and nonconvex lattice energies. A simplified version of the model is used to study the energetics and dynamics of an opposite-sign dislocation pair. The lattice energy keeps the cores compact, and an external applied shear either assists annihilation or, above a threshold, separates the pair.
Read the full abstract
General Relativity and Quantum Cosmology
Stealth scalars on prescribed dark-energy backgrounds: reconstruction, branch regularity, and perturbative bounds
We reconstruct homogeneous stealth scalar fields on prescribed Friedmann–Lemaître–Robertson–Walker backgrounds with constant or evolving dark-energy equations of state. The scalar stress tensor vanishes, so the expansion history is supported by an independent background sector. Eliminating the self-interaction potential yields a Riccati equation for the scalar profile. We use it to distinguish exact stealthness from global regularity and from the stronger identification of a term in the scalar stress tensor with the prescribed dark-energy density. We prove that this sector-preserving identification is not dynamically invariant in a mixed flat constant-$w$ cosmology. For radiation-free $w$CDM backgrounds, a hypergeometric representation allows the regular and singular branches to be separated by a background-dependent threshold. At conformal coupling the reconstruction instead reduces to an exact comoving-distance quadrature, valid for arbitrary regular flat histories, including radiation. A CPL phantom-divide crossing does not singularize the background stealth equation, although the perturbations of the prescribed dark-energy sector require an independent treatment. The scalar–tensor description gives luminal high-frequency propagation under a positive effective Planck mass. If the field tracks its cosmological value locally without screening, Cassini constraints strongly suppress the corresponding late-time gravitational responses.
Read the full abstract
Machine Learning
RAD-TFM: Robust and Domain-Adapted Tabular Foundation Models
The development of tabular foundation models (TFMs) has accelerated in recent years, outperforming traditional ML methods on many tabular benchmarks. A key finding is that TFMs can be pretrained entirely on synthetic datasets, opening opportunities to design data generators that encourage desirable model properties. Prior work has mainly focused on crafting high-quality priors over generators to improve overall pretraining performance. Our key insight is that parameterizing the generator distribution enables an adversarial, distributional robustness perspective: during training, we can adapt the generator to emphasize datasets that are particularly challenging for the model, which we formalize by introducing an optimality gap measure. Further, we develop a method to align the synthetic data generation with real-world datasets from a given domain, constraining the adversary to generate realistic data. Together, these algorithms comprise the Robust and Domain-Adapted Tabular Foundation Models (RAD-TFM) pipeline, a model-agnostic adversarial training framework. Applied to the TabPFN V2 classifier, RAD-TFM improves performance across 6 diverse tabular benchmarks, with up to a 10.8 percentage-point improvement in mean normalized AUC over the strongest baseline, using fewer than 100k additional synthetic training datasets, less than 0.1% of the original pretraining data. Applying the same pipeline to TabICL V2 also improves performance across three major benchmarks, demonstrating the generality of the framework. These results highlight a promising new direction for targeted adversarial continued training of TFMs using synthetic data alone.
Read the full abstract
Game Theory
Proportionality in Elections under Feasibility Constraints and Ranked Ballots: The Solid Coalition Refinement Rule
We study elections with ranked ballots and general feasibility constraints, which indicate which sets of candidates are allowed as outcomes. We introduce the first proportionality axiom for this general setting and show that it extends the established notions from specific models that fit our general framework. We extend the Solid Coalition Refinement (SCR) rule to general constraints and prove that it satisfies our axiom exactly for matroid constraints. Although the naive implementation of SCR involves a step that is computationally intractable for weak orders, we show that in practical cases it can be implemented in polynomial time or efficiently in FPT time. We also illustrate limitations of this rule for approval ballots and provide its first empirical evaluation.
Read the full abstract
Cosmology and Nongalactic Astrophysics
FAIR Universe Weak Lensing ML Uncertainty Challenge: Dataset and Cosmological Inference Competition
Weak gravitational lensing is a powerful cosmological probe that enables direct mapping of the matter distribution and precise constraints on the fundamental parameters governing structure formation in the universe. In recent years, machine learning has shown strong potential for extracting cosmological information from weak-lensing data beyond traditional summary statistics. However, progress has been limited by the lack of standardized benchmarks that capture the complexity of the dataset, simulator imperfections, and various systematic biases encountered in realistic analyses. To address this gap, we present the FAIR Universe Weak Lensing Machine Learning Uncertainty Challenge, one of the NeurIPS 2025 competitions, as a benchmark dataset and evaluation study. Our dataset reflects realistic weak-lensing observations, incorporating multiple systematic biases, and thereby providing a controlled testbed for robust cosmological inference under uncertainties. We summarize the main methodological advances developed by the winners of our challenge, including data augmentation with conditional optimal transport, an AI-agent-driven approach for parameter inference, ensemble learning, Vision Transformer-based architectures, and physics-informed feature integration. By unifying the benchmark dataset and the winning solutions, this work provides a community resource for developing and evaluating machine learning methods for weak lensing cosmology and, more broadly, for scientific inference under systematic uncertainties.
Read the full abstract
Astrophysics of Galaxies
Local Pitch Angles of Galactic Structures in Nearby Galaxies with JWST: II. Dust Lanes and Gas Flows in Bar-Dominated Systems
We present a spatially resolved study of local pitch angles in six barred spiral galaxies observed by PHANGS–JWST, focusing on bar-dominated regions where dust lanes are the main structures. We combine a two-dimensional Fourier transform (2D-FT) with a circular-window correlation method to map pitch angles across multiple wavelengths and spatial scales, extending our previous analysis of spiral arms and interarm substructures in disk-dominated galaxies. The dust lanes show relatively large pitch angles, typically $|\alpha|\simeq35^\circ$–$70^\circ$, well above those measured for spiral arms. In several galaxies, the pitch angle changes systematically along the dust lane, with smaller values in the inner regions and larger values farther along the bar, tracing variations in local curvature. The multi-scale analysis shows that larger windows recover the overall geometry of the dust lanes, whereas smaller windows reveal local irregularities and secondary structures. MIRI images generally provide the clearest view of the dust morphology and allow a more precise characterization of local pitch-angle variations, while NIRCam offers a complementary view of the underlying stellar distribution. Unlike spiral-arm pitch angles, which are commonly associated with density-wave patterns, the values measured here characterize the local orientation and curvature of dust lanes associated with bar-driven gas flows. Together with the results of our companion study of disk-dominated galaxies, these findings support a unified, scale-dependent framework in which local pitch angles serve as observational diagnostics of distinct dynamical regimes in nearby galaxies.
Read the full abstract
Systems and Control
Sample-Covariance Regularization for Data-Driven Predictive Control from Frequency Response Functions under Colored Noise
In this work, we investigate how sample covariance obtained from frequency-response functions (FRFs) can be exploited to improve performance of data-driven frequency-domain predictive control (FreePC) in the presence of colored measurement noise. In case of periodic multisine excitation, this covariance is obtained directly from repeated periods and provides a frequency-dependent measure of uncertainty without requiring an additional parametric noise model. To incorporate this information into both indirect and direct predictive control, we develop a frequency-domain subspace predictive controller (FD-SPC) and establish its equivalence to the direct FreePC formulation in the noiseless setting. Regularization terms for both formulations are then motivated by expected prediction-error criteria and constructed from the sample covariance of the FRF estimates. Across all tested noise models and data sizes, regularization consistently reduces the median realized control cost relative to the corresponding controller without covariance regularization, with reductions of up to 35%.
Read the full abstract
Materials Science
Temperature-Driven Evolution of Magnetization Dynamics in Cobalt-Doped Fe$_5$GeTe$_2$
Understanding magnetization dynamics in metallic van der Waals ferromagnets is essential to uncover their magnetic interactions and evaluate their potential for high-frequency spintronics applications. In this study, we investigate the temperature-dependent magnetization dynamics of a single crystal of cobalt-doped $Fe_{5}GeTe_{2}$ (CFGT) using broadband ferromagnetic resonance (FMR) spectroscopy. At high temperatures, the FMR response is dominated by a single broad resonance that shifts toward lower magnetic fields upon cooling. Fits to the frequency dependence of the resonance field and linewidth yield an effective $g$ factor of approximately 2.45 and an effective Gilbert damping parameter of approximately 0.15 near room temperature. The effective magnetic anisotropy evolves strongly with temperature, with an increasingly pronounced easy-plane contribution upon cooling toward 200 K. Below approximately 200 K, an additional resonance emerges and evolves into a distinct two-mode response that persists to the lowest measured temperature of 25 K. The two branches exhibit different temperature dependences of their resonance fields and linewidths, and the emergence of the additional mode is not accompanied by a comparably pronounced anomaly in the static magnetization. These results reveal a dynamical crossover in CFGT that is not readily apparent from bulk magnetometry and highlight the sensitivity of FMR to the magnetic behavior of chemically substituted van der Waals ferromagnets.
Read the full abstract
Computation and Language
Jais 3: Introducing Arabic-centric Agentic AI
We present Jais 3, a fleet of six open, Arabic-centric agentic language models ranging from 0.9B to 375B parameters. Jais 3 is built to complete tasks for Arabic speakers, not only to answer questions in plain text: it is trained to plan multi-step tasks, call tools and Model Context Protocol (MCP) servers, operate a terminal, search the web, and read, edit, and produce PDF, spreadsheet, and presentation files, while communicating with the user in Arabic. Instead of pretraining from scratch, we adapt each model of the fully open K2 Horizon fleet to Arabic through continual pretraining, followed by supervised fine-tuning (SFT), final token preference optimization (FTPO), and reinforcement learning with verifiable rewards (RLVR). We combine natural Arabic text with translated, transliterated, and synthetic Arabic, and replay the base model's own training data to keep its reasoning, agentic, and English abilities. Jais 3 covers Modern Standard Arabic and the major regional dialects, follows instructions with Arabic-specific constraints, and inherits the 512K-token context window of K2 Horizon (all models except the 0.9B). To our knowledge, its 375B model is the largest open Arabic-centric language model to date. For evaluation, we revisit how Arabic models are assessed, moving to fully generative evaluation on audited Arabic benchmarks. Across model sizes, Jais 3 matches or outperforms open-weight models of comparable size on Arabic and agentic benchmarks, with particularly strong agentic capabilities and Arabic instruction following, and far fewer hallucinations. We release the weights and the checkpoints from every training stage of all six models, together with samples of the training data, the data recipes and training configurations, the training logs, the audited benchmarks, and the evaluation code.
Read the full abstract
Machine Learning
Lower-Level Agnostic Bilevel Optimization
Bilevel optimization is a fundamental framework for machine learning problems, where an upper-level (UL) objective depends on the solution of a nested lower-level (LL) problem. Existing bilevel algorithms typically assume that the LL objective is available for evaluation, differentiation, or unrolling. However, this assumption can fail when the LL response is produced by a black-box solver, simulator, adversary, or learned predictor. In this paper, we study LL agnostic bilevel optimization, where the UL learner has access only to an approximate LL solution $\bar{\mathbf{y}}(\mathbf{x})$ and cannot query the LL objective, its gradients, Hessian/Jacobian information, or the procedure that generates the LL response. We propose LeGo-BiO, a coordinate-wise pseudo-hypergradient method that estimates the missing Jacobian $\nabla _{\mathbf{x}}\bar{\mathbf{y}}(\mathbf{x})$ using finite differences of $\bar{\mathbf{y}}(\cdot)$. For each UL coordinate, LeGo-BiO constructs a decision vector subject to 1-sparse updates, thereby isolating the corresponding Jacobian column needed in the UL gradient computation. This avoids the directional-projection limitation of full-parameter response differences and, unlike derivative-free/zeroth-order approximations of the entire UL gradient, preserves the available first-order gradient information with respect to the UL variables for more accurate UL gradient evaluation. We establish convergence guarantees for LeGo-BiO in both deterministic and stochastic settings under an LL error-bound/Polyak-Lojasiewicz-type condition, without requiring LL strong convexity. Our bounds yield an $O(T^{-1})$ rate in the deterministic setting and an $O(T^{-1/2})$ rate in the stochastic setting when the LL approximation error decreases geometrically. Experiments demonstrate competitive performance of LeGo-BiO despite being agnostic to the LL objective.
Read the full abstract
Astrophysics of Galaxies
Local Pitch Angles of Galactic Structures in Nearby Galaxies with JWST: I. Spiral Arms and Interarm Substructures in Disk-Dominated Systems
We present a spatially resolved study of spiral arms and feather substructures in seven nearby galaxies observed with JWST. Using eight images per galaxy, from four NIRCam and four MIRI filters, we combine a two-dimensional Fourier transform (2D-FT) with a circular-window correlation method to measure both global and local pitch angles. The 2D-FT identifies the dominant azimuthal mode and the corresponding $p_{\rm max}$, allowing us to reconstruct the spiral pattern and define arm and interarm masks. We then apply cosine filters with different pitch angles and window sizes in the $(\ln (R),\theta)$ plane to map the pitch-angle distribution across multiple spatial scales. NIRCam images provide the most robust measurements of the main spiral pattern, whereas MIRI data reveal dusty interarm features, particularly feathers, with greater contrast at the smallest window size ($D_w=0.25$). Across the sample, the main arms typically have $|\alpha|\simeq15^\circ$–$30^\circ$, while feathers show systematically larger values, often above $50^\circ$. These results demonstrate that pitch angle is scale dependent and that different structural components within the same galaxy can occupy distinct angular regimes. NIRCam and MIRI therefore offer complementary views of spiral structure, with the former tracing the large-scale stellar backbone and the latter exposing finer dust substructures. Together with the companion study of barred galaxies, this work provides a common framework for using local pitch angles to investigate distinct dynamical environments in nearby galaxies.
Read the full abstract
Populations and Evolution
Turing Instability of a Closed Nutrient-Phytoplankton-Zooplankton-Detritus Model
We study Turing instability in a closed Nutrient-Phytoplankton-Zooplankton-Detritus (NPZD) ecosystem model with nutrient recycling, formulated as a reaction-diffusion system without an explicit time-delay. The detritus compartment represents the biomass during the recycling process and acts to introduce a time delay in this process. Total biomass in the system is conserved, leading to a closed model. Although spatial diffusion typically promotes stability, our focus is on understanding how different diffusion rates for different species can destabilize steady states and give rise to spatial patterns. To address this, we first perform linear stability analysis to identify conditions under which Turing instability may occur. We then validate these theoretical predictions through numerical simulations. Our approach develops both intuition and rigorous analysis of how diffusion influences instabilities in the context of a conservative system, offering a comprehensive framework for understanding pattern formation in multi-species plankton ecosystems.
Read the full abstract
Solar and Stellar Astrophysics
A panchromatic photometric and spectroscopic modeling of SN1987A from 200 to 800d
Type II supernovae (SNe) at 1-3yr after explosion can document not only the preSN mass loss history of massive stars but also the chemistry of the inner ejecta rich in metals and eventually molecules and dust. With the aim of modeling SN1987A, we present a grid of radiative-transfer models of Type II SNe at Zsun/3, covering an initial mass range between 11 and 21Msun, a fixed ejecta kinetic energy of 1.2e51erg, and ~0.08Msun of 56Ni. The influence of metallicity at such late times is found to be relatively weak. In this model set, the [OI]6330A flux at 1-2yr varies non-monotonically with Minit. It first rises with Minit due to the growing O abundance but drops for M>17-19Msun due to the higher ejecta density, with little change to the rest of the optical spectrum. Assuming CO molecules account for 90% of the cooling in the C/O shell leads to a 10-20% reduction in the [OI] flux, equivalent to an O(1Msun) reduction in Minit at M<19Msun. The fractional [OI] flux rises from 200 to ~600d but decreases beyond. We obtain a good agreement to SN1987A for all Minit apart from [OI], which appears in the optical as the only discriminant between models. The 13Msun model yields the best match, from UV to IR, but a higher Minit may result with allowance for strong CO and SiO cooling. Past 500d, a growing amount of dust below ~2000km/s is essential to account for the ~1.5mag optical excess fading relative to predictions from decay power. Adopting 0.1mic silicate grains, we find a dust mass of 1e-4Msun at 600d rising to a few 1e-3Msun at 700-800d. We also found that the 200-800d International Ultraviolet Explorer spectra of SN1987A are all contaminated by a background hot star, which also causes the relative UV brightening past 500d. Such contamination may complicate the interpretation of more distant SNe II at >~2yr once or if they fade to O(1e5Lsun) [abridged].
Read the full abstract
Artificial Intelligence
Multi-GPU MoE Inference with Adaptive Expert Offloading
Modern Mixture-of-Experts (MoE) inference is constrained by GPU memory. This constraint is mitigated by expert offloading, which stores part of the model's weight in host memory and executes on demand. Though MoE architecture has computation sparsity, more GPUs are needed for serving increasingly larger frontier models. However, existing expert offloading systems overlook this need and fail to scale performance with additional GPUs. We present AdaOffload, a general and efficient multi-GPU hybrid expert execution framework that effectively handles heterogeneous hardware and workloads. We co-design GPU-CPU parallelism and GPU-CPU co-execution for decode and prefill with online workload observation that adapts to different hardware, workload, and batch size heterogeneity. We evaluated AdaOffload on five different platforms with varying GPU capability (L4, Pro6000, B200), number of GPUs, CPU capability, and NUMA topology on three different frontier open-weight models (284B to 753B). AdaOffload achieves $1.9\times$ to $15.5\times$ throughput speedups across different models over state-of-the-art systems.
Read the full abstract
High Energy Physics (Phenomenology)
$B\to$ invisible decays as probes of light new physics
Rare $B$ decays with missing energy probe heavy particles contributing to $b\to s\nu\bar\nu$ and new light states produced on shell. We discuss both possibilities in light of the Belle II excess in $B^+\to K^+\Emiss$. Fits to the missing-mass distribution favour a two-body contribution near $2.1\,\GeV$, while scalar and fermion pairs of masses around $0.4$-$0.6\,\GeV$ remain viable alternatives. Extracting their couplings would profit from additional measurements, including the $B\to K^*\Emiss$ rate and longitudinal $K^*$ polarization, together with invisible $B_s$ decays which constrain scalar-pair interactions. We discuss how an aligned Higgsed $U(1)'$ model can realize the single-vector explanation through a vector-like top partner. Loop-induced electroweak mixing makes visible $Z'$ decays unavoidable, requiring an additional invisible channel to evade dimuon searches. We show that a light dark fermion can provide this channel and also account for the observed dark matter abundance through resonant thermal annihilation.
Read the full abstract
Machine Learning
Rethinking How to Leverage Cheap Physics for Data-Efficient Neural Operators
Neural operators learn mappings from PDE inputs, such as initial conditions and coefficient fields, to the corresponding solutions. Once trained, they can rapidly predict solutions for new inputs. However, obtaining input-solution pairs for supervised training often requires expensive numerical simulations, whereas PDE inputs and governing equations provide inexpensive physical information. This raises a question: how can cheap physical information reduce the need for labeled solutions? A natural approach is to use physics to select inputs for labeling, but our acquisition experiments show no consistent advantage over random sampling. This observation motivates exploring an alternative: using cheap physics to improve neural operator initialization before supervised training. We propose Physics-Informed Initialization (PI-Init), a simple yet effective method that pretrains neural operators on unlabeled inputs using physics-informed losses and retains the learned parameters for supervised fine-tuning on the labeled dataset. A linearized analysis shows that pretraining can improve downstream prediction accuracy by reducing initialization errors that limited labeled data cannot effectively correct. Experiments across diverse PDE benchmarks and neural operator architectures demonstrate improvements in predictive accuracy, out-of-distribution generalization, and label efficiency. Representative results show error reductions of up to 62.7% at the same label budget and comparable or better accuracy with 90% fewer labels, relative to supervised-only training.