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 811 records · Page 45Linked to original sources

arXiv2026

RAN-JEPA: An Action-Conditioned World Model for Energy-efficient Mobile Network Control

Operational mobile networks are rich in observations but poor in interventions: they continuously generate large volumes of passive telemetry, while controlled action--response trajectories remain comparatively scarce and costly to collect. This observation--intervention asymmetry is a central obstacle to building practical world models for RAN control. We address this problem with RAN-JEPA, an action-conditioned world model that separates representation learning from intervention learning. The model first learns a predictive latent representation from observation-only network data, adapts that representation to the target deployment domain, and then learns action-conditioned latent dynamics from controlled sleep/wake trajectories while keeping the encoder fixed. We evaluate RAN-JEPA using public operator performance data for pretraining and a real-equipment four-cell 5G testbed for domain adaptation, post-training, and control. Across four future 15-minute intervals, RAN-JEPA achieves lower action-conditioned prediction error than a GPT-2 world-model baseline and a mechanistic rule, while remaining close to a strong persistence reference. What-if evaluation further shows that the model produces distinct and coherent future trajectories under matched SLEEP and KEEP plans. In closed-loop control, RAN-JEPA consistently supports higher sleep ratios than the two baselines at comparable congestion cost. These results show that observation-first, action-conditioned learning can turn abundant passive telemetry and limited intervention data into a practical predictive model for energy-aware RAN control.

cs.AI↗

arXiv2026

Global existence of large strong solutions to the 3D compressible micropolar equations with density-dependent viscosities

In this paper, we consider the Cauchy problem for the three-dimensional (3D) compressible micropolar equations with density-dependent viscosities. By classical $L^p$ estimates to elliptic system, defining suitable energy functions, after the elaborate index analysis, we establish the global existence of strong solutions to the system with all viscosities depending density as long as the initial density is large enough. By product, We also prove the existence of global large strong solutions with only large initial shear viscosity $μ_1(ρ)$ and angular viscosity $μ_2(ρ)$, which consists with the research work \cite{ZZ2014} of 3D compressible Navier-Stokes equations with constant viscosities and improve the recent result of \cite{HLZ} in some extent where they requiring both initial shear viscosity and bulk viscosity to be large.

math.AP↗

arXiv2026

High-Throughput LLM-Based Compression and What It Takes: A Systems Study of Inference Engines and Entropy Coding Optimizations

Large language models are attractive text compressors because next-token probabilities can be converted into code lengths. In practice, however, the main bottleneck is often not modeling quality but systems cost: a compressor must obtain logits quickly enough to compete with strong classical baselines. This paper examines systems optimizations for LLM-based compression, including multistream pipelined arithmetic coding and target-interval capture, and evaluates their trade-offs across inference engines, configurations and datasets. We evaluate LLM-based text compression on three different inference engines: Transformers, vLLM, and TensorRT-LLM. Notably, we obtain a speedup of 18.84x compared to prior work's Transformers implementation and a maximum recorded inference speed of 181 kilobytes per second at a compression ratio of 7.17, when running on an NVIDIA A100.

cs.DB↗

arXiv2026

Inverse design of poles and zeros: tailoring phase control in metasurfaces

We develop a physically interpretable inverse-design framework for generalized Huygens metasurfaces, using prescribed scattering poles and zeros to engineer resonance interference and transmission-phase control. Compared with direct optimization of transmission or reflection coefficients, this approach offers a physically transparent route to rationally controlling resonance properties, including linewidth and group delay, while reducing computational cost. The resulting designs demonstrate how near-unity transmission and $2π$ phase excursion can arise from resonance interference. These signatures of the Kerker effect extend beyond the conventional alignment of electric and magnetic dipole resonances to other mechanisms, including guided-mode resonances and quasi-bound states in the continuum. We further optimize the Kerker condition across a continuum of refractive-index states in a reconfigurable phase change material metasurface, enabling high-transmission $2π$ phase tuning. Our results establish a versatile and practical approach to metasurface phase control through the targeted design of scattering poles and zeros.

physics.optics↗

arXiv2026

Taming Softmax at the Limits of Muon: A Unified Approach through Hilbert Geometry

Muon controls weight updates through the RMS movement of a linear layer's output. This implicitly treats the linear output as a proxy for the full module, including the nonlinearity that follows. This proxy can be misleading for softmax, where the RMS shift in activations does not faithfully measure the change in the output distribution. We propose Hilbert distance as a natural metric for bounding changes in these output distributions. Hilbert distance bounds changes in cross-entropy loss and limits probability redistribution. For softmax, it takes the form of the variation seminorm of the preactivation perturbation, allowing probability constraints to be expressed as parameter constraints. Formulating the steepest descent problem with a Hilbert distance constraint brings two recently proposed methods under a unified mathematical framework. For the LM head, solving a row-wise approximation to the resulting diameter-constrained problem leads us to RowNorm \citep{pethick2025trainingdeeplearningmodels}. For scaled dot-product attention, the same principle yields the first-order query--key steepest descent problem underlying Compositional Muon \citep{keigwin2026compositional}. These connections motivate using Hilbert distance as a guiding principle for optimizer design in softmax modules.

cs.LG↗

arXiv2026

Roth's theorem on weighted projective lines

Let ${\mathbb P}^1_{\mathfrak q}$ be the weighted projective line with coprime weights ${\mathfrak q}=(q_0,q_1)$ and $L=q_0q_1$. The weighted Veronese map $Φ\colon {\mathbb P}^1_{\mathfrak q}\to {\mathbb P}^1$ satisfies $h\circΦ=L\,{\mathfrak s}$. Pulling back Roth's theorem gives the optimal exponent $2L$ for the full rational orbit set. For points represented by coprime integral coordinates, counting suggests the smaller exponent $q_0+q_1$. We formulate this as a conjecture and prove an $S$-adic radical form under the $abc$ conjecture. For ${\mathfrak q}=(2,3)$ the resulting inequality at the target $[1:1]$ is the coprime Hall-type inequality, and Danilov--Elkies examples show that the exponent $5$ is sharp. We also prove the expected exponent unconditionally over characteristic-zero function fields for constant targets and polynomial approximants, and for almost every real target. A weighted $abc$ formulation is introduced; in signature $(3,2)$ finitely many coefficient pairs are equivalent to $abc$, and weighted Belyi maps transfer this input to arbitrary algebraic targets. Applications include a pullback form of Schmidt's subspace theorem, weighted Thue inequalities, and a counterexample to a naive weighted rescaling of Roth's index construction.

math.NT↗

arXiv2026

Revisiting the 16x Pretraining Compute Gap between Diffusion and Autoregressive Language Models

Diffusion language models (DLMs) have attracted wide attention for their fast, parallel decoding, but they are believed to be far less efficient to train than autoregressive language models (ARMs): a widely cited study reports that a DLM needs roughly 16x the training compute of an ARM to match its likelihood, a figure that has made pretraining DLMs from scratch look prohibitive at scale. We find that the methodology behind this figure has fallen out of step with current DLM practice: it scores the model with its evidence lower bound (ELBO) and trains a full-sequence, decoder-only model. We therefore revisit the scaling laws of DLMs, adopting the exact likelihood under the model's own decoding order, block diffusion training, an encoder-decoder architecture, and a sweep over the block sizes used in practice. Under this new protocol, we study across ten model variants, about 1,700 models in total, and find that pretraining DLMs is far cheaper than previously thought. Concretely, scoring with the exact likelihood reduces the compute multiple to 12.8x; block diffusion training and an encoder-decoder architecture further reduce it to 7.8x at block size 32, and smaller block sizes narrow it to 4.6x at block size 8 and 3.0x at block size 4. We complement these scaling analyses with an empirical rule for the optimal decoder depth of encoder-decoder block diffusion, a held-out compute budget beyond the fitting range at which the gap between variants persists, and downstream benchmark results where likelihood improvements carry over to accuracy. Overall, the 16x figure substantially overstates the pretraining cost gap between DLMs and ARMs, and pretraining DLMs from scratch is far more viable than previously assumed.

cs.LG↗

arXiv2026

Physical-Field-Bridged Material Parameter Inference from Multi-View Deformation Videos

Estimating simulation-ready material parameters from visual observations is crucial for deformable-object manipulation but challenging because material properties are only indirectly revealed through physical responses. We introduce physical-field-bridged material inference, a single-pass framework that uses dense physical responses to connect multi-view RGB deformation videos with constitutive-type classification and continuous parameter regression. Unlike inverse-physics methods requiring per-scene reconstruction, online tuning, or simulation-in-the-loop optimization, our model predicts force/contact masks, projected motion-flow fields, and projected stress-response maps, then conditions material inference on these fields to reduce reliance on appearance or category shortcuts. We construct DeformFieldBench, a 3DGS MPM learning environment comprising 5,000 simulations across 100 object shapes and five mechanically informative loading modes, with dense pixel-aligned physical supervision. Our method achieves the best median errors on all four continuous parameters, improves stress-hotspot recall from 0.0762 to 0.5121, and attains 0.8676 SSIM/36.82 dB PSNR in same-simulator replay on unseen objects. Under a protocol-matched GSO evaluation, inference and future-state prediction take approximately 30 seconds, compared with 15-120 minutes for per-scene optimization. The method also yields measurable replay agreement on five real-world sequences. These results support dense physical fields as an effective bridge for fast material inference, while cross-simulator and real-world tests reveal remaining domain and identifiability gaps.

cs.CV↗

arXiv2026

Training Generalist Navigation Policies via Action Self-Distillation

Language provides a natural interface for humans to specify and reason about goals, but training language-conditioned robotic policies and reasoning typically requires expensive manual annotations. We propose Action Self-Distillation (ASD), a method for learning language-conditioned reasoning for robotic navigation without human-annotated language or reasoning labels. ASD uses an open-source VLM, which contains broad world knowledge and reasoning capabilities, as both a teacher network and a student policy. During training, the self-teacher receives privileged trajectory feedback and infers the underlying navigation goal together with goal-oriented reasoning. The student is then trained to condition on the inferred navigation goal, predict future ego-frame trajectory, and provide a description of the scene semantics, plan, and reasoning, from onboard observations alone. Across 13 challenging indoor and outdoor real-world navigation tasks, our method improves success rate over the strongest baseline by $30\%$ in indoor settings, $30\%$ on multi-stage tasks, and achieves $95\%$ success rate on outdoor tasks, validating that privileged trajectory feedback can enable self-distillation for scalable generalist navigation policies without human annotations.

cs.RO↗

arXiv2026

Effects of Coulomb Collisions on the Structures of Fast Magnetosonic Shocks

We present fully kinetic particle-in-cell simulations to demonstrate qualitative effects of binary Coulomb collisions on fast magnetosonic shocks. We find that with sufficiently high collisional frequencies, shock rippling and reformation can be inhibited, leading to stationary, laminar shock fronts. We observe that ion-ion collisions disrupt the motion of particles reflected back upstream that is the key for cyclical reformation. Meanwhile, the electron-ion collisions provide a Spitzer resistivity that can dissipate energy from the instability that causes shock rippling. This result helps us understand the role of collisions in laboratory shocks and possibly the outer heliosphere.

physics.plasm-ph↗

arXiv2026

Gas radiation and the gravity limit of flicker in a co-flowing hydrogen jet diffusion flame

Although weak at normal gravity, gas radiation sets the gravity at which a co-flowing hydrogen jet diffusion flame stops flickering. Flamelet-based simulations with unity Lewis numbers, which reproduce the flicker frequency measured by femtosecond coherent anti-Stokes Raman scattering within 4%, are combined with local stability analysis. At one tenth of normal gravity, the adiabatic flame flickers, the radiating flame does not, and switching the emission off restores flicker. Radiation acts through two mechanisms that do not weaken at reduced gravity. The hot outer layer loses buoyancy in proportion to the gas residence time, and a disturbance that displaces the flame also displaces its emission, which damps the disturbance on the lean flank (displaced-emission damping). Freezing the radiative source, which removes only its response to disturbances, slows the downstream recession of the oscillation more than tenfold. Both mechanisms slow the upstream spreading of disturbances. This spreading, competing with the co-flow, sets the onset location and the critical co-flow. Without radiation, the spreading speed scales with the square root of gravity; its radiative reduction does not. At reduced gravity, the flicker limit therefore becomes a straight line relating the square root of the co-flow Froude number to a radiation number, the ratio of buoyant to radiative time scales. The calibrated line places the critical gravity for the experimental co-flow at 0.13 times normal gravity, 2.6 times the adiabatic value. It correctly predicted whether two further flames keep flickering. Beyond the limit, transient wave packets can mimic flicker throughout a short drop-tower test.

physics.flu-dyn↗

arXiv2026

Tanawpamana: Benchmarking Vision-Language Models on Philippine Cultural Heritage Sites

Vision-language models are increasingly used to identify and describe places, yet their recognition and knowledge of specific cultures remain poorly understood, particularly in underrepresented regions such as the Philippines. This study introduces the Tanawpamana Dataset, a repository of Philippine heritage site images with metadata based on the Talapamana registry of the National Commission for Culture and the Arts of the Philippines. We also introduce the Tanawpamana Benchmark, comprising 4,046 tasks designed to evaluate models' recognition and knowledge of heritage sites. Evaluating seven vision-language models, we find that they struggle more with fine-grained visual discrimination among similar heritage sites than with coarse recognition, and that their geographic knowledge is weaker when inferred from image inputs than from text.

cs.CV↗

arXiv2026

TDR: Tagalog Diacritic Restoration Dataset

Homographs are words that share identical spelling but differ in meaning and pronunciation. In Tagalog, these distinctions are disambiguated in writing through tuldik (diacritical marks), which are rarely preserved in contemporary Tagalog text. This scarcity of diacritics limits diacritic restoration, which is useful for downstream tasks such as prosody selection for text-to-speech synthesis. To address this gap, this study introduces TDR, an annotated corpus of 40 frequent Tagalog homographs constructed from news text. It also introduces a collocation-driven retrieval method to expand the dataset and mitigate severe class imbalance among minor variants. Finally, we show that TDR supplies a capability absent from pretrained models, and that this capability can be learned by small, inexpensive models as well as by fine-tuned LLMs.

cs.CL↗

arXiv2026

On extending induced subgraphs

When characterizing graphs defined by forbidden induced subgraphs, it is often desirable to understand how a graph $G$ can be generated incrementally from a given induced subgraph $H$. In this paper, we study a number of problems within this framework. First, for several notions of connectivity, we determine all minimal induced subgraphs of $G$ that contain $H$ and attain the connectivity of $G$. Second, we establish how to grow $H$ inside $G$ while preserving twin-freeness and certain forms of connectivity. Third, we establish {\it finite test theorems} for various graph classes, proving that agreement with the recursively generated graphs for a few initial iterations forces agreement at all subsequent iterations. Consequently, such an infinite graph family can be completely determined by finitely many of its initial generations. As an application, we characterize $\{\mathbb L_1,\mathbb L_3\}$-free graphs, where $\mathbb L_1$ is the claw and $\mathbb L_3$ is another forbidden graph for line graphs. Our structural results reduce this problem to a finite verification, which is completed through computer-assisted computation.

math.CO↗

arXiv2026

Fast charging of quantum batteries assisted by nonlinear couplings

Charger-mediated quantum batteries offer promising opportunities for efficient energy transfer and storage at the quantum level, with potential advantages beyond their classical counterparts. Here, we theoretically demonstrate that nonlinear charger-battery interactions can substantially accelerate the charging of quantum batteries. In the weak-driving regime, sufficiently strong nonlinear coupling can dominate over the effect of dephasing and significantly reduce the charging time. In the strong- and optimal-driving regimes, where the optimal driving strength maximizes the steady-state ergotropy in the linear-coupling regime, nonlinear coupling gives rise to a new optimum in the charging time beyond that of the linear-coupling regime, accompanied by nearly critically damped dynamics. Remarkably, even in the absence of external driving, nonlinear coupling alone can substantially reduce the charging time, providing an alternative route toward fast battery charging. Our results establish nonlinear charger-battery coupling as an effective resource for accelerating quantum-battery charging and broaden the scope of interaction-enabled quantum advantages in energy transfer and storage.

quant-ph↗

arXiv2026

Strong convergence rate of the implicit midpoint method for stochastic KdV equations

We establish a strong convergence rate for the implicit midpoint method applied to the periodic stochastic Korteweg-de Vries equation with additive noise, addressing a question that has remained open since the work of Debussche and Printems. To control nonlinear error growth, we obtain exponential integrability estimates for the exact and comparison solutions using the coercivity of a classical KdV invariant. We then derive local error bounds from estimates for the comparison solutions and control error accumulation through an $L^2$ error propagation identity that isolates a martingale increment. Under $H_0^{9/2}$ regularity assumptions on the noise and the deterministic initial data, we prove that, on every fixed finite time interval, the maximum $L^2(\mathbb{T})$ error at the time points is $O(h)$ in $L^p(Ω)$ for every $1\le p<\infty$. Notably, the exponential integrability and the convergence rate hold for every damping coefficient $λ\ge0$, including the undamped case $λ=0$. Numerical experiments confirm the predicted first-order convergence.

math.NA↗

arXiv2026

Existence of positive solutions to anisotropic elliptic equations in the entire Euclidean space

In this paper, we research an anisotropic elliptic equation in the entire Euclidean space R^{N} . Under the assumption that the nonlinearity of the equation has exponential or critical growth, we prove, by means of the mountain-pass theorem, that the equation admits a nontrivial positive weak solution. Moreover, applying the mountain-pass theorem and Ekeland's variational principle, we show that a perturbation of the equation possesses two distinct positive weak solutions.

math.AP↗

arXiv2026

Clean handshake sequences

After a basketball game, each of the $n$ players on Team A must shake hands with each of the $n$ players on Team B. What is the minimum number of (reversible, ambidextrous) gloves required so that zero germ transmission occurs? We show that the answer is $3n/2+o(n)$ and develop a general theory for prescribed handshake sets.

math.CO↗