Cosmology and Nongalactic Astrophysics
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Friday 11 September
We present a fully nonlinear relativistic description of the observed pulsar timing modulations, defined through the ratio of the proper-time intervals between two successive emissions of radio pulses and their observations by a pulsar timing array. Working in the limit of a vanishing emission interval, we derive the observed pulsar timing modulation to second order in relativistic perturbation theory around a Minkowski background, without fixing a gauge and retaining the full scalar, vector, and tensor content of the metric perturbations. Although the derivation naturally introduces several gauge-dependent intermediate quantities, we explicitly demonstrate that they combine into a coordinate-independent expression. Moreover, we explicitly show that the second-order timing modulation from two neighboring radio pulses coincides with the observed redshift, defined as the fractional change in wavelength along a single geodesic. The agreement represents a nontrivial second-order realization of the exact nonlinear equivalence proven in Magi & Yoo 2026.
Cosmology and Nongalactic Astrophysics
The Loud Tail of the Supermassive Black Hole Binary Population: Multimessenger Candidates and Prospects for SKAO
We assess whether electromagnetically identified supermassive black hole (SMBH) binary candidates populate the rare, high-amplitude tail of the gravitational wave (GW) amplitude distribution. Using an SMBH binary population model fitted to the NANOGrav 15-year data, we find that 3C~66B and Mrk~501 are likely to stand out from the GW background, making them promising targets for individual-source searches. Furthermore, we show that individual detections can probe the binary hardening mechanism, as models with strong environmental hardening predict a higher abundance of individually resolvable binaries than purely GW-driven models. Applying the same population fits and accounting for the confusion noise from the unresolved population, we forecast that the Square Kilometre Array Observatory (SKAO) will resolve tens of individual binaries, with 3C~66B robustly detectable.
Read the full abstract
Analysis of PDEs
Nondegeneracy of the Sadovskii vortex
We prove nondegeneracy, modulo horizontal translation, of the normalized Sadovskii vortex patch. Building on the free-boundary and root-map framework developed by Huang and Tong, we first obtain a precise quantitative approximation of an exact Sadovskii profile, trapping its free boundary between two explicit barriers constructed around a rational approximation. This localization is then used to control the natural weighted quadratic form at the touching configuration. In the odd class, a ground-state identity shows that horizontal translation is the only zero mode. In the even class, dilation gives a negative direction, while a finite-rank comparison on a compact part of the boundary together with analytic contact-tail estimates yields a positive gap on a codimension-one subspace. Consequently, the Sadovskii dipole is locally rigid at the linearized level once translation is fixed. The computer-assisted part is confined to rigorous interval verification of a finite collection of explicit inequalities; the continuum, contact and operator reductions are proved analytically.
Read the full abstract
Combinatorics
Matroids with cycle systems are regular
A cycle system for a matroid $M$ is a collection of cycles (unions of circuits) whose intersection properties mimic the cut sets of a graph. Cycle systems were introduced by Corry, the first author, McClain, Perkinson, and Yi, who showed that the $h$-vector of any matroid that admits a cycle system is a pure $O$-sequence, confirming a conjecture of Stanley for this class. Those authors also conjectured that any matroid admitting a cycle system must be binary. Here we answer this conjecture in the affirmative, and prove the stronger result that any such matroid must in fact be regular (representable over any field). From this, we conclude that if $M$ is connected, every cycle system for $M$ is a basis for its circuit space of. We establish other properties of cycle systems along the way, which may be of independent interest.
Read the full abstract
Quantum Physics
Structured Quantum Kernels for Chaotic Forecasting
Quantum kernels promise exponentially large feature spaces, but expressive circuits render their Gram matrices uninformative, bandwidth tuning collapses them toward classical RBF, and the practical consensus is that quantum kernels add nothing on classical data. We answer another productive question, an architectural one: whether the structure of an encoding circuit can carry an inductive bias that tuned classical kernels lack. We introduce AeRot, a quantum kernel fusing amplitude encoding of l2-normalized delay windows with a grouped single-qubit rotation layer that assigns contiguous temporal blocks to each qubit, making the circuit delay-window-aware. On Lorenz-63 kernel ridge regression with cross-validated bandwidths over 100 seeds, AeRot outperforms tuned RBF and Matern-5/2, the advantage emerging at physical horizons >= 0.15 tu and reaching +0.137 mean R^2 at 0.25 tu (5 qubits, window 32). Linear-stability analysis of the window tail localises the advantage to the unstable saddle-approach regime: AeRot wins 83% of windows in the most unstable decile, with the win rate rising monotonically with tail instability and a sign flip at the local stability boundary. The difficulty is fold-branch ambiguity: trajectories approaching the saddle are locally diverging, and Euclidean kernels struggle to resolve which lobe the trajectory will commit to. Structural diagnostics confirm Gram matrices structurally distinct from tuned RBF, with strictly higher target-kernel alignment at every horizon. The gain is architectural: a finite-sample inductive-bias effect of temporally structured encoding on a folded attractor, with no computational-separation claim attached. To our knowledge this is the first mechanistic localisation of quantum-kernel advantage to a specific dynamical regime of a classical system.
Read the full abstract
Logic
Medvedev logic is undecidable
We show that Medvedev's logic of finite problems, a well-known superintuitionistic logic, is undecidable. The key method is a reduction from the periodic tiling problem to non-theoremhood in Medvedev's logic. This settles a longstanding open problem. Using similar techniques, but reducing instead to the ordinary tiling problem, we likewise obtain undecidability of Skvortsov's logic of infinite problems, and the fact that the two logics are distinct – in fact, they are separated by any aperiodic tiling of the plane. Due to the fact that Medvedev's logic figures in so many different areas, these results have implications for several fields – for example, the study of schematic fragments of logics such as propositional dependence logic, or the study of internal logics of toposes. The core idea and technical work of the undecidability proof were obtained using ChatGPT Sol 5.6, and formally verified in Lean by Claude Opus 5. A detailed methodology section outlines how such results were obtained.
Read the full abstract
Analysis of PDEs
Hadamard states for spacetimes with timelike boundaries
We formulate and prove existence of pure quasifree Hadamard states for the Klein–Gordon field with Dirichlet boundary condition on a globally hyperbolic spacetime with timelike boundary. The microlocal condition is imposed on the one-particle Hilbert-space-valued distribution $p\circ G_D$ rather than on the two-point function. This is expressed by a $b$-wave-front set condition on the compressed future cone. The analysis on the manifold with corners $\bar M\times\bar M$ is then almost completely avoided. We prove that our version of the Hadamard condition is universal in the class of quasi-free Hadamard states. We also prove a propagation theorem for positive-energy microlocal splittings which uses only a microlocal splitting in the small $b$-calculus and well-posedness of the Dirichlet problem, and in particular requires no propagation theorem for reflected or gliding rays; combined with an explicit computation in the ultrastatic case and a Fulling–Narcowich–Wald deformation argument this yields the construction of a pure quasi-free Dirichlet–Hadamard state.
Read the full abstract
Chemical Physics
Can Autonomous LLM Agents Execute Multireference Quantum Chemistry Calculations?
Multireference electronic-structure calculations remain difficult to automate because critical workflow decisions, including active-space selection, state averaging, convergence recovery, and state identification, traditionally rely on expert judgment. Here, we investigate whether an autonomous large language model (LLM) agent can perform these tasks without human intervention. The agent selects active spaces using literature-grounded analogies or explicitly documented chemical reasoning, generates and submits ORCA calculations, analyzes outputs, and records all decisions in an auditable reasoning log. Benchmarking against 558 vertical transition energies (VTEs) from QUESTDB shows that an unguided baseline agent achieves 24.9% coverage with a mean absolute error (MAE) of 0.373 eV. Introducing a structured decision ladder increases coverage to 44.1% while reducing the MAE to 0.339 eV. The largest gains are observed for double and Rydberg excitations, demonstrating that expert-informed procedural guidance substantially improves active-space construction and state identification. When provided with complete workflow information, the agent successfully reproduces published QUEST calculations with an MAE of only 23 meV, resolving 75% of target configurations within seven attempts. These results demonstrate that contemporary LLM agents can autonomously execute and reproduce complex multireference quantum-chemical workflows, while highlighting the importance of structured reasoning frameworks for achieving reliable high-throughput and high-fidelity electronic-structure calculations.
Read the full abstract
Artificial Intelligence
ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search
In this work, we present ZGCM-1, a fully open 7B dense foundation model trained from scratch with extreme data, system, and algorithmic efficiency. ZGCM-1 is founded on a core premise: compact models cannot passively memorize the open web, but can overcome parametric capacity limits by coupling deliberate internal thinking with active external tool use. To support this paradigm across a 256K context, we develop an end-to-end, high-efficiency open training recipe: Architecture & System Co-design: interleaved gated sliding-window and full attention, and a stable FP8 Muon optimizer; Progressive Curriculum & MDP Mid-Training: context scaling across 16K, 64K, and 256K, and the reformulation of interaction traces into Markov Decision Processes. Furthermore, we establish an AI-native R&D workflow where agent swarms autonomously manage cluster operations, data curation, and rapid diagnostic evaluation. Extensive evaluations show that ZGCM-1-7B is competitive across 7B model family on general benchmarks. On several challenging mathematical reasoning and agentic search suites, it remains competitive with frontier models orders of magnitude larger, such as Qwen3-235B-A22B and GLM-5.1. We also show that our pre-training design offers a ~4.2x efficiency improvement in 16K pre-training time-to-loss. Across the full development lifecycle, we distill eight actionable empirical findings-spanning architectural scaling, SFT quality pruning, long-context generalization, and agentic co-training dynamics. To facilitate community research, we open-source model weights from the pre-training, mid-training, and post-training stages, intermediate checkpoints, training code, per-stage data and data recipes, and W&B logs.
Read the full abstract
Computers and Society
LOCO 2026 Lightning Talk Abstracts: 2nd International Workshop on Low Carbon Computing
This volume contains the accepted lightning-talk contributions from the 2nd International Workshop on Low Carbon Computing (LOCO 2026), held at Lancaster University, United Kingdom, on 10-11 September 2026. The collection brings together 14 short papers presenting emerging research, early-stage results, practical tools, and critical perspectives on reducing the environmental impacts of computing. Topics include carbon and energy measurement, carbon-aware scheduling, sustainable artificial intelligence and machine learning, high-performance computing, computational sufficiency, hardware longevity, interoperability, frugal computing, and post-growth approaches to digital technology. The contributions were selected by the LOCO 2026 Programme Committee for their relevance to the workshop and their potential to stimulate discussion across technical, organisational, and societal dimensions of low-carbon computing. They are published together in this edited volume as a companion collection to the main LOCO 2026 proceedings index.
Read the full abstract
Optics
Momentum microscopy of ultrafast electron emission from a strongly driven optical nanoantenna
Metallic nanostructures in combination with femtosecond lasers are a well-suited platform for the control of photoelectrons by strong and nano-localized driving fields, with high relevance for ultrafast, coherent electron emitters and petahertz electronics. Photoelectron spectroscopy resolves such photoelectron dynamics, but lacks nanoscale spatial resolution, making it only suited for single emitters or homogeneous arrays. We report the first strong-field experiment combining both photoemission electron microscopy and momentum microscopy, two complementary techniques providing spatial and momentum resolution within the same instrument. We apply this new methodology to a double-hole nanoantenna with sub-10 nm apex radii, demonstrating the potential of this approach for the control of photoelectrons in heterogeneous nanostructured samples using few-cycle light fields. Our measurements reveal distinct signatures of two classes of electron trajectories in the near-field. Quiver trajectories result in directed, angularly more focused emission, whereas subcycle trajectories give rise to a broader transverse momentum distribution. Surprisingly, this observation disagrees with previously reported emission characteristics from nanotip emitters, which we classify as a special case of a broader class of curved emitter surfaces driven by ultrashort light fields. This demonstrates both the impact of sophisticated electron detection methods and the potential of strong-field control of electrons in nanoscale geometries.
Read the full abstract
Cryptography and Security
SkillAtlas: An Attack Trace Library for Agent Skills
Agent skills are reusable units for language-model agents, but their risks emerge through model decisions, user context, tool calls, and execution feedback rather than through stable signatures or a single sandbox run. Existing static, dynamic, and benchmark-style evaluations rarely preserve public evidence that can be inspected, searched, and reused. We present SkillAtlas, a hosted attack trace library that converts private agent-skill security report bundles into reviewed, redacted, and searchable public cases. The library contains 3,014 cases, 6,589 traces, 151,131 steps, 233 affected skills, and 8 risk categories; 42.5% of successful cases first become successful after a non-success initial round, and trajectory-grounded labels improve pre-execution guard accuracy to 0.770.
Read the full abstract
Human-Computer Interaction
Real-time Learning and Evolution in Robotic Art Installations
We present three robotic art installations which explore the aesthetics of adaptive behavior. Through embodied machine leaning and digital evolution, these works draw viewers into an artificial ecosystem in which open-ended novelty, trial-and-error learning, competition, and cooperation emerge in real time. Research-creation practices are examined in relation to these works, focusing on how they redefine the role of artists within a human-machine collective while examining points of convergence and divergence between artistic and engineering approaches to adaptive robotics. The systems in question use learning and evolutionary processes not as a means to optimize a specific solution, but as an aesthetic experience on its own, suggesting new modes of interdisciplinary art-science research. Finally, we discuss strategies and practices to elevate the aesthetic experience for audiences, including contexts of presentation as well as temporal and material considerations for artworks based on embodied adaptive systems.
Read the full abstract
Information Theory
A New Upper Bound on the Binary Deletion Channel Capacity
We prove that the capacity of the binary deletion channel satisfies $C(d)\le (1-d)/4$ for every $13/20\le d<1$. The proof describes the output from right to left, using a six-bit context to assign a description length. We bound the increase in expected description length minus output entropy when one input bit is added. A relative-entropy identity reduces this bound to finitely many linear inequalities. A potential on input windows of length 26 makes the inequalities telescope, giving a bound for every input word. Deletion composition extends the result from $d=13/20$ to all larger deletion probabilities. We also obtain finite-block bounds on mutual information and decoding error, with explicit $O(\log n/n)$ corrections. The finite certificate is checked using exact integer arithmetic, and the proof is formalized end to end in Lean.
Read the full abstract
Combinatorics
Avoiding four-term progressions from finitely many starts
For every finite set $A\subset\mathbb{Z}$, we construct a bijection $p:\mathbb{N}_0\to\mathbb{Z}$ beginning with $0,1$ that contains no four-term arithmetic progression, in occurrence order, whose first value belongs to $A$. More generally, for each $n\ge2$ the enumeration can begin with $0,1,3,\ldots,2^{n-1}-1$ and simultaneously avoid every such progression starting in $A$ or among these first $n$ entries. The construction uses finitely branching prerequisite relations and an explicit bounded integer potential. This proves finite prerequisite closure and gives an exhaustive enumeration, rather than merely a total order. We also give an extension theorem for finite prefixes with compatible binary-tail constraints. These results do not determine whether every enumeration of the integers contains an ordered four-term arithmetic progression.
Read the full abstract
Populations and Evolution
Speciation, extinction and explosions induced by sudden environmental changes
We study the evolution of sexually reproducing populations subjected to sudden environmental changes. Using an individual based model we consider scenarios where abrupt changes can affect the carrying capacity, the mating range or the strength of assortativity, that controls the minimum similarity between individuals that allows mating to occur. We show that a boost in carrying capacity can either lead to new speciation events, thereby increasing biodiversity, or to widespread extinctions, causing the system to become dominated by a single species. A similar effect is observed with respect to the mating range. We show that the key factor determining the outcome of the process is genome length, i.e., the number of genes controlling reproduction and assortativity. We also show that when mating becomes more restrictive, increasing the degree of assortativity, a burst of speciation occurs. The explosion lasts only for a few generations, converging to values larger than before the change, but not as high as at the peak.
Read the full abstract
Image and Video Processing
PiMiX 2.02: Toward AI-Driven Data Fusion in Radiographic Imaging and Tomography
PiMiX (Physics-informed Meta-instrument for eXperiments) was introduced for multi-instrument, multi-experiment, and simulation-experiment data fusion in radiographic imaging and tomography (RadIT). Here we present PiMiX 2.02 as an evolving AI-enhanced cyber-physical meta-instrument integrating imaging sensors, near-sensor computing, data fusion, physics-informed inference, and human-supervised AI workflows across X-ray, neutron, and other modalities. Demonstrated capabilities include multimodal CMOS radiation imaging, simulation-assisted sub-pixel neutron localization, and edge-deployed optical-neural-network (ONN) inference; GPU and ONN implementations achieved greater than 96% precision for neutron-event detection with sub-micron localization. A further advance is human-in-the-loop agentic-AI co-analysis of X-ray and neutron images from inertial-confinement-fusion experiments. Beyond conventional preprocessing, the workflow generates competing feature hypotheses, ranks contours using physics-informed evidence, estimates confidence, and presents alternatives for human review. The same architecture adapts to different physics: X-ray analysis emphasizes dark, nonuniform ring structures using deformable closed paths and multi-scale evidence, whereas neutron analysis targets bright emission envelopes using fractional-emission levels, intensity gradients, and cross-filter persistence. We also highlight automated comparison of an as-designed stereolithography model with an X-ray CT reconstruction of an additively manufactured metal lattice. Together, these examples show progression from AI assistance in specific processing tasks to multi-task, multi-domain scientific co-analysis. PiMiX 2.0 further provides a pathway toward PRISM, a RadIT scientific foundation model, and tighter integration of diagnostics, digital representations, inference, and experimental control.
Read the full abstract
Probability
A Dynamic Central Limit Theorem for Mean-Field Quantum Filtering
We study Open Problem 3 of Kolokoltsov's survey "Quantum filtering and propagation of chaos for open quantum systems" (2026) on $H=\mathbb{C}^d$ ($2\le d<\infty$ fixed): a dynamic central limit theorem for $N$ mean-field coupled quantum particles under continuous diffusive (homodyne) measurement. The tagged conditional state $\Gamma^{(1)}_t$ is compared with the one-particle filter $\gamma_t$ driven by its own innovation, the mean-field Hamiltonian evaluated at the deterministic Hartree-Lindblad mean $\eta_t$. The fluctuation $F^N_t=\sqrt{N}(\Gamma^{(1)}_t-\gamma_t)$ is not autonomous in the limit: it is driven through the linearised filter by its own innovation, by an idiosyncratic Gaussian martingale carried by the connected correlations with the other particles, and by the empirical field $\hat G^N_t=\sqrt{N}(\Gamma_t-\eta_t)$, which itself satisfies an autonomous linear Gaussian equation, so that $(F,\hat G)$ solves a closed system. Unconditionally, $(F^N,\hat G^N)$ is tight, every limit point solves this system driven by a Brownian motion orthogonal to a square-integrable martingale pair, and the mean covariance of the idiosyncratic martingale is identified as the limit of the rescaled pair correlations. The remaining clauses, concerning signs rather than amplitudes, reduce to a single decorrelation statement fixing the law of the driving pair. It survives in a conditional form, with the common response to the empirical field and the tagged-slot component projected out; without those projections it is false for generic data, which is what fixes them. The conditional form is proved at leading order on a shrinking horizon and from below at its exact orders on every horizon; it is needed only for the concentration of the driving covariances, the Gaussianity of the pair, and uniqueness, and what remains open is one-sided and above the leading order.
Read the full abstract
Machine Learning (Statistics)
Beyond Point Forecasts: A Survey on Probabilistic Forecasting for Time Series and Spatiotemporal Data
Probabilistic forecasting is central to decision-making under uncertainty, yet its methodological landscape has become increasingly fragmented across temporal and spatiotemporal forecasting, statistical modeling, machine learning, and deep generative modeling. This survey develops a unified perspective by organizing probabilistic forecasting methods according to where and how uncertainty is introduced into the forecasting pipeline. Our taxonomy connects model-agnostic approaches including ensembles and distribution-free calibration, with model-intrinsic approaches spanning Bayesian modeling, parametric predictive distributions, distributional regression, and modern generative models, and further examines the emerging role of time series foundation models. Beyond methodological synthesis, we identify the assumptions, computational demands, and forms of uncertainty represented by different paradigms, and translate these distinctions into data-driven and domain-specific guidance for method selection. We complement the survey with a cross-paradigm empirical study on univariate, multivariate, and spatiotemporal forecasting tasks. The results reveal that no single uncertainty-quantification paradigm dominates across settings. Calibration, sharpness, predictive accuracy, and computational efficiency can lead to substantially different model preferences, while expressive generative models and zero-shot foundation models exhibit markedly different accuracy-efficiency trade-offs. Lastly, we identify unresolved challenges surrounding uncertainty in evolving dependency structures, physics-informed predictive distributions, forecasting extreme events, handling count-valued, directional, and continuous-time series, and the development of unified software resources. Our survey provides both a conceptual framework and a practical roadmap for probabilistic forecasting research.
Read the full abstract
General Physics
The geometrization of electromagnetism
Previous works [1,2,3] expose a synthetic, Riemannian geometrization of electromagnetism that emerges naturally from the experimental bases provided by the Lorentz force and the Maxwell equations. Since the components of the associated space-time metric tensor depend on the four-vector velocity, Riemannian geometry does not fully describe the space-time geometry. The present work aims to provide a suitable geometrical basis for the problem, applying a pseudo-Finsler geometry – a Finsler geometry with a non-homogeneous norm.
Read the full abstract
Machine Learning (Statistics)
Stochastic Gradient Descent over P2
Stochastic gradient descent (SGD) admits diffusion approximations that replace the complicated randomness of stochastic gradients by Gaussian noise, providing a powerful tool for understanding its dynamics and long-time behavior. We investigate whether an analogous approximation principle holds for optimization over probability measures, where the objective is a functional defined on the Wasserstein space P2. The nonlinear geometry and infinite-dimensional nature of P2 prevent a direct extension of the classical Euclidean theory. Using Lions differentiability, we lift the problem to a linear Hilbert space, where higher-order differential calculus becomes available. We then construct a Gaussian random-field approximation whose velocity field matches the mean and covariance of the original stochastic gradient. By exploiting this moment matching through higher-order Taylor expansions, we show that the Gaussian approximation captures the SGD dynamics with second-order weak accuracy. Our result provides a rigorous foundation for replacing sample-driven randomness by analytically tractable Gaussian fluctuations in stochastic optimization over probability measures.
Read the full abstract
Functional Analysis
Improved Bounds For The Lipschitz Retraction Constant In Classical Banach Spaces
We improve the upper and lower bounds for the smallest Lipschitz constant K for which there exists a K-Lipschitz retraction from the unit ball to the unit sphere of the infinite-dimensional Banach space X, for several classical spaces X.
Populations and Evolution
GrassTop: Grassmannian k-mer Topology for Viral Classification and Phylogenetic Analysis
We introduce GrassTop, a genome representation that integrates Grassmann manifolds and algebraic topology for viral classification and phylogenetic analysis. The framework begins by constructing multiscale topological and spectral descriptors of (k)-mer positional patterns. It then extracts a low-rank subspace that summarizes variation across the filtration and compares genomes using a Grassmannian distance. Although the reported implementation uses the chordal distance, the framework is not restricted to this particular choice. We evaluate GrassTop on four families of viral classification datasets, four phylogenetic clustering datasets, and a sequence perturbation experiment. Under the reported 5-nearest-neighbor protocol, GrassTop achieves higher scores than five published alignment-free reference methods across all reported classification metrics and datasets. Its UPGMA (unweighted pair-group method using arithmetic averages) trees achieve an average label purity of 1.0 on every phylogenetic dataset. The perturbation experiment provides a more nuanced result: the subspace representation differs most clearly from direct comparison of the unprojected feature matrices for SARS-CoV-2, whereas the differences are smaller or non-monotonic for the other datasets. Overall, these results support GrassTop as an effective topological-geometric representation for viral classification and phylogenetic analysis.
Read the full abstract
Computer Vision
ProtoCAM: Interpretable Few-Shot Mask-Guided Prototypical Learning for Breast Lesion Classification in Ultrasound Imaging
Breast ultrasound imaging plays an important role in the early detection and diagnosis of breast cancer, particularly for patients with dense breast tissue. However, developing reliable deep learning models for ultrasound analysis is challenging due to limited annotated medical data and the need for interpretable predictions. To address these challenges, this paper proposes ProtoCAM, an explainable few-shot learning framework for breast lesion classification that integrates mask-guided feature encoding, prototypical metric learning, and gradient-based visual explanations. The proposed approach leverages lesion masks to guide feature extraction and constructs class prototypes within an embedding space to enable robust classification under limited training samples. The framework was evaluated on the BUSI dataset using a stratified group k-fold cross-validation protocol to prevent patient-level data leakage. Experimental results demonstrate ProtoCAM's high performance in low-data scenarios. In a 3-way 5-shot setting, the proposed method achieves a macro F1-score of 0.910, representing a substantial improvement over standard supervised CNN models. Among the evaluated backbone networks, ResNet18 achieved the best performance, reaching a macro F1-score of 91.65% under a 15-shot configuration, providing interpretable insights into the classification decisions. These results highlight the potential of explainable few-shot learning frameworks for reliable computer-aided breast cancer diagnosis in data-scarce medical imaging environments.
Read the full abstract
Information Retrieval
PCGNet: Unifying Shared and Specific Information for Fashion Matching Recommendations
In fashion domain, recommending complementary clothing items that match selected pieces is a crucial cross-selling technique that improves customer satisfaction. Nevertheless, fashion matching presents significant challenges, as recommendations must not only align with individual user fashion preferences but also ensure compatibility between garments. These challenges are twofold. First, existing models often assume an overly simplified decoupled relationship between product compatibility and personalized user preferences, overlooking the natural complexity between the two. Second, existing data-driven approaches are not optimized for real-world fashion data, which is typically sparse and characterized by noisy interactions. To address these challenges, we propose Personalized Compatibility Graph Network, a multi-objective graph learning framework that organically unifies the modeling of product compatibility and personal preferences. PCGNet uses contrastive mutual information maximization to extract and align shared and view-specific patterns, thereby capturing the complex interplay between compatibility and personal preferences. Moreover, we introduce a correlation-aware neighbor sampling and a learnable global graph augmentation, which enhance the model by incorporating self-supervised signals mined directly from the graph, ensuring more stable and informative representations. Finally, PCGNet generates recommendation scores through the joint optimization of BPR ranking loss and multi-view mutual information losses. Experimental validation on two benchmark datasets demonstrates that PCGNet significantly outperforming state-of-the-art methods across all four evaluation metrics.
Read the full abstract
Number Theory
The Strong 24-Conjecture
Restricted forms of Lagrange's four-square theorem, in which the representing variables are required to satisfy additional arithmetic conditions, have been studied from several points of view. Among the refinements proposed by Zhi-Wei Sun is the conjecture that every nonnegative integer has a representation \[ N=x^2+y^2+z^2+w^2,\qquad x,y,z,w\in\mathbb Z_{\ge0}, \] for which \[ x\ \text{and}\ x+24y\ \text{are perfect squares}. \] We prove this conjecture for all sufficiently large integers. The proof starts from an elementary parametrization of the two square conditions and combines a semi-linear two-squares sieve with Poisson summation, finite-field cancellation, and switching.
Read the full abstract
Algebraic Geometry
Stringy invariants for abelian character varieties
We compute all stringy invariants, encoding orbifold cohomology numbers, of (the normalization of) the identity component of $G$-character varieties of free abelian groups and of moduli spaces of $G$-Higgs bundles on abelian varieties, for all complex connected reductive groups $G$. As an application, we provide a direct proof of a topological mirror symmetry statement: the equality of the stringy invariants for Langlands dual groups. In the framework of Ginzburg–Kaledin local resolutions, we discuss the meaning of these stringy invariants, showing that symplectic resolutions of these moduli spaces can only exist in the cases of groups of Dynkin type $A$, $B$, or $C$. When such resolutions exist, the computed stringy Hodge numbers agree with the actual Hodge numbers of the resolution.
Read the full abstract
Data Structures and Algorithms
NOC NOC, who's there? Clustering systems of tree-child and normal networks
Clustering systems provide a natural way to encode structural information contained in phylogenetic networks. In this note, we study the clustering systems of normal and tree-child networks through an overlap-based property of set systems, called not-overlap-covered (NOC). We show that the NOC property is equivalent to inclusion-visibility, a memberwise formulation of the strict-compatibility condition previously used for tree-child clustering systems. We characterize normal networks as precisely the semi-regular networks whose clustering systems satisfy NOC. Consequently, a clustering system is realized by a normal network if and only if it satisfies NOC, or equivalently, if every one of its clusters is inclusion-visible. In this case, the Hasse diagram provides a canonical normal realization. These are exactly the clustering systems realized by tree-child networks. The NOC formulation yields a sharp quadratic upper bound on the number of distinct clusters of tree-child and normal networks and a direct polynomial-time recognition algorithm. Finally, we explore several consequences of the NOC perspective beyond the phylogenetic setting. These include connections to the enumeration of normal networks, an order-theoretic interpretation of inclusion-visibility, structural properties of NOC set systems, and a tractable special case of Minimum Set Cover, which is NP-hard in general.
Read the full abstract
Robotics
Bridging Thought and Action: Taming Long-Horizon Instability in Open-Source LLM Agents with a MetaTool-Enhanced ROS Framework
Large Language Models (LLMs) have enabled more natural human-robot interaction, but open-source models often exhibit unstable long-horizon reasoning and inefficient action execution when deployed in agentic robotic frameworks. This paper presents an enhanced ROS-Agent based architecture that improves task reliability and execution efficiency for agentic robotic systems using open-source LLMs. The proposed system introduces a novel intermediate mechanism, termed the MetaTool, which enforces structured planning prior to action execution. Given a natural-language command, the MetaTool induces the LLM to generate a pseudo-code plan of intended tool invocations, which is stored in the ROS-Agent's scratchpad and persists throughout execution. By explicitly separating planning from execution, the proposed approach reduces execution loops and improves deterministic behavior. The architecture is validated on a custom mobile robotic platform with multimodal perception and motion control capabilities. Experimental results on real-world interactive tasks demonstrate improved task completion and contextual consistency, with up to ~24% gains on complex tasks compared to the baseline framework.
Read the full abstract
Software Engineering
The Agentic Company OS: Substrate Inversion for Sustained Enterprise Agent Deployment
Enterprise AI agents often succeed in a demonstration and then stall once they must operate day after day. An industry report estimates that most pilots never reach production and that deployed systems rarely retain feedback or improve over time, while agent benchmarks show single-run successes masking unreliable repetition. We argue that these failure modes share a common architectural root: agents reason over data structured for human operators and traditional applications, not for the language models that power them. This position paper proposes that companies deploying agents in sustained operation should rebuild their cognitive substrate, the shared environment agents read as working context, around representations matched to that reasoning surface, isolating schema translation to the action boundary. Markdown is the instantiation available today, not a proven agent-native primitive. Two mechanisms ground the argument: context-bandwidth asymmetry, the gap between one-pass reading of connected prose and field-by-field typed access that strips relations; and cross-loop coupling, the claim that action, skill, and policy loops compound only if they share one substrate. A four-layer framework (Data, Knowledge, Intelligence, Governance) operationalizes the position, with a Sync Agent enforcing the action boundary and a per-skill trust gradient, making governance and auditability structural properties of the substrate. The position revives the shared-substrate tradition of classical multi-agent systems under LLM-era economics. We analyze the main objections and risks, including indirect prompt injection on the compile path, and outline a research agenda for evaluating substrates directly.