Machine Learning
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Wednesday 9 September
In continual reinforcement learning, carefully managing the stability-plasticity tradeoff remains a core challenge. Recent work by Abel et al. (2025) formalized this dilemma by defining plasticity as the generalized directed information from an agent's observations to its actions, and empowerment as the generalized directed information from its actions to its observations. This formulation successfully reframes the traditional stability-plasticity tradeoff as an empowerment-plasticity tradeoff. However, while extensive literature exists on optimizing for empowerment, there is currently no research addressing the optimization of plasticity under this new definition. This paper presents preliminary work toward optimizing plasticity within Markov decision processes. We show that there exists a Bellman optimality equation for optimizing plasticity similar to previous work for empowerment.
Spectral Theory
An Improved Bound for the Ovals Problem
Let $\gamma\subset\mathbb R^{m},\,m\geq2,$ be a closed curve of length $2\pi$ with its curvature $\kappa$, parametrized by arc length, and let $\lambda_\gamma$ be the first eigenvalue of the periodic curvature Schrödinger operator $-d^2/d s^2+\kappa(s)^2$. We obtain \[ \lambda_\gamma\geq \frac{\sqrt{\pi}}{2} \left(\frac{\Gamma(7/6)}{\Gamma(5/3)}\right)^3. \] This is a near-sharp lower bound for the Ovals problem. Our proof introduces a new geometric approach. We derive a convolution identity from the closure condition and combine it with projection averaging over tangent directions and sharp Poincaré inequalities on antipodal arcs. As applications, we provide an improved two-state kinetic Lieb-Thirring inequality and the corresponding two-eigenvalue constant.
Read the full abstract
Systems and Control
Impedance-Based Sensitivity Analysis for Stability Enhancement of LCC-HVDC Links Connected to Weak Grids Using Grid-Forming Converters
This paper presents a frequency-domain, impedance-based sensitivity methodology for stability assessment and enhancement of line-commutated converter HVDC (LCC-HVDC) links operating under weak-grid conditions. The methodology integrates frequency-domain identification tailored for black-box systems, the Generalized Nyquist Criterion (GNC) for multivariable stability assessment, and modal impedance decomposition with participation-factor analysis to locate and interpret interaction mechanisms. The approach is validated against a detailed linearized state-space model and nonlinear EMT simulations of an LCC-HVDC benchmark. A sensitivity study varying the grid short-circuit ratio (SCR) reveals a stability limit for the standalone LCC-HVDC link and demonstrates that the integration of a grid-forming voltage source converter (GFM-VSC) substantially increases the stability margin.
Read the full abstract
Fluid Dynamics
Mixing dynamics and transport mechanisms during laminar stirring flows
We investigate laminar mixing of a passive dye in an infinite, two-dimensional domain filled with a Newtonian fluid by simulating a cylindrical stirrer that rotates at constant speed along a circular path, stirring an initially quiescent fluid. The fluid is marked by a passive dye in the lower half of the domain, enabling a systematic analysis of dye-interface evolution and mixing dynamics. By varying the stirring Reynolds number within the laminar regime, we identify how transitions in flow topology govern mixing and distinguish three mixing regimes spanning diffusion-dominated and advective mixing. In the diffusion-dominated regime, mixing is characterized by the formation of a well-mixed central region and the development of a spiral dye pattern that evolves in an approximately self-similar manner. Advective mixing is marked by substantial deformation of the dye interface beyond the central region. We provide a mechanistic interpretation of advective mixing by relating mixing events to specific flow features: (i) direct interaction between the stirrer and the dye interface, and (ii) vortex shedding near the stirrer and away from the central region. Enhanced mixing occurs when vortical structures are able to escape the central region and transport scalar gradients nonlocally. When vortical activity remains confined near the stirrer's path, mixing per stirrer period exhibits only weak dependence on stirring speed despite increasing stirring intensity. Significant enhancement of mixing per period occurs only when escaping vortices introduce a new transport mechanism that carries scalar gradients far beyond the stirrer's path. Overall, we present a mechanistic link between flow topology and mixing by identifying the transport mechanisms through which flow features govern scalar evolution, providing a bridge between the kinematics of mixing and the underlying fluid dynamics.
Read the full abstract
Solar and Stellar Astrophysics
Solar Flare Prediction Using a Hybrid Convolutional Neural Network and Transformer Model
Solar flares are intense bursts of electromagnetic radiation that occur when stored magnetic energy in the Sun's atmosphere is suddenly released. They are categorized into five classes – from least to most powerful: A, B, C, M, and X – with each successive class representing a ten-fold increase in energy output. The electromagnetic radiation emitted by the stronger classes (C, M, and X) is capable of causing significant disruptions to communication systems, satellites, and power grids on Earth. Accurate prediction of solar flares is crucial for mitigating their adverse effects and ensuring the functionality of critical infrastructure. This research introduces a novel model named ResNet-Transformer, which combines a convolutional neural network (CNN), ResNet50, with a standard Transformer architecture. The hybrid model effectively processes both spatial and time-series data derived from solar images to predict the occurrence, class (C, M, and X), and probability of solar flares within 24-hour, 36-hour, and 72-hour windows. This hybrid deep learning model represents the first of its kind in the domain of image-based, multiclass solar flare prediction. We evaluated the model's performance using a comprehensive set of metrics, including weighted precision, recall, and F1 score, together with balanced accuracy, the Matthews correlation coefficient (MCC), Cohen's kappa, and the area under the receiver operating characteristic curve (ROC-AUC). Results show that ResNet-Transformer surpasses traditional machine learning methods, such as support vector machines (SVMs) and standalone CNN models, across all evaluated metrics. This study highlights the potential of integrating convolutional neural networks with Transformers to enhance predictive capabilities in solar physics, paving the way for more reliable and timely solar flare forecasting.
Read the full abstract
Optics
New Method for Path-length Equalization of Long Single-mode Fibers for Interferometry
The ability to use single mode (SM) fibers for beam transport in optical interferometry offers practical advantages over conventional long vacuum pipes. One challenge facing fiber transport is maintaining constant differential path length in an environment where environmental thermal variations can lead to cm-level variations from day to night. We have fabricated three composite cables of length 470 m, each containing 4 copper wires and 3 SM fibers that operate at the astronomical H band (1500-1800 nm). Multiple fibers allow us to test performance of a circular core fiber (SMF28), a panda-style polarization-maintaining (PM) fiber, and a lastly a specialty dispersion-compensated PM fiber. We will present experimental results using precision electrical resistance measurements of the of a composite cable beam transport system. We find that the application of 1200 W over a 470 m cable causes the optical path difference in air to change by 75 mm (+/- 2 mm) and the resistance to change from 5.36 to 5.50 {\Omega}. Additionally, we show control of the dispersion of 470 m of fiber in a single polarization using white light interference fringes ({\lambda}c=1575 nm, {\Delta}{\lambda}=75 nm) using our method.
Read the full abstract
Logic
Simultaneous extensions of states on quantum logics
Suppose that L is a quantum logic (= an orthomodular poset) and A, B are Boolean subalgebras of L. Suppose further that s1(resp. s2) is a state on A (resp. on B). We ask when both s1 and s2 allow for a simultaneous extension over L. Under a natural necessity condition on A, B, s1, s2 we find a sufficient condition for the existence of the extension. This condition is the requirement of an abundance of two-valued states on L. We then show by the Greechie pasting technique that this condition is far from necessary even for state-space-rich logics.
Read the full abstract
Dynamical Systems
Symplecto-rigidity and bootstrap for Anosov symplectomorphisms and contact Anosov flows
This paper is a sequel to the authors' paper on rigidity of higher dimensional contact Anosov flows~[GRH]. At the time the authors were unaware of much earlier work of Hamenstädt~[Ham] devoted to the same problem. The methods of~[GRH] and~[Ham] are overlapping but not entirely the same. In this paper we strengthen some the results of Hamenstädt, by further bootstrapping the regularity of the conjugacy of 2-pinched contact Anosov flows. We also introduce a notion of symplecto-rigidity for symplectomorphisms and prove that some Anosov diffeomorphisms are symplecto-rigid. We further combine symplecto-rigidity with various rigidity techniques to establish a number of smooth rigidity results of Anosov symplectomorphisms. Some new phenomena are present in the realm of Anosov symplectomorphism. For example, while the well-known de la Llave example on the 4-torus demonstrates absence of rigidity in the space of smooth Anosov diffeomorphisms, however, we prove that it is rigid in the space of Anosov sympelctomorphisms.
Read the full abstract
Signal Processing
Bridging Echolocation Gaps in Automated Beaked Whale Tracking
Passive acoustic monitoring (PAM) is an effective and widely used tool for tracking marine mammals, particularly beaked whales, which are infrequently observed visually because of their deep-diving behavior. However, the large data sets generated by PAM methods often require time-consuming hand labeling to identify whale trajectories in the recorded audio. Automated multi-target tracking (MTT) methods could significantly reduce human workload, but current methods have difficulty forming continuous tracks because of the irregularity of beaked whale echolocation clicks. More precisely, regular sequences of clicks are often interrupted by longer pauses that occur when whales face away from the sensors or stop clicking. Consequently, the probability of detection is difficult to model accurately, and MTT trajectories become fragmented at these pauses. In this paper, we propose a multistage target-estimation method aimed at bridging large gaps in click sequences by combining belief propagation-based MTT with track smoothing and stitching. We validate our method using acoustic recordings of clicks from goose-beaked whales (Ziphius cavirostris), and demonstrate that it improves track estimates and reduces fragmentation in the presence of consecutive missed detections. When evaluated with the generalized optimal subpattern assignment (GOSPA) metric, our method outperforms existing MTT reference methods through reductions in missed-target errors.
Read the full abstract
Machine Learning (Statistics)
Weighted Empirical Risk Minimization for Machine Learning under Long-Range Dependence: Exact Pathwise Rates and Learning-Error Geometry
We develop an exact almost-sure learning theory for smooth parametric models trained by regularly weighted empirical risk minimization on long-range dependent data. The training observations are generated from a fixed finite window of a stationary Gaussian sequence, and the sample weights are regularly varying. If the loss gradient at the population minimizer has Wiener-chaos rank $m$ and a nonzero low-frequency coefficient, then, in the long-memory interior regime, the finite-lag score reduces on the iterated-logarithm scale to a single weighted Hermite chaos. This yields an almost-sure Bahadur representation, an exact limsup law for the learned parameter, and, for $m\ge2$, the functional cluster set of the complete learning trajectory. The polynomial learning exponent is determined by the memory parameter and the chaos rank and is invariant under the admissible power weighting, whereas the sharp pathwise constant and cluster geometry depend on the weights. In the rank-one case, global optimization over the admissible power exponents shows that every optimizer is positive. Time-series prediction and classification examples illustrate the results.
Read the full abstract
Solar and Stellar Astrophysics
Hbeta Spicules, Small-scale Jets, and Hbeta Microflashes: Sub-arcsecond Dynamic Events and their Magnetic Origins in the Lower Solar Chromosphere Observed by DKIST
We examine an on-disk solar quiet region of enhanced magnetic network using Hbeta images from Daniel K. Inouye Solar Telescopes (DKISTs) Visible Broadband Imager (VBI) and line-of-sight magnetograms from DKISTs Visible Spectro-Polarimeter (ViSP). We also compare the ViSP magnetograms with co-aligned co-temporal line-of-sight magnetograms from the Solar Dynamics Observatory (SDO)/Helioseismic and Magnetic Imager (HMI). We find:(i) Two types of chromospheric jet-like features: Hbeta spicules, and small-scale jets. Both are rooted near edges of magnetic network lanes. Several sit close to ViSP-detected tiny islands of either minority-polarity flux or dips in majority-polarity flux. Hence, those several plausibly stem from mixed-polarity magnetic flux, and are caused by chromospheric magnetic reconnection. For He̱ta} spicules, the average width, length, lifetimes, and speeds are 420$\pm$400 km, 2600$\pm$1600 km, 4.3$\pm$0.25 min, and 12$\pm$4.7 kms. For small-scale jets, those are 385$\pm$100 km, 620$\pm$40 km, 5$\pm$3min, and 4.4$\pm$1.5 kms. (ii) H$\beta$ microflashes sit in evidently unipolar flux. Their widths, lengths, lifetimes, and speeds are 180$\pm$65 km, 365$\pm$100 km, 4.3$\pm$2.6 min, and 3.1$\pm$0.2 kms. (iii) At the base of a coronal plume, due to its higher spatial resolution than HMI, ViSP shows both some very strong (>800 G) majority-polarity flux not shown by HMI and some sub-arc-second minority-polarity-flux inclusions not shown by HMI. These sub-arcsecond chromospheric transients carry sufficient energy (10^{24}erg) to transiently heat the local chromosphere and corona and potentially contribute to solar-wind acceleration. They may represent the small-scale end of a continuum of magnetic-reconnection-driven activity, highlighting the importance of DKIST's high-resolution magnetic-field measurements for understanding small-scale chromospheric dynamics.
Read the full abstract
Audio and Speech Processing
From Metrics to Natural Dialogue: French Full-Duplex Benchmark for Spoken Dialogue Models
Full-duplex spoken dialogue models aim to make voice agents more natural by allowing them to listen, speak, pause, and respond during ongoing conversation. However, it is not clear whether full-duplex benchmarks behave the same way when models are evaluated in a different language. To investigate this, we introduce a French full-duplex benchmark (FDB) with two variants, CALLFC-FDB for Canadian French and MEDIA-FDB for European French, and compare them with an English FDB. Built from real spoken resources, these benchmarks evaluate key full-duplex skills, including pause handling, turn-taking, backchannels, and interruptions. Beyond introducing French FDBs, we evaluate human–human conversations with FDB metrics to better understand the values these metrics take in real-world dialogue. Our analysis reveals that most timing-based metrics behave similarly across languages, while content-based evaluation degrades under language mismatch. We also find a trade-off between optimizing benchmark metrics and preserving conversational naturalness.
Read the full abstract
Algebraic Geometry
Torelli theorem for the product of moduli spaces of vector bundles over a curve
We prove a Torelli-type theorem for a product of the moduli spaces of semistable vector bundles over smooth projective curves of genus $g\,\ge\, 4$. A similar result is proved for the product of the moduli stacks of semistable vector bundles. This is proved using a decomposition theorem for the product of the normal projective varieties with Picard rank one and discrete Picard group. As an application, we compute the automorphism group of the product of the moduli spaces and moduli stacks.
Read the full abstract
Classical Analysis and ODEs
Dimension-free estimates for full discrete maximal functions associated with Euclidean balls and spheres
We prove that the full discrete Hardy-Littlewood maximal operator associated with Euclidean balls satisfies dimension-free bounds on $\ell^p(\mathbb Z^d)$ for every $1<p<\infty$. We also establish analogous dimension-free bounds for the full discrete spherical maximal operator when $d\geq 5$ and $2\leq p<\infty$. The main new idea is to approximate the relevant Fourier multipliers by finite linear combinations of normalized discrete Gaussian multipliers and their translates. We obtain these approximations uniformly in frequency and with uniformly bounded coefficients through a refined saddle-point analysis. The ball result resolves a question of E.M. Stein, while the spherical result gives, in the range $p\geq 2$, a dimension-free strengthening of the theorem of Magyar, Stein, and Wainger.
Read the full abstract
Cryptography and Security
Beyond Static Guarantees: Measuring the Static-Pass Dynamic-Fail Gap in Security-Sensitive and LLM-Generated Python Code
Advances in large language models (LLMs) fuel the quest for scalable methods to assess the security of generated and security-sensitive software. Static analysis is widely adopted as a scalable, reproducible, and inexpensive security gate, but cannot directly observe runtime exploit behaviour. Vulnerabilities dependent on adversarial inputs, execution context, or exploit chaining may evade static checks while remaining exploitable in practice, yet passing static analysis is often treated as evidence of secure behaviour. This paper introduces the Static-Pass Dynamic-Fail (SPDF) phenomenon and a three-stage agentic pipeline combining static scanning, LLM-driven Common Weakness Enumeration (CWE) reasoning, and autonomous exploit verification in isolated Docker containers. We evaluate 1,355 Python samples from SecurityEval, RedCode, and CyberNative datasets. Of the 654 samples producing no findings under the composite Bandit-Semgrep gate, the LLM detection stage identified 394 candidate vulnerabilities across 235 files. Dynamic verification confirmed or partially confirmed exploitability in 95 files, yielding an inclusive pipeline rate of 14.53% (roughly 1 in 7 statically clean samples). This rate represents the proportion of Bandit-Semgrep-clean samples for which the pipeline identified a candidate vulnerability and obtained runtime evidence supporting exploitability. Outcomes varied by dataset: among candidate file–CWE pairs, confirmed exploitability was 33.7% for RedCode, 28.6% for CyberNative, and 5.4% for SecurityEval. Several frequently confirmed classes, including CWE-338 and CWE-916, were flagged by neither Bandit nor Semgrep. These findings indicate that static-analysis success and runtime security are hierarchical layers of software assurance rather than interchangeable measures, and have the potential to reshape how AI-generated and security-sensitive code is evaluated.
Read the full abstract
Differential Geometry
Twistor theory of Loxodromes
The twistor correspondence provides a duality between curves in $S^4$ and ruled surfaces in $\mathbb{CP}^3$ in which the conformal properties of the former are reflected in the projective properties of the latter. We use this to characterise a $14$–dimensional family of homogeneous ruled surfaces in $\mathbb{CP}^3$ which correspond to loxodromes in $S^4$.
Read the full abstract
High Energy Physics (Theory)
Complete set of tree-level $2\to 2$ scattering amplitudes of ghost-free bimetric theory
Bimetric theory is an extension of general relativity that perturbatively describes one massless and one massive graviton. It admits observationally viable cosmologies, passes local tests of gravity, and provides candidates for dynamical dark energy and spin-2 dark matter. Scattering amplitudes provide a complementary probe of its consistency, e.g. through analyticity, unitarity, and causality. While the tree-level $2\to2$ amplitudes of the closely related theory of massive gravity have previously been computed and analysed, the richer amplitude structure of bimetric theory has mostly been studied for selected processes and helicity sectors. We present the complete set of tree-level $2\to2$ scattering amplitudes of ghost-free bimetric theory around proportional Minkowski backgrounds. We expand the action through quartic order in the mass eigenstates and use a computer-algebra workflow to compute all of the $199$ symmetry-inequivalent helicity amplitudes. We find a clear hierarchy: amplitudes with a single massive external state vanish, those with exactly two are independent of the nonlinear bimetric parameters, and nonlinear ghost-free parameter dependence first appears with three massive external states. Taking the massive-gravity limit yields the complete set of tree-level $2\to2$ massive gravity amplitudes, which agree exactly with previous results. Finally, all amplitudes grow with energy at most as $E^6$, corresponding to the characteristic $\Lambda_3$ strong-coupling scale, and this maximal growth cannot be eliminated by any nontrivial choice of the theory parameters.
Read the full abstract
Game Theory
Where Should Society Draw the Line? A Social Choice Approach to Collective Consent
Society constantly has to determine the boundaries of what it deems acceptable, from legislative decisions to the guardrails governing autonomous systems. We initiate the axiomatic study of collective consent: given individuals' attitudes toward options, which options should receive societal consent? We organize our analysis around three principles: sufficient support, minority protection, and dominance by decisively better options. Each captures a distinct reason for withholding societal consent from an option. For each principle, we develop a corresponding solution concept that transparently implements the principle and is canonical in a mathematically precise sense. For example, for minority protection, the resulting concept is a consent-adapted version of Moulin's Proportional Veto Core. Balancing multiple principles simultaneously is more challenging. To address this, we develop a game-theoretic characterization of our veto core that naturally gives rise to a family of related concepts. From this family, we identify the Approval-Weighted Veto Core as particularly desirable. By making minorities' blocking power depend on the approval support of the options being challenged, it smoothly interpolates between proportional minority protection and majority support. Experiments on five datasets spanning high-stakes decision-making (such as political elections, ethical AI evaluations, and moral decision-making) show that solution concepts violating a principle in theory also violate it empirically.
Read the full abstract
Computation and Language
Multilingual in Name Only? Cultural and Linguistic Weaknesses of LLMs in Urdu
Multilingual large language models (LLMs) are increasingly used for open-ended text generation, yet their behaviour in low-resource languages remains poorly understood. In this work, we question how correct and reliable is the generation of multilingual LLMs when used for the task of story generation. We consider Urdu language as a representative low-resource language. We generate Urdu-Stories, a corpus of 93 stories generated using three contemporary LLMs (GPT-5.1, Qwen-3-Max, DeepSeek-3.1). We manually annotate the errors present in them under a nine-label linguistic, semantic, and cultural taxonomy. Our notable findings suggest that LLMs often make basic errors of grammar and semantics. The stories lack coherence, have unnatural repetition and show pervasive cultural shallowness. We further show using few-shot prompting that the cultural and context errors largely remain unresolved. Our findings highlight the limitations of current LLMs as a reliable source of content generation and information retrieval for low-resource languages.
Read the full abstract
Functional Analysis
Antitonicity property of the Moore-Penrose inverse for selfadjoint operators
In this article we study the antitonicity property of the Moore-Penrose inverse in the class of selfadjoint operators, with respect to the Löwner order. For this purpose, we employ different positive decompositions that selfadjoint operators admit. In addition, we relate a weak version of the antitonicity property with Thompson components.
Read the full abstract
Computer Vision
GRADE: Single-Frame Generative Radar Depth Estimation Under Visual Degradation
Dense 3D depth perception fails under smoke, fog, and darkness because optical sensors cannot penetrate airborne particulates. mmWave radar remains usable and measures range accurately under these conditions, but its small aperture limits angular resolution. We present GRADE, which grounds a pretrained generative prior in single-frame radar geometry to estimate high-fidelity metric depth. GRADE first maps raw 4D radar spectra to coarse metric depth. A latent diffusion backbone then recovers structural detail while conditioning every denoising step on this estimate. A pixel-space adapter uses residual camera cues when available and is trained across clear, smoke-degraded, and occluded inputs so the full output approaches the radar-conditioned path as visibility degrades. Trained and evaluated on ~95K frames across 12 buildings with real smoke, GRADE achieves an MAE of 0.303 m in clear scenes and 0.313 m under smoke, outperforming existing baselines. Code and datasets are available at https://phi-lab-rice.github.io/GRADE.
Read the full abstract
General Relativity and Quantum Cosmology
Massive quantum divergence on the Cauchy horizon of a black hole
We investigate a quantum massive scalar field in the interior of a charged and spherically-symmetric (Reissner-Nordström) black hole. We examine the behaviour for varying values of the black hole charge, field mass and coupling constant when the field is in two quantum states: Hartle-Hawking (representing a black hole in thermal equilibrium) and Unruh (representing a black hole evaporating via the emission of Hawking radiation). We show that the vacuum polarization (in both states) generically diverges on the Cauchy horizon, in stark contrast to what happens for massless fields; also, the angular components of the quantum stress-energy tensor generically diverge on the Cauchy horizon and, in at least one of the two states, they do so more strongly than for massless fields. Furthermore, we calculate the energy fluxes in Eddington-Finkelstein coordinates $\{u,v\}$. We show that these fluxes in both states on the Cauchy horizon do not generically vanish. This implies, in particular, that in regular, Kruskal coordinates $\{U,V\}$ the ingoing flux diverges like $V^{-2}$ on the Cauchy horizon (where $V=0$). This divergence suggests that its backreaction via the semiclassical Einstein equations would yield a strong singularity, as opposed to its weaker, classical counterpart. Interestingly, there are exceptions, in which the energy fluxes in Eddington-Finkelstein coordinates vanish: (i) in the extremal limit (where the black hole is maximally charged); (ii) certain fine-tuned regions of parameter space, where the fluxes change sign.
Read the full abstract
Nuclear Experiment
Island of Inversion in neutron-rich cobalt isotopes revealed from mass measurements
Mass measurements of the ground and isomeric states of $^{68-70}$Co have been performed using the JYFLTRAP Penning-trap mass spectrometer at the IGISOL facility. The masses were measured, either for the first time for the isomeric states of $^{68}$Co and $^{70}$Co, or with greatly improved precision for the others, removing ambiguities in the mass surface beyond $N=40$. The ordering of the low and high-spin states in $^{68}$Co and $^{70}$Co has also been established. The results, supported by Large-Scale Shell Model and Discrete Non-Orthogonal Shell-Model calculations, show a gradual lowering of intruder states with increasing neutron number, eventually leading to an inversion in $^{70}$Co. These findings clarify previously proposed contradictory interpretations. Finally, we demonstrate the importance of including induced effective 3N forces for a consistent description of binding energies in the island of inversion near $N=40$.
Read the full abstract
General Relativity and Quantum Cosmology
Binary-black hole spin population results may be driven by prior degeneracies
Gravitational waves carry information on the spins of merging binary-black holes. The orientations of their spins relative to their orbits ("tilts")—while challenging to measure—differentiate the astrophysical formation mechanisms by which merging black hole binaries may form. Multiple analyses have reported tentative evidence that black hole spins are preferentially oriented in the plane of the binary orbit. We show that these results are likely extrapolated from more easily measurable "effective spin" parameters which characterize the gravitational-wave inspiral of black hole binaries but may not uniquely constrain their astrophysical formation. In particular, we reproduce a peak in the spin tilt distribution with information about the marginal effective spin distributions alone. We propose a geometric picture to compare constraints on effective spins to standard population modeling assumptions, suggesting that current inferences of a preferred spin orientation are spurious.
Read the full abstract
Machine Learning
Adaptive Margin Ordinal Loss: Penalizing Center-Class Hedging in Ordinal Classification
Standard cross-entropy loss causes neural networks trained on ordinal classification tasks to hedge predictions toward center classes, a failure mode we term center-class hedging. This occurs because predicting the middle class minimizes expected symmetric loss, making it the path of least resistance regardless of the true label. Existing ordinal losses address related problems such as large-error penalization and rank consistency, but none directly suppresses center-class hedging as a function of where the true label lies relative to the ordinal center. We propose the Adaptive Margin Ordinal Loss (AMOL), a multiplicative weight applied to per-class loss terms of the form $m(k,y) = 1 + \alpha \cdot (1 - |k-c|/c) \cdot (|y-c|/c)$, where $c$ is the center class, $k$ is the candidate class, and $y$ is the true label. The weight encodes a joint condition: it is large only when the candidate class is near center and the true label is far from center, collapsing to standard behavior otherwise. We further introduce the Center-Hedging Rate (CHR) as a diagnostic metric that directly quantifies this failure mode. Across four ordinal classification benchmarks and five random seeds, AMOL achieves the best or tied-best Quadratic Weighted Kappa (QWK) on all four datasets compared to cross-entropy, OLL, and SORD baselines. An asymmetric variant (AMOL-asym) eliminates center-class hedging entirely on the Abalone dataset ($\text{CHR} = 0.000 \pm 0.000$ across all five seeds, $n \approx 266$ extreme-class test samples per run), compared to $0.074 \pm 0.005$ for standard cross-entropy.
Read the full abstract
Algebraic Geometry
The Reye geometry inside the 64 lines of the Schur quartic
We identify the classical geometry hidden in the Naskręcki–Pokora $(24_4,32_3)$ configuration on the Schur quartic. Using Höhn's identification of the $24$ selected lines with the $24$ roots of $D_4$, the antipodal involution on the roots induces a fixed-point-free quotient of the incidence configuration, and this quotient is precisely the classical Reye configuration. We also determine the symmetry of the complete $64$-line incidence geometry: its automorphism group has order $4608$, the two Naskręcki–Pokora configurations form a single orbit, and the stabilizer of either has order $2304$ (projectively, $576$). The $64$ lines extend canonically to a $176$-line arrangement carried by six projectively equivalent Schur quartics, with $176=16+16+9\cdot16$ and induced surface permutation group $S_3\times S_3$. Finally, the antipodal quotient itself extends coherently through this six-quartic geometry: on each Schur quartic it produces two Reye configurations sharing the same $16$-element incidence skeleton, and on the full $176$-line arrangement it gives a compatible global quotient. This reveals a precise incidence-theoretic connection with classical desmic geometry, while showing that this connection is not a literal identification with the two Reye configurations arising from the classical desmic construction in $P^3$.
Read the full abstract
Information Retrieval
When Synthetic Data Hurts: On Catastrophic Forgetting in Skill Retrieval for LLM Agents
LLM agents increasingly rely on external skills retrieved at runtime, making skill selection from large repositories a critical challenge. We present a production skill router over 34,396 skills and a large-scale study of skill retrieval using limited real supervision and synthetic data. We found that the synthetic-data fine-tuning improves in-distribution retrieval but it causes catastrophic forgetting on real and out-of-distribution (OOD) data. We evaluate several forgetting mitigation fine-tuning approaches inspired by continual learning, including embedding-anchor regularization, Learning without Forgetting (LwF), Elastic Weight Consolidation (EWC), and L2-initialization. The results show that these approaches not only retain the performance on OOD skills retrieval but also improve the retrieval on synthetic in-distribution skills by 13.98% for 0.6B Qwen retriever and reranker. Our results provide a practical benchmark and a robust fine-tuning recipe for scarce, multi-positive supervision.
Read the full abstract
Computer Vision
Meta-Learning for Data-Efficient Plant Growth Estimation via Vision Transformers and Fuzzy Clustering
Accurate plant growth estimation is essential for greenhouse monitoring, yet obtaining labeled data remains costly and time-consuming. To address this, we propose a few-shot regression framework that combines Vision Transformer (ViT) feature embeddings, clustering-based task construction, and gradient-based meta-learning, and show that task construction in embedding space is a primary driver of performance. The approach leverages an unlabeled image pool to organize data into structured tasks using fuzzy c-means clustering, enabling efficient learning from a small number of labeled samples. We systematically evaluate meta-learning methods and show that second-order methods (e.g., Model-Agnostic Meta-Learning variants such as MAML++) outperform classical baselines in the few-shot regime. Furthermore, intra-cluster support selection has a limited and dataset-dependent impact. Experiments on two plant datasets show that structured task design combined with meta-learning enables reliable plant growth estimation under severe label scarcity.
Read the full abstract
Robotics
Lie-Algebraic Bell Recurrences for Arbitrary-Order Twist Jets and Parallel-Mechanism Closure
This paper develops an arbitrary-order kinematic construction that links serial propagation, parallel-mechanism closure, and rigid-platform point fields within one dual screw framework. A cylindrical joint is retained as one native physical block, with revolute and prismatic joints obtained as special cases. For each fixed joint axis, ordinary Bell polynomials organize the derivatives of the exponential factor; across a chain, the noncommuting factors remain in their physical order. Initial-frame prefix and terminal-resolved covariant formulas then produce equivalent representations of the serial twist jet. For a parallel mechanism, repeated Leibniz differentiation, with joint-level derivatives organized by Bell polynomials, yields an arbitrary-order triangular active-passive closure recurrence: the same passive Jacobian is solved at every derivative order at a regular configuration, while the right-hand side contains only prescribed active data and lower-order jets. The resulting platform twist jet is mapped exactly to the point-independent affine invariants of the velocity, acceleration, jerk, and snap fields. The validation is deliberately complementary: a generic 3C chain with noncoplanar axes and nonzero rotational and translational cylindrical coordinates tests ordered serial propagation, an RR+RRR spherical wrist tests active-passive closure, and a Hunt-type 6-RUS mechanism with six active revolute joints tests an independently reconstructed platform jet and its affine fields. Independent differentiation of the rigid motion, evaluation of the affine fields, and the differentiated branch closures all agree through fourth order with residuals below $10^{-12}$ in the corresponding SI units. The formulation is purely kinematic and applies at configurations where the selected active-passive partition is regular.
Read the full abstract
Functional Analysis
Structural consequences of the Schur $p$-property for Lipschitz-free $p$-spaces
Let $0<p<q\leq 1$. We show that the Schur $p$-property provides a powerful structural principle for Lipschitz-free $p$-spaces. Our main result asserts that $\mathcal{F}_p(M)$ has the Schur $p$-property for every $q$-metric space $M$; when $M$ is compact, it has the strong Schur $p$-property, with a constant depending only on $p$ and $q$. As consequences, $\mathcal{F}_p(M)$ is $\ell_p$-saturated, contains no isomorphic copy of an infinite-dimensional $r$-Banach space for $p<r\leq 1$, and every bounded operator from a $p$-Banach space into $\mathcal{F}_p(M)$ is either compact or fixes a copy of $\ell_p$. These results settle Questions 6.1, 6.2, 6.5, and 6.6 from our recent work [F. Albiac, J. L. Ansorena, J. Bíma and M. Cúth, Lipschitz free p-spaces for $0<p<1$ in the light of the Schur $p$-property and the compact reduction, J. Geom. Anal. 36 (2026), Paper No. 54] on the Schur $p$-property, together with several related problems. They also confirm a prediction of Kalton and the first-named author from 2009: no infinite-dimensional $q$-Banach space has the $p$-Lipschitz lifting property. Further applications reveal a sharp contrast between the linear and Lipschitz structures of nonlocally convex spaces.