Computer Vision
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Wednesday 9 September
This paper addresses the issue of supervised classification in the context of hyperspectral satellite images. It deals with two fundamental aspects: dimensionality reduction of data and the selection of appropriate supervised classification techniques. Firstly, we delve into dimensionality reduction, a critical step in simplifying the management of hyperspectral data. The reduction aims to decrease complexity in terms of memory and computing time. We examine two commonly used methods: Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). Subsequently, we explore the selection of the most suitable supervised classification algorithms for hyperspectral images. We compare the performance of three methods: K-Nearest Neighbors (KNN), Support Vector Machines (SVM), and Random Forest (RF) using real hyperspectral data. The results highlight that the combination of PCA and RF yields the highest overall accuracy and Kappa coefficient.
Computer Vision
Learning to Adapt and Calibrate: Score Distribution Alignment for Few-Shot Uncertainty Prediction in Medical VLMs
Uncertainty estimation for medical vision–language models (VLMs) using conformal prediction has gained increasing attention due to its distribution-free coverage guarantees. However, standard conformal prediction relies on exchangeability between calibration and test data and typically requires a sufficiently large calibration set to obtain reliable coverage. These assumptions are difficult to satisfy in few-shot transfer settings, where only a small labeled support set is available to adapt a pretrained VLM to a new medical task, while an unlabeled query set is used for evaluation. Supervised fine-tuning on the support set changes the model parameters and consequently shifts the nonconformity score distribution, breaking exchangeability between calibration and query samples and leading to unreliable coverage under distribution shift. Existing transductive conformal adaptation methods often preserve validity by avoiding supervised updates. While this helps maintain conformal assumptions, it underutilizes the scarce labeled support data and limits task adaptation, which is the primary objective in few-shot learning. In this setting, conformal prediction should serve as an uncertainty estimation layer that supports the adapted model, rather than preventing adaptation itself. To this end, we propose AlignCP, a framework that reconciles supervised few-shot adaptation with conformal uncertainty estimation under non-exchangeability. AlignCP learns a reweighted calibration distribution that reduces the score-level discrepancy between the labeled support set and the unlabeled query set. By aligning the one-dimensional nonconformity score distributions, AlignCP aims to close the coverage gap induced by adaptation without requiring query labels.
Read the full abstract
General Relativity and Quantum Cosmology
Generalized Black Holes with Fully Non-aligned Electromagnetic Fields
We develop a Plebański–Demiański-adapted parametrization of the general Ovcharenko–Podolský solution and construct new families of generalized black holes in four-dimensional Einstein–Maxwell theory. These Petrov type~D spacetimes possess non-null, fully non-aligned electromagnetic fields. A restriction imposed in the previous parametrization is not required by the field equations in shifted coordinates. Removing it retains an independent charge parameter $q$, which may be real or purely imaginary. For $q^2\geq0$, both the metric and electromagnetic field admit a smooth aligned limit to the charged Plebański–Demiański solution with $\Lambda=0$ and $e^2+g^2=q^2$. We impose angular conditions in the Griffiths–Podolský form allowing Killing horizon cross sections with spherical topology. The resulting eight-parameter class extends the seven-parameter class of Ovcharenko and Podolský [arXiv:2508.04850]. The remaining normalization condition is generically quartic, defining four algebraic branches. We give explicit metrics and electromagnetic fields for the non-twisting and $l=0$ twisting families and analyze their black hole and acceleration horizons, extremality, conicity, and further limits. We determine the effective-NUT-free branches of the latter family and their degenerate cases. Using the genuine Griffiths–Podolský form, we establish an explicit correspondence between the non-accelerating effective-NUT-free subclass and the Kerr–Newman–Bertotti–Robinson family and express its electromagnetic flux charges in our parameters. The $q^2<0$ sector includes the genuine uncharged Kerr–Bertotti–Robinson solution recently identified by Ovcharenko and Podolský [arXiv:2608.21672].
Read the full abstract
Numerical Analysis
Weakly Symmetric and Traceless Tangential-Normal Tensor Finite Elements: Application to the Brinkman Equations
We develop a family of weakly symmetric and pointwise traceless tangential-normal tensor finite elements in arbitrary space dimension and for all polynomial orders. Symmetry is imposed through local cell moments, while the only globally coupled stress degrees of freedom are tangential-normal facet moments; no vertex degrees of freedom are required. In dimensions three and higher, a lowest-order linear enrichment restores the rigid-motion facet control required for discrete Korn stability. As a principal application, we construct a distributional mixed method for the incompressible Brinkman equations using the physical viscous stress. Coupled with divergence-conforming BDM velocities and discontinuous pressures, the method is stabilization-free, uniformly stable with respect to the viscosity parameter, exactly divergence-free, and pressure-robust. We establish optimal-order error estimates in the natural norms. Under suitable parameter-explicit regularity assumptions, we also obtain a parameter-uniform boundary-layer estimate with optimal Darcy approximation order. Relaxing tangential-normal continuity yields an algebraically equivalent stress-hybridized formulation and a stabilization-free virtual element realization.
Read the full abstract
Combinatorics
The VC-dimension of strongly regular graphs
A graph $G$ is $n$-existentially closed or $n$-e.c. if, for all subsets $S\subseteq V(G)$ with $|S|=n$ and for all partitions $S=A\sqcup B$, there exists a vertex in $V(G)\setminus S$ adjacent to all vertices in $A$ and no vertices in $B$. We study the minimum number of edges $m(v,n)$ of a $v$-vertex $n$-e.c. graph, and show that $m(v,2)=3v+O(1)$ while $m(v,n)=\Theta(v\log v)$ for fixed $n\ge 3$. The latter result uses a connection to binary covering arrays. A related parameter is the VC-dimension of $G$, defined as the size of the largest subset of vertices shattered by the neighborhoods of vertices in $G$. We initiate systematic study of the VC-dimensions of strongly regular graphs (SRGs). We characterize the sufficiently large SRGs with VC-dimension 2. Furthermore, we determine the VC-dimension of sufficiently large Latin square graphs and of all SRGs of order at most 28, and prove upper bounds for SRGs arising from orthogonal arrays or Steiner systems.
Read the full abstract
High Energy Physics (Theory)
An Effective $S$-Matrix Approach to Low-Frequency Waveforms from Black Hole Mergers
We develop an on-shell description of low-frequency gravitational waveforms from black-hole mergers beyond leading order. Treating the strongly coupled merger as effective hard $S$-matrix data, we organize its long-wavelength response using soft theorems and the KMOC formalism. At next-to-leading order, the quantum soft theorem contains logarithmic terms absent from the classical soft theorem. We show that these extra terms cancel in the full KMOC in-in observable between the one-loop radiative amplitude and the corresponding graviton cut, leaving precisely the classical logarithmic contributions associated with gravitational drag and early-time acceleration. We also identify the $1/\omega$ corrections from remnant recoil and Christodoulou non-linear memory. These results reveal a hierarchy of merger information accessible at low frequency: logarithmic tails depend only on asymptotic hard data, recoil probes total radiated momentum, while non-linear memory probes the angular distribution of the emitted radiation.
Read the full abstract
Classical Analysis and ODEs
Positivity and Asymptotics for Chenevier's Orthogonal Polynomials
We prove the strict positivity conjectured by Chenevier for the critical vectors in the unconditional part of his automorphic Hermite–Minkowski theorem. The proof establishes strict negativity of all Verblunsky coefficients of a circle measure associated with the weight $(\arcsin x)/x$ on $(-1,1)$. A positive-kernel formula and the classical Schur algorithm give these signs, and a para-orthogonal transformation yields positivity in every degree. We also compute the positive density representing the negative of the Schur function as a Hausdorff moment generating function. After rescaling their indices to $[0,1]$, the normalized critical vectors converge weakly to the arcsine law, while Chenevier's critical scale is asymptotic to $8\pi/n$. At the critical boundary, a single nonzero effective integral vector is negative for every admissible test function exactly in odd degree and in degree zero. Finally, we prove an exact first-variation formula for exponential perturbations of the Legendre measure and derive its asymptotics for endpoint cusps. For the perturbation leading to Chenevier's weight, the derivative at the Legendre measure has a $(\log n)/(\pi^2n^2)$ term and an explicit constant at order $n^{-2}$. The corresponding nonlinear asymptotic remains conjectural.
Read the full abstract
High Energy Physics (Phenomenology)
Towards new D meson fragmentation functions
The Heavy Meson (Hymn) collaboration presents a new extraction of D meson fragmentation functions using experimental data from LEP and LHC. We focus particularly on kinematical regimes where perturbative QCD should be safely applicable to avoid contamination from higher-twist effects which could lead to an apparent process dependence of fragmentation functions. We account for the initial-state radiation and, as a novel ingredient, consider the prompt and non-prompt contributions separately. The analysis is carried out at next-to-leading order accuracy including uncertainty estimation based on Monte-Carlo replica technique. We disucss the exemplary case of $\mathrm{D}^0$ here and defer the results for $\mathrm{D}^\pm$ and $\mathrm{D}^{*,\pm}$ to a forthcoming publication.
Read the full abstract
Cosmology and Nongalactic Astrophysics
Primordial turbulence from inflation: a new inflaton-driven turbulent regime
For the first time, we establish a direct connection between inflationary dynamics and the subsequent magnetohydrodynamic evolution. Starting from vacuum fluctuations, we self-consistently evolve the coupled inflaton-plasma system through inflation, reheating, and into the radiation-dominated era. We find an inflaton-driven inverse cascade followed by freely decaying turbulence, with kinetic energy dominating over magnetic energy and driving dynamo amplification. The inflaton-driven turbulence causes steeper magnetic energy decay than helical turbulence and should therefore be accounted for subsequent evolution of primordial fields and their present-day observables.
Read the full abstract
Geometric Topology
Extremal entropy for products of Fuchsian representations
In this paper, we count the number of (almost) extremally stretched closed geodesics for a pair of (non-conjugate) Fuchsian representations of a closed surface group, and show that it has subexponential growth. We then deduce that, as a discrete subgroup of $\mathsf{PO}(2, 1) \times \mathsf{PO}(2, 1)$, the growth indicator of the product representation vanishes on the boundary of the Benoist limit cone. We also prove that for a general Zariski dense Borel Anosov subgroup of $\mathsf{PO}(2, 1) \times \mathsf{PO}(2, 1)$, its growth indicator vanishes on at least one boundary component of the Benoist limit cone, but not necessarily on both. One may view our first result as a sharpening of Thurston's result that there is a unique geodesic lamination $\lambda$ such that every measured lamination maximizing the ratio of lengths with respect to the two representations has support contained in $\lambda$. We hope this will be a starting point for a more general study of extremal entropy.
Read the full abstract
Number Theory
Sharp local estimates for smooth numbers
We establish sharp local estimates for smooth numbers. Corollaries are derived, including optimal estimates for square-free smooth numbers.
Differential Geometry
Unbounded normalized scalar curvature integrals in dimension four
We give counterexamples to Yau's question on normalized scalar curvature integrals in every dimension $n\geq4$. For each such $n$, there exists a smooth complete Riemannian metric $g$ on $\mathbb{R}^n$ with nonnegative Ricci curvature and a pole such that \[ \lim_{R\to\infty}R^{2-n}\int_{B_q(R)}\mathrm{Scal}_g\,\mathrm dV_g=+\infty \] for every fixed $q\in\mathbb{R}^n$. Here $B_q(R)$ denotes the geodesic ball of radius $R$ centered at $q$, and $\mathrm{Scal}_g$ is the scalar curvature of $g$.
Read the full abstract
Computer Vision
Geometry Without Coordinates: LiDAR Diffusion as a 3D Feature Bridge
Transferring the rich priors of large 2D foundation models to sparse 3D LiDAR remains challenging, as training native 3D foundation models at comparable scale is limited by data and annotation scarcity. We introduce a LiDAR-conditioned diffusion model trained on pseudo-labels from off-the-shelf 2D foundation models. The model supports multiple output modalities, including depth, semantic segmentation and instance prediction, selectable via a textual task prompt. Because the model is conditioned on LiDAR, both its outputs and its intermediate UNet features can be projected back onto the input point cloud, enabling analysis of a 3D representation learned entirely under 2D supervision. We study this representation directly in point-cloud space, explicitly excluding raw spatial coordinates to isolate feature content from projection geometry. Linear probes recover up to ~23% Mean Intersection over Union (MIoU) on 3D semantic classes, compared to ~3.5% for a matched Gaussian-noise control, indicating substantial non-trivial structure. Pairwise cosine similarity across modality-specific feature streams reveals a layered organization. Early encoder layers remain weakly aligned across modalities while individually decodable, intermediate layers converge toward a shared representation, and decoder layers re-specialize toward task-specific outputs. These findings indicate that LiDAR-conditioned diffusion models can induce structured 3D representations from 2D supervision alone, with a modality-dependent manifold that locally unifies near a shared bottleneck. This positions diffusion as a viable mechanism for transferring large-scale 2D priors into sparse 3D domains.
Read the full abstract
Computation and Language
On-Policy Distillation for Vision-Language Model Adaptation, an Effective Paradigm on Low-Quality Multimodal Data
Knowledge distillation offers an efficient route to transfer a task-adapted vision-language teacher to a compact student. The training target in current vision-language distillation methods is typically constructed from the teacher prediction and applied uniformly to all training samples, making it unreliable under class and domain shifts. In this paper, we argue that distillation target construction should be treated as a dynamic training decision rather than a fixed recipe. To this end, we propose OnPoKD, an on-policy distillation framework for vision-language model adaptation. To the best of our knowledge, OnPoKD is the first framework that applies on-policy distillation to vision-language model adaptation by learning target construction as a policy decision. OnPoKD learns a lightweight controller that constructs sample-wise adaptive targets using reliability and disagreement cues from the teacher model, student model, and zero-shot prior. Instead of relying on a fixed teacher prediction, the controller dynamically balances teacher supervision, zero-shot prior guidance, and hard-label anchoring through bounded policy actions, allowing the distillation target to adapt to varying sample reliability and training stages. The policy controller is updated with validation feedback, encouraging target construction to optimize transferability rather than merely fitting the training distribution. Since the controller is only used during training, OnPoKD can be seamlessly integrated into existing vision-language distillation pipelines while preserving the original inference architecture and test-time cost. Extensive experiments on Base-to-novel generalization and Cross-dataset transfer benchmarks show that OnPoKD consistently improves over strong vision-language distillation baselines.
Read the full abstract
General Relativity and Quantum Cosmology
Kaniadakis Holographic Dark Energy with Particle Horizon as IR Cutoff
We construct a holographic dark-energy model using Kaniadakis entropy with the particle horizon as the infrared cutoff, consistently modifying both the holographic density and the Friedmann background. In standard Einstein gravity, noninteracting particle-horizon holographic dark energy (HDE) does not produce the sufficiently negative pressure required for late-time accelerated expansion. We show that the Kaniadakis deformation changes this behavior while the particle horizon is retained. For the representative parameter choice $K=1.90\times10^{-36}$, with $\Omega_{\rm DE0}=0.7$ and $c^2=0.64$, the deceleration parameter changes sign at $z\simeq0.585$ in the noninteracting case. Including the interaction strengths $b^2=0.03$ and $0.06$ changes the transition only slightly, giving $z\simeq0.586$ and $0.589$, respectively. Thus, for the parameter set considered here, the interaction does not generate the acceleration; rather, the transition is already present in the noninteracting Kaniadakis model and the interaction produces only a small shift in its timing. We derive the autonomous evolution equations, the effective dark-energy equation of state, and the deceleration parameter. The adiabatic squared sound speed is also examined as a diagnostic of the effective-fluid stability, while statefinder variables are used to characterize deviations from $\Lambda$CDM. Finally, we verify that the $K\to0$ limit continuously recovers standard Einstein-gravity particle-horizon HDE, for which the noninteracting model remains decelerating.
Read the full abstract
Astrophysics of Galaxies
Extending the Little Red Dot population at intermediate redshift with VIPERS
Little red dots (LRDs) are a recently identified population of compact, red sources characterised by a distinctive V-shaped UV-to-optical continuum and broad emission lines. Their physical structure and evolution remain uncertain, while their properties at intermediate redshifts are largely unexplored. We identify and characterise LRD-like sources at 0.5<z<1.75 using the VIMOS Public Extragalactic Redshift Survey (VIPERS), extending observational studies to a previously poorly explored cosmic epoch. We select candidates from their UV and optical slopes measured from broadband photometry and refine the sample using broad emission lines in VIPERS spectroscopy. We investigate their photometric, spectroscopic, and morphological properties, constrain their physical sizes with HSC imaging, measure their comoving number density and evolution, and characterise their local environments using the VIPERS galaxy density field. We also model the multi-wavelength emission of the unique X-ray-detected source with CIGALE, including X-ray and AGN emission. We identify 14 LRD-like sources, including one X-ray-detected object, whose HSC morphology remains unresolved. We measure their number density in two redshift bins, 0.5<z<1 and 1<z<1.75, and combine our results with literature measurements. We find a rapid decline in number density after cosmic noon. The two sources with sufficiently constrained environments are located in underdense regions. The SED of the X-ray-detected source is best reproduced by a radiatively efficient accretion disk viewed at low inclination. Our results demonstrate that LRD-like sources are present at z>0.5, bridging local and high-redshift studies. Their abundance, compact morphology, environmental properties, and multi-wavelength emission provide new constraints on the evolution and physical origin of this population.
Read the full abstract
Differential Geometry
Alexandrov-Type Rigidity for Minimal Capillary Hypersurfaces on Rotational Supports
In this paper, we prove that, under suitable conditions on the generating profile, every connected, compact embedded minimal capillary hypersurface supported on a one-ended rotational hypersurface in the Euclidean space is a horizontal slice. We also construct a smooth one-ended rotational support carrying a non-horizontal planar capillary $n$-ball, showing that the main slope condition cannot in general be omitted. For obtuse contact angles, we further obtain a catenoid-band classification under a reverse profile inequality.
Read the full abstract
Computer Vision
Decoupled Self-Forcing Distillation for Streaming Talking Head Generation
Streaming talking-head generation produces each frame as its driving audio arrives, yet fidelity and efficiency have so far pulled in opposite directions: end-to-end methods condition a video diffusion model on audio directly and achieve high quality but only at large scale, while cheaper two-stage methods generate an intermediate motion representation and trail in fidelity. We argue the cost of the former lies in the target of fusion: the video latent is dominated by identity, appearance and background, none of which audio bears on, so coupling audio to every pixel blurs detail and wastes capacity. We instead fuse conditions in a low-dimensional identity-disentangled motion space, routing audio and motion captions by their temporal granularity, and generate motion latents with a small causal autoregressive transformer that a pretrained diffusion renderer turns into video. Conditions thus control video transitively, and high fidelity no longer requires a large backbone. Streaming this decomposition needs both models to be causal, and the exposure-bias problem could be solved by self-forcing given a bidirectional teacher. But there is no such teacher in motion space. Our decoupled self-forcing distillation resolves both models under one frozen teacher: conditioned on motion, it distills the renderer into a block-causal student; unconditionally, it scores rendered rollouts against real videos, supervising motion by the video it produces. This lifts the fidelity ceiling from the motion generator onto the stronger renderer. The two models run as parallel causal streams, reaching 15.4 FPS at 1.3 s latency with no quality degradation.
Read the full abstract
Software Engineering
Ensembling LLMs for AI-Augmented Cybersecurity Software Requirements Generation
Translating high-level controls from security standards into concrete, system-specific requirements is central to cybersecurity requirements engineering. Large language models (LLMs) can accelerate this labor-intensive, recall-sensitive task, but any single run is unreliable: it misses valid safeguards while introducing plausible hallucinations, and outputs shift across runs and models. We reframe this variability as a resource: rather than selecting one output, we study post-generation ensembling, aggregating stochastic runs with information-retrieval data-fusion operators. We propose two strategies: Uniform fusion rewards mere cross-run agreement, whereas Naive-Bayes fusion weights each run by its estimated reliability. We evaluate both over 24 runs from 12 configurations across four model families, generated for ten ISO/IEC 27002:2022 controls and expert-judged against a gold standard of 72 valid requirements. Pooling every run's output recovers all 72 (whereas single configurations recover on average under half) but also 111 hallucinations. Fusion separates the wheat from the chaff, ranking valid requirements well ahead of hallucinations. In the areas under the precision-recall and ROC curves, Uniform fusion alone largely surpasses every original run and configuration by 0.142 and 0.118 over the best configuration. Naive-Bayes weighting adds a further 0.039 and 0.052, reaching 0.864 and 0.869 while attaining useful operating points earlier. Internal validation confirms the stability of these gains: they stay positive in at least 92% of out-of-bag bootstrap resamples and every structured-perturbation sample. Post-generation fusion thus turns apparent noise into a practical asset: a lightweight layer giving analysts broader coverage and a better prioritized review queue, using affordable, below-frontier models alone.
Read the full abstract
Artificial Intelligence
TRACE: Training Reasoning Agents for Causal Exploration with Synthesized Rewards
Reinforcement learning with verifiable rewards (RLVR) has advanced language-model reasoning in domains such as mathematics and code, where objective answers are inexpensive to check. Diagnostic reasoning over complex data lacks this advantage: establishing the true cause of an anomaly often requires costly expert investigation and may remain ambiguous after the fact. We ask whether this asymmetry of verification can instead be engineered. We sample an intervention, inject it into a controlled simulator, and generate the observations it would produce. The hidden intervention provides an oracle label and objective reward, while the agent must still investigate noisy, confounded, and distributed evidence. We instantiate this approach in TRACE, a digital-advertising diagnostic environment with 12 root causes and fine-grained segment attribution. Agents investigate each episode using Python and SQL and must identify both the root cause and, when applicable, the affected segment assignment. On a held-out 235-episode test set, the strongest prompted baseline, Claude Opus 5, reaches 0.686 FullAttr@1. Supervised fine-tuning raises Qwen3.5-35B-A3B from 0.159 to 0.637, and subsequent RL with synthesized rewards reaches 0.757, outperforming all evaluated prompted baselines, including frontier closed-source models and a prompted Qwen3.5-122B-A10B model. The resulting policy also uses substantially fewer tool calls than the prompted 35B base. These results provide evidence that access to a scalable, objective training signal can be a more important constraint than model scale alone. More broadly, simulation-based verification can make otherwise ambiguous diagnostic reasoning tasks amenable to scalable reinforcement learning.
Read the full abstract
Optimization and Control
An $O(1/T^3)$ algorithm for minimizing convex quadratic functions over the $L_1$ ball
We study convex quadratic minimization over the unit $L_1$ ball in which the maximum eigenvalue of the Hessian matrix is bounded by a positive constant $L$. We propose a novel first-order algorithm with objective value error bounded by $O(L/T^3)$ after $T$ gradient evaluations, assuming that the subproblems involved in the algorithm can be solved exactly. To the best of our knowledge, the best convergence rate of algorithms in the literature is $O(L/T^2)$. From the perspective of information-based complexity theory, our proposed algorithm is the first in the literature that achieves the $O((L/\varepsilon)^{1/3})$ first-order oracle complexity, although its current version is not necessarily practical for implementation. We hope that our proposed algorithm could shed some light on future implementable and efficient $O(L/T^3)$-convergence-rate algorithms. The proposed algorithm incorporates a decomposition of components of vectors in the unit $L_1$-norm ball to "good" and "bad" parts, and uses symmetric rank-1 (SR1) updates on the bad parts. The proposed algorithm was developed after the author instructed the OpenAI ChatGPT 6 (Astra) model to study the problem using ideas of weak-type $L^1$ estimates and good-bad part decomposition in harmonic analysis and a recent result.
Read the full abstract
Statistical Methodology
Semiparametric Inference for Conditional Shapley Feature Importance
Shapley values are widely used for post-hoc feature attribution, but most estimators return point quantities and do not quantify uncertainty, and popular implementations sample out-of-coalition features from their marginal distribution, which misattributes importance when features are dependent. This paper studies the conditional formulation, in which out-of-coalition features are integrated out under their true conditional distribution. The target is a global, loss-based importance that pairs a conditional value function with a SAGE-style loss aggregation. We propose a one-step estimator with K-fold cross-fitting and a U-statistic correction of the squared loss that removes the Monte Carlo bias of the naive plug-in; it is $\sqrt{n}$-consistent and asymptotically normal under double-robust rate conditions, and the resulting Wald interval attains nominal coverage. A Pinsker-type bound quantifies the bias from misspecifying the working copula class, while vine copulas keep conditional sampling tractable. In a Gaussian design study with n= 500, the empirical coverage of the 95% interval lies between 0.91 and 0.96 across all features, the test holds its Type-I rate at 0.05, and it reaches power one for moderate signals. Applied to the UCI Concrete and California Housing data, the method identifies the conditionally informative features with Bonferroni-controlled significance.
Read the full abstract
Quantum Physics
Programmable photonic state fusion via heralded storage of asynchronously generated resources
Probabilistic photonic sources generate elementary states in different trials, whereas multiphoton protocols require them to interfere in common temporal modes. We propose a fusion protocol that overcomes this mismatch by successively loading independently heralded photonic states into active storage loops. Conditioning on vacuum in monitored dump modes selects events in which each newly generated state is transferred into the same circulating modes as the photons already stored, thereby removing its generation time label. Provided the two alternatives of each elementary state undergo the same loading transformation, the accumulated state is described by a product of programmable linear factors, allowing a target superposition to be constructed by polynomial factorization. Adjusting the storage loop coupling as the state grows substantially improves the loading efficiency, changing the faster-than-exponential penalty of fixed balanced couplers to exponential scaling. We apply the protocol to two-photon path–frequency states for photonic clock interferometry and estimate the detected rate including source waiting time and round-trip loss.
Read the full abstract
Machine Learning
One Loop, Two Gains: Can Active Learning win the Lottery for Free?
The lottery ticket hypothesis posits the existence of winning tickets: sparse subnetworks that, when trained in isolation from their original initialization, match the accuracy of the full dense network. The predominant method for discovering such tickets, iterative magnitude pruning, alternates pruning with full retraining from scratch until convergence over many cycles. Similarly, deep active learning also retrains a model from scratch after each acquisition round as new labels become available. Despite this shared reliance on iterative retraining with a substantial computational overhead, the two paradigms have been studied separately. We observe that the iterative training loop inherent to pool-based active learning already provides the exact computational structure that iterative magnitude pruning exploits, and propose Improve & Prune (I&P), a method that integrates magnitude pruning into each active learning retraining cycle at practically no additional cost. This raises a key empirical question: can iterative magnitude pruning produce winning tickets under the non-stationary data regime of active learning? We investigate this question across multiple acquisition functions, architecture families, and image classification datasets, including an active fine-tuning scenario. Our results demonstrate that I&P yields sparse, deployable models at each active learning iteration. Those match the accuracy of their dense counterparts at sparsities up to 95%, effectively obtaining winning tickets as a byproduct of the active learning pipeline. These per-iteration sparse models can address two computational bottlenecks - per-round model retraining and acquisition scoring over the unlabeled pool - that currently prevent the practical adoption of DAL on large architectures and large unlabeled pools.
Read the full abstract
Materials Science
Constrained Optimization of Higher-Order Cluster-Expansion Hamiltonians for Alloys Using Simulated Bifurcation
Identifying ground-state and low-energy atomic configurations is a central problem in alloy design. The cluster-expansion (CE) method represents configurational energetics on a fixed lattice as an effective Hamiltonian; for binary alloys, higher-order CE models become polynomial Ising Hamiltonians. Using Au-Cu as a model binary alloy, we formulate cubic and quartic cluster-expansion Hamiltonians as penalty-augmented polynomial unconstrained binary optimization (PUBO) problems under fixed-composition constraints. We optimize these PUBO problems using SQBM+, a simulated-bifurcation-based solver that can treat higher-order polynomial binary objectives directly. This direct PUBO treatment avoids the need to construct an explicit quadratic reformulation with auxiliary variables. Composition constraints are imposed through quadratic penalty terms, whose weights are estimated from derivative coefficients of the continuous relaxation of the CE objective. Benchmark calculations for systems up to 2048 atoms show that SQBM+ robustly obtains low-energy feasible configurations for cubic CE models and remains effective for many quartic instances. Formation-energy convex hulls constructed from the optimized configurations recover the CuAu and Cu3Au ordering trends and reveal finite-size effects at off-stoichiometric compositions. These results demonstrate simulated bifurcation as a practical route to constrained higher-order CE optimization for alloy configuration search.
Read the full abstract
Number Theory
Quasi-modularity of symmetric quasi-shuffles
We develop an algebraic framework for the quasi-modularity of symmetric multiple $q$-zeta values. We identify natural classes of symmetric quasi-shuffles whose $q$-zeta values exhaust the algebra of level-one quasi-modular forms, and further classes whose $q$-zeta values are quasi-modular forms of finite level. We also obtain explicit symmetrization formulas and relate symmetric quasi-shuffles to symmetric and quasisymmetric functions.
Read the full abstract
Robotics
Deformable Object Manipulation under Partial Observability via Real-Time Full-Shape Estimation
Manipulating deformable objects (DOs) is challenging due to their high-dimensional state space, underactuated dynamics, and partial observability. In this paper, we propose cRVAE, a lightweight conditional recurrent variational autoencoder that estimates the full DO state from only partial corner-node observations during inference. The resulting model is used as the forward model in a receding-horizon optimal control framework for obstacle-aware collaborative DO manipulation. In simulation on rope and fabric, cRVAE estimates the full DO state from the available corner-node measurements alone, matching the accuracy of a parameter-identified XPBD model. At inference it uses no physical parameters as model inputs and performs no online parameter identification. It also runs approximately 350 times faster on the rope and over 1500 times faster on the fabric per forward pass, keeping horizon-based planning within the 100 ms control budget where XPBD exceeds it already at short horizons. Full-shape estimation from corner sensing at in-loop speed is what makes the model deployable on hardware, which we demonstrate on a Unitree Go2 robot.
Read the full abstract
Machine Learning
View-Structured Conformal Prediction for 3D Gaussian Splatting
3D Gaussian Splatting (3DGS) renders novel views in real time, but an uncertainty heatmap does not certify that a rendered view meets a certain prediction coverage. We treat novel-view synthesis as structured regression and ask that, with probability at least $1-\alpha$, RGB prediction boxes cover at least a $1-\beta$ fraction of pixels in a new view. We propose View-Structured Conformal Prediction (VSCP). It splits the pre-calibration scale into a spatial shape from the renderer and a transferable view-difficulty factor, which predicts the smallest view-wise multiplier that shape needs. A held-out quantile over views (View-CP) then gives finite-sample validity even when transferring to new scenes. The same factorization makes the analysis exact: a conformity score is the ratio of oracle to predicted view difficulty, and excess width separates into a test-side and a calibration-side term. Across 13 real scenes, pixel-pooled calibration reaches 89.9% marginal pixel coverage but only 61.4% view-event coverage at a 90% target, while View-CP reaches 91.7–92.0%. At matched coverage VSCP cuts width by 22.1% against a constant scale, and matches a ten-model ensemble's 21.0% reduction using only one model per scene and four rather than ten rasterization passes per query. VSCP also improves on the closest single-model baseline, the 3DGS-U field, by 4.7 points ($p=0.0225$). The view predictor transfers from bounded source families to all nine unbounded Mip-NeRF~360 scenes. There the full scale beats the constant scale with 20.7% width saving on all nine scenes. It also keeps an 18.3% saving under a different densification backbone and runs at 216–280 FPS on an RTX~4090.
Read the full abstract
High Energy Physics (Phenomenology)
An improved partial-wave projection of the one-particle exchange in relativistic three-body scattering
A critical component of relativistic three-body scattering amplitudes is the one-particle exchange (OPE) process, wherein a single particle is exchanged between incoming and outgoing two-body subsystems. We present an improved, finite-sum expression for the partial-wave OPE that is valid for three massive, spinless particles in arbitrary angular momentum configurations. A collection of analytic and numerical checks demonstrate that our finite-sum expression reproduces the known properties of the partial-wave OPE, including its threshold scaling behavior and singularity structure. Compared with previous work, this result more efficiently handles the rapid proliferation of partial waves contributing to any total angular momentum $J$, enabling the construction of robust wavesets for three-body amplitude analyses.
Read the full abstract
Computation and Language
RiLM: Parameter-Efficient Language Modeling via Geodesic Decoding
Language models under one million parameters matter for edge deployment, domain adaptation, and reproducible research, yet a two-layer LSTM or Transformer at embedding width d = 128 still spends roughly one third of its capacity on the output matrix W_out in R^(d x |V|). We propose Riemannian Language Models (RiLM), which remove that layer entirely: context unfolds as a trajectory on a Riemannian manifold, and next-token probabilities arise from squared geodesic distance between the current state and vocabulary embeddings. The same embedding map serves input and output – decoding is geometry. We instantiate the framework on flat R^d (Flat RiLM) and the Poincare ball H^d (HypRiLM) with a shared MLP composition map phi (~290k parameters, d = 128, |V| = 2000). Across five seeds on WikiText-2, HypRiLM reaches 54.2 +/- 0.2 validation perplexity versus 87.6 +/- 0.6 for Flat RiLM; tied and matched LSTM, Transformer, and SSM controls remain at 113-147 PPL on WT-2 – HypRiLM leads by roughly 2x over the strongest tied recurrent baseline (SSM, 113.0 +/- 3.8). Penn Treebank and a 10k-vocabulary stress test confirm that geodesic decoding transfers across corpora and larger |V|, while hyperbolic curvature helps selectively. We also characterize boundary collapse in naive hyperbolic recurrence and show how Mobius stabilization restores trainability. Claims are scoped to controlled small-model comparisons, not full-vocabulary state of the art.