arXiv subjects

Search papers

Search indexed arXiv papers on artificial intelligence, large language models, computer vision and robotics. Read source abstracts and follow links to arXiv.

At least 1,135 records · Page 63Linked to original sources

arXiv2026

TerraNostra-Bench: Benchmarking Spatial Downscaling and Temporal Rollout for Global 1-km Land-Use Reconstruction

Uncertainty in the terrestrial carbon cycle remains a major constraint in climate projections, partly due to imperfect land surface representations in Earth system models. We propose a data-driven framework for high-resolution land-use reconstruction, treated as a semantic-segmentation task, that combines coarse-resolution scenario data and climate classification with static geophysical features to produce annual land-use maps at 1 km resolution. We introduce TerraNostra-Bench, a change-aware evaluation protocol that separates spatial downscaling from autoregressive temporal rollout and scores models on the rare land-use-change pixels rather than the dominant static background. Using it, we compare convolutional (U-Net) and transformer (Swin) backbones in autoregressive and non-autoregressive schemes against persistence, linear, and shallow convolutional baselines. The autoregressive prior is the most informative input yet accumulates error over long rollouts: autoregressive models lose 5.7-7.2 accuracy points over 13 years, against 2.2-2.8 for non-autoregressive models. This trade-off is governed by the autoregressive choice rather than the backbone. It can be tuned by an inference-time prior-trust threshold that requires no retraining and by the training-time weighting of reconstruction and change objectives; the two levers trace complementary frontiers, with calibration recovering up to 10 accuracy points while keeping change sensitivity above the persistence floor. Extrapolating beyond the training window, backward to 1955-1959 and forward to 2015-2019 under scenario forcing, the models retain most of their in-window skill and show the same trade-off, indicating it is intrinsic to the design. This work provides the methodological foundation and evaluation surface for a forthcoming temporally continuous land-use product.

physics.ao-ph↗

arXiv2026

Multi-View Latent Action Pretraining for Cross-Embodiment Robot Learning from Videos

Action-free videos are an appealing substitute for costly teleoperated data. Latent action models (LAMs) turn videos into a supervision signal by inferring abstract actions between frames. Existing LAMs observe a single camera view, which resolves 6D robot actions only ambiguously missing out-of-plane movements and potential occlusions. We propose a LAM that ingests synchronized multi-camera views using cross-attention to resolve depth and occlusions before compressing them into latent actions. The latent actions are fed into a policy in which a latent expert and an action expert are coupled by sequential flow matching around a frozen VLM backbone. We further evaluate in strict cross-embodiment settings where task-specific demonstrations come from other embodiments. We show that dual-view LAM pretraining improves the decodability of ground-truth 6D actions by 17% compared to single-view, while scaling single-view data by two orders of magnitude is not sufficient to fully recover out-of-plane motions. The downstream policy improves success rate by +13.5% compared to +6.2% for single-view LAM. A single cross-embodiment is sufficient for skill transfer, and adding a second viewpoint outperforms adding up to four further embodiments. The results transfer to hardware where demonstrations come from a human hand improving success rate by +20% over action-only policies in 850 rollouts. We show that video and robot data are complementary, and that benefits of learning from video are task-dependent. Website: https://learnsyslab.github.io/multiview_latent_action_pretraining/

cs.RO↗

arXiv2026

Metrization of the Fell Topology and the Wijsman Topology for the Box Metric on Closed Downward Families

We study nonempty Box-closed families of metric measure spaces that are downward under the Lipschitz order. If a space is below a member in this order, it also belongs to the family. Such families need not be pyramids, because directedness is an additional condition. We define a metric on these families by taking the supremum, over all finite metric measure spaces, of the normalized absolute difference between the infimal Box distances from each finite space to the two families. We divide each difference by the sum of the support cardinality and diameter of the finite test space. The topology induced by this metric agrees on the full domain with the relative Fell topology and with the restriction of the Wijsman topology for the Box metric. On the subspace of pyramids, it induces the weak topology and makes the natural inclusion into the hyperspace of nonempty closed subsets a topological embedding for either hyperspace topology. Pyramids associated with metric measure spaces are dense in the pyramid space, which is the completion of metric measure spaces under the metric pulled back along the map that sends each space to its associated pyramid.

math.MG↗

arXiv2026

Absolute winning properties of exceptional sets for rational maps

We study Schmidt's game on the Julia set $J(f)$ of weak hyperbolic rational maps $f:\CC\to\CC$. For each $y\in J(f)$, let $E(f,y)$ be the set of points in $J(f)$ whose forward orbits stay a positive distance from $y$. We prove that $E(f,y)$ is absolute winning on $J(f)$ for each $y \in J(f)$. To do this, we first prove the absolute winning property for the exceptional set associated to the critical points by a potential game argument, and then use uniform expansion to obtain the result for each target point $y$.

math.DS↗

arXiv2026

Toward Agentic AI for the 6G Physical Layer: From Algorithm Discovery to Verified Baseband Silicon

6G will be designed faster than it can be built: AI-driven search already proposes physical-layer algorithms at machine speed, while carrying a single one onto embedded baseband silicon remains specialist work. That silicon imposes hard memory and numeric limits, but also exposes cost structures a GPU lacks, and a design flow that never runs on the target sees neither. We present an agentic workflow in which an LLM coding agent implements algorithms on a commercial baseband SoC, optimizes them down to agent-written vector assembly, and validates them on silicon. The agent is guided by a knowledge base of measured platform facts, checked at every step against evaluators it is instructed never to edit, and metered per phase. We showcase the workflow on an AI-discovered orthogonal time frequency space (OTFS) detector portfolio on the open 6G4All platform. It yields twelve kernels that run cycle-exactly on silicon, their optimized versions at 18k-405k cycles (30-660 $μ$s) per frame. It also yields a new detector shaped by the chip's measured costs that needs 13.9% fewer cycles than the port of the discovered detector while matching its error rate, with a measured edge our evaluation cannot statistically resolve, a structure that a hardware-blind search objective had no reason to find. What is not specific to OTFS is the workflow: an LLM agent exploring an implementation space under real hardware limits, with the cost of every choice measured rather than assumed.

cs.AI↗

arXiv2026

Accelerating real-time hybrid-functional calculations with learned ACE and symmetry-based FFT reuse

Real-time hybrid-functional calculations require repeated reconstruction of nonlocal exchange, whose cost grows rapidly with Brillouin-zone sampling. We develop two methods to improve the efficiency of time evolution and spatial processing. The first learns the time evolution of adaptively compressed exchange (ACE) factors and predicts their updates, reducing the frequency of explicit reconstruction. The second uses crystal symmetry to reuse exchange fast Fourier transforms (FFTs) among equivalent real-space points. This symmetry-based FFT reuse (SBFR) reduces the number of FFTs while retaining the full $\bm k$-point sum. The methods can be used with a variety of hybrid functionals, and we use HSE06 in this work. Relative to an ACE reference using full-row mesh FFTs, learned prediction provides more than fivefold acceleration, SBFR more than fourteenfold, and their combination more than sixty-threefold. The combined method shows efficient parallel scaling. After learning, propagation takes less than 0.5 s per step. The reduced computational cost allows us to study nonlinear and nonequilibrium electron dynamics with hybrid functionals and to examine the role of excitonic effects in semiconductors and insulators.

cond-mat.mtrl-sci↗

arXiv2026

New Directions in Irrationality

Since Apéry's proof of the irrationality of $ζ(3)$, decades of work have explored integrals, hypergeometric identities, and recurrences with exceptional arithmetic properties, to discover the strongest quantitative bounds on irrationality measures for zeta values and related constants. Yet, most findings have relied on specially tailored families of such constructions, making their systematic extension difficult. Here we show how these constructions can be connected and generalized into a common multidimensional structure based on conservative matrix fields (CMFs), enabling a systematic discovery of new irrationality proofs and quantitative bounds for irrationality exponents. CMFs are finite-dimensional cocycles of compatible rational matrix shifts, and are shown to capture all the classical constructions developed since Apéry as one-dimensional trajectories. The initial point encodes the mathematical constants; varying the trajectory within a suitable convergence region produces different approximation schemes for the same constants. We use this structure to improve the irrationality measures of famous constants, yielding for example $μ(ζ(2))<4.62675$ and $μ(ζ(3))<5.4197$, and proving that $1,ζ(3),ζ(4)$ are linearly independent over $\mathbb{Q}$, yielding $μ(Aζ(3)+Bζ(4))<66.3$ for every $(A,B)\in\mathbb{Q}^2\setminus\{(0,0)\}$. The same approach demonstrated over other CMFs provides irrationality proving trajectories for infinitely more mathematical constants. We further construct higher-rank CMFs with explicit initial states for $1,ζ(2),ζ(3),\ldots,ζ(d+1)$, and identify the arithmetic conditions needed to obtain further, higher-order independence results.

math.NT↗

arXiv2026

Decoupled Representation Learning for Face Presentation Attack Detection under Smart Glass Capture

Face presentation attack detection is typically studied under fixed camera or handheld acquisition, whereas smart glass capture introduces a more challenging setting with ego motion, changing viewpoint, background clutter, distance variation, and less controlled illumination. To address this challenge, we present \emph{SmaGla-FPAD}, a smart glass based face presentation attack database acquired using Ray-Ban Meta Wayfarer Smart Glasses across indoor, outdoor, and in-car settings. The database contains 297,000 sample images with both bona fide and silicone mask based attacks. We further propose \emph{DeCoupPAD-FAS++}, a decoupled representation learning framework that separates spoof related evidence from nuisance sensitive capture variation through complementary attack and style branches, pseudo-domain supervision, adversarial suppression, and prompt guided semantic alignment. Experiments on SmaGla-FPAD show that the proposed method achieves the best balance between attack rejection and bona fide acceptance in both in domain and cross scenario evaluation. Ablation results further confirm the contribution of disentanglement, pseudo-domain supervision, and prompt based regularisation under smart glass capture.

cs.CV↗

arXiv2026

Looks Fine to Me: Understanding Oversight in Agentic Interactions

Agentic AI systems are deployed to execute complex tasks across domains such as writing, coding, and data analysis. As execution is offloaded to AI agents, the human role shifts toward oversight. Effective oversight requires recognizing when closer scrutiny is needed, even though flawed outputs may appear convincing. Our formative study (N=12) revealed that users accepted plausible outputs without thorough evaluation, and only recognized these lapses in retrospect. To examine this systematically, we conducted a controlled within-subject study (N=36) in which participants completed programming tasks with an agent that produced correct outputs, silent defects, or visible failures. Combining task outcomes, interaction logs, gaze, physiological signals, and self-reports, we examined perceived, behavioral, and effective oversight. Participants were often highly confident while missing silent defects: they performed generic checks, failed to generate diagnostic evidence, or even misinterpreted defect evidence. We discuss how studying oversight across multiple dimensions can support future work on timely, task-calibrated re-engagement.

cs.HC↗

arXiv2026

Collective Hall quantization in 3D Chern vector systems

Understanding the three-dimensional (3D) quantum Hall effect, characterized by a Chern vector rather than a single Chern number, has been a central pursuit in topological physics. While quantized Hall transport is established for the special case with one nonzero Chern component, recent realizations of more general Chern-vector states raise a fundamental question: how does such topology protect quantized transport? Here, we predict a distinct 3D quantized Hall effect protected by $\mathbf{C}=(0,m,n)$. For samples with dimensions $L_y$ and $L_z$, global topology enforces collective quantization of $mL_y+nL_z$, despite the fragility of individual edge modes to disorder. This reveals a novel transport regime in which robust quantization emerges as a holistic property of a unified chiral network,rather than of its constituent modes. Crucially, this collective response manifests as a quantized Hall conductance $G_{xy}=\pm(mL_y+nL_z)e^2/h$, with a distinctive dependence on both Chern-vector components and sample dimensions. We propose candidate materials to facilitate experimental observation.

cond-mat.mes-hall↗

arXiv2026

The strong type endpoint estimate for the local spherical maximal function

Let $p_d=(d^2+1)/(d(d-1))$ and $q_d=(d^2+1)/(d-1)$. We prove that the local spherical maximal operator $\sup_{1\leq t\leq 2}|f*dσ_t|$ is bounded from $L^{p_d}(\mathbb{R}^d)$ to $L^{q_d}(\mathbb{R}^d)$ for every $d\geq 2$. Previously only the restricted weak type estimate was known at this endpoint. Together with known results, this completes the characterization of the strong type $L^p\to L^q$ bounds for the local spherical maximal operator. We also prove the maximal estimate \[ \Big\|\sup_{t\in\mathbb{R}}|e^{it\sqrt{-Δ}}f|\Big\|_{L^{2(d+1)/(d-1)}(\mathbb{R}^d)} \lesssim\|f\|_{\dot H^{d/(d+1)}(\mathbb{R}^d)} \] for the wave equation at the critical exponent. Both results are obtained by exploiting the gain in bilinear estimates that comes from the interaction between high and low frequencies.

math.CA↗

arXiv2026

Weyl Bounds for $\operatorname{GL}_2\times \operatorname{GL}_2$ Rankin--Selberg $L$-functions via Short Spectral Moments

Let $f$ and $g$ be Hecke--Maass cusp forms of $\operatorname{SL}_2(\mathbb{Z})$, with spectral parameters $t_f$ and $t_g$, respectively. Let $T$ be a large parameter and put $L = |T-t_g|.$ We prove the Lindelöf-on-average estimate \[ \sum_{ |t_f-T|\leq H} \left|L\!\left(\frac12,f\otimes g\right)\right|^2 \ll_{\varepsilon} T^{1+\varepsilon}H, \] uniformly in the range \[ T^{1/3+\varepsilon}\ll H,\qquad HT^\varepsilon\ll L\leq T. \] As an immediate consequence, in the generic range when both $|t_g|$ and $L$ are of size $T$, we obtain the Weyl bound \[ L\!\left(\frac12,f\otimes g\right) \ll_\varepsilon C(f\times g)^{1/6+\varepsilon}. \]

math.NT↗

arXiv2026

Certified Adaptive List Decoding for Error-Correcting Codes

Maximum-likelihood (ML) decoding of linear block codes is NP-hard, so neural decoders are evaluated by how closely they approach it. We observe that proving a candidate optimal can be far easier than finding it, and build a decoder that does both. We cast decoding as a search over an absorbing-state discrete diffusion model, whose sequential commitment of coordinates turns generation into a tree, and run a probability beam search over it. A dual bound from a linear-programming relaxation and a distance-based test, both derived from the code, then certify per frame that the current candidate is the global ML solution; because both are theorems, certified frames incur no accuracy-computation trade-off. Across $15$ codes and five operating points we certify up to $99.9\%$ of frames, the two tests dividing the space as their derivations predict, one strong for sparse parity-check matrices and the other for codes with a large minimum distance. Certification also behaves as a per-frame reliability signal, with $96.7$-$100\%$ of all frame errors falling on uncertified frames. Against an exact ML reference at $5$ dB the decoder is within one standard deviation on three of the four codes where the reference is uncensored, and against four neural baselines it attains the best bit-error rate in $71$ of $75$ cells.

cs.IT↗

arXiv2026

Probing the Optical/UV Continuum Variability of Active Galactic Nuclei with the Damped Random Walk Model

Optical/UV continuum variability over baselines of up to 20 yr is investigated using the damped random walk (DRW) model for 1214 active galactic nuclei (AGNs) compiled from Y. Shen et al. and C. J. Burke et al. Among these, 115 AGNs with reliable DRW fitting and unbiased DRW parameters (timescale $τ_{2500}$ and amplitude $σ_{2500}$ at the rest-frame 2500~Å) are used to examine the relations between the variability and the accretion properties. We found a significant correlation of $τ_{2500}$ with black hole mass ($M_{\rm BH}$) and dimensionless accretion rate ($\dot{\mathscr{M}}$) of the form $τ_{2500} \propto M_{\rm BH}^{0.50\pm0.07} \dot{\mathscr{M}}^{0.27\pm0.04}$, while the dependence on the continuum luminosity at 5100~Å and redshift becomes insignificant after controlling for $M_{\rm BH}$. $τ_{2500}$ also correlate with the mass accretion rate ($\dot{M}$) as $τ_{2500} \propto \dot{M}^{0.41\pm0.04}$, which is consistent with the thermal timescale expected from the standard thin-disk model within $3σ$ uncertainties; while the average value is consistent with the expectations of the model, the scatter in the relation can be due to differences in the viscosity parameter ($α$) or in disk temperature profiles. Assuming that \tauuv\ corresponds to the accretion disk thermal timescale, the disk-emitting radius at rest-frame 2500~Å ($R_{2500}$) is found to correlate with $M_{\rm BH}$ as $R_{2500} \propto M_{\rm BH}^{0.53\pm0.05}$. Using the microlensing-based $R_{2500}-M_{\rm BH}$ relation of C. W. Morgan et al., $α$ is statistically estimated to be $0.04\pm0.01$.

astro-ph.GA↗

arXiv2026

What Shapes the Ultraviolet Spectra of Type Ia Supernovae?

Ultraviolet (UV) observations of Type Ia supernovae (SNe Ia) exhibit significant diversity, but their physical origin remains poorly understood. We investigate UV spectra of SNe Ia observed with the Swift UltraViolet and Optical Telescope. Our expanded pre-maximum sample probes dependencies on decline rate, ejecta properties, and host environment. We recover the correlation between UV flux and decline rate near maximum in the near-UV (NUV), and find correlations at pre-maximum phases in both the NUV and mid-UV (MUV). This trend is strongly wavelength dependent, with slow decliners showing higher NUV but lower MUV flux than fast decliners. Synthetic TARDIS spectra varying the 56Ni abundance show that enhanced 56Ni suppresses the MUV while increasing the NUV, broadly reproducing the trend. We find no significant correlation between UV flux and ejecta velocity, while Si II 6355 pEW shows trends similar to those with decline rate. We also find no significant correlations with host-galaxy properties, although host environment may contribute to the remaining UV diversity among SNe with similar decline rates. Early-excess SNe show enhanced flux mainly in the NUV rather than a broad UV excess, suggesting that strong circumstellar interaction is not necessarily required to explain the observed excess emission. Peculiar subclasses show distinct UV properties, with brighter and hotter subclasses showing higher UV flux, although the current sample is small and does not cover all SN Ia subtypes. Our results suggest that the UV diversity of SNe Ia is shaped by multiple mechanisms that contribute differently across the UV wavelength range.

astro-ph.HE↗

arXiv2026

Atomic identification via the Prandtl-Tomlinson stick-slip mechanism

Distinguishing atomic species is an ongoing challenge in the field of scanning probe microscopy. Typically, atomic force microscopy makes use of electrostatic interactions to perform atomic identification. Here we perform lateral force microscopy (LFM) measurements with a CO terminated tip of a partially oxidized Cu(110) surface and show clear differences between the metal and oxygen sites over a large z-range. This allows for atomic identification on a typical metal oxide. Density functional theory based calculations show that the tip-sample interaction is dominated by short-ranged orbital interactions and that the electrostatic contribution is negligible. At small tip-sample distances, LFM can probe the Prandtl-Tomlinson stick-slip mechanism on the single-atom scale. It is visible as a striking asymmetry in the forward and backward scan directions, experimentally revealing this fundamental model for nanotribology on the single-atom scale.

cond-mat.mtrl-sci↗

arXiv2026

Non-null ruled surfaces and pseudo hyper-dual tangent sphere bundles

In this paper, the tangent bundles of hyper-dual pseudo spheres are defined. Then, a new family of non-null ruled surfaces associated with hyper-dual Lorentzian curves is defined via the E.Study hyper-map in Minkowski space and the definitions of the tangent bundles of hyper-dual pseudo spheres. Within this framework, the resulting non-null ruled surfaces are classified. Necessary and sufficient conditions for developability are obtained with their geometric interpretations in terms of hyper-dual Lorentzian curves and arc-length parameters. Moreover, explicit examples are presented to verify the theoretical results.

math.DG↗

arXiv2026

Phonon Dispersion Relation in Weakly Coupled Bose Gas Beyond Bogoliubov Approximation

We study the quasiparticle dispersion relation for a homogeneous Bose gas in a large cubic box with periodic boundary conditions at zero temperature. Inspired by the Bogoliubov method, we rearrange the usual Bose gas Hamiltonian in a form adapted to perturbation theory in the weak-coupling regime, both in the canonical and grand-canonical setting. We compute the first non-trivial corrections to the dispersion relation in both settings. We show that for small momenta they are linear in the absolute value of the momentum. Thus, in this approximation phonons have a well-defined velocity, which we compute.

math-ph↗