Search arXiv⌕ Search

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 991 records · Page 55Linked to original sources
arXiv2026

Thermodynamic-limit bootstrap: Solving fractional quantum Hall ground states

Strongly interacting ground states are usually computed in finite systems and then extrapolated to the thermodynamic limit (TDL). Here we develop a quantum many-body bootstrap that solves fractional quantum Hall ground states directly in the TDL. The method gives rigorous and quantitatively tight lower bounds on the ground-state energy and identifies the ground-state structure and topological order. For Laughlin and Pfaffian parent Hamiltonians, it reproduces exact pair-correlation functions and low-lying orbital entanglement spectrum (OES) nearly level by level. For Coulomb interactions, the rigorous TDL energy bounds closely match thermodynamic estimates from exact diagonalization and density-matrix renormalization group (DMRG) extrapolations, and the local diagnostics recover established Laughlin and Moore--Read OES counting and Friedel oscillations of the composite Fermi liquid. Beyond these benchmarks, the bootstrap resolves the thermodynamic ground states at $ν=1/4$ and $ν=12/5$, identifying a composite Fermi liquid and $\overline{\mathrm{RR}}_3$ topological order, respectively. The same framework applies equally well to Abelian and non-Abelian, gapped and gapless quantum Hall states. The TDL bootstrap therefore provides a general ground-state solver that determines both the energy and the state directly in the TDL.

cond-mat.str-el↗
arXiv2026

Loud and Clear: Making Visual Evidence Count in Multimodal Document RAG

Multimodal document retrieval-augmented generation (RAG) has focused largely on retrieval, yet its answers depend on how well a vision-language model reads the retrieved pages. We find that evidence in retrieved pages remains underused even when the correct pages are retrieved. On SciEGQA oracle pages, five prior methods give the generator 0.7K to 2.7K visual tokens per query on average under a common image processor, yet raising the page-only budget from 1K to 32K improves accuracy by 15.22 points. At a fixed 16K budget, adding ground-truth crops improves accuracy by 3.75 points over page-only input. We study what the generator needs from retrieved pages through two requirements: Loud, making full use of visual token capacity, and Clear, refining the page so that its evidence becomes prominent. For Loud, increasing either the visual token budget or visual fidelity, the image resolution those tokens preserve, improves accuracy with the other held fixed, and full-page scaling outperforms repetition and tiling at matched budgets. For Clear, suppressing surrounding content improves accuracy without enlarging selected regions, and adding OCR of these regions to page images outperforms either input alone. We implement both as Loud and Clear (LoC), which combines higher-resolution pages with OCR from query-relevant regions without ground-truth annotations or additional training. LoC outperforms eight prior methods on SciEGQA, ViDoRe-v3, and MMLongBench-Doc, improving over the strongest, ColQwen3, by 8.88, 3.09, and 4.22 points with the same retriever and generator. It also outperforms its page-only variant under the same visual token budget. These results locate a bottleneck beyond retrieval: how the retrieved pages are given to the generator.

cs.CV↗
arXiv2026

Constrained Multi-Agent Reinforcement Learning in Imagination

Predicting cost in a multi-agent world model moves most constrained policy updates into imagination: the stronger model-free baseline needs $10$ to $20$ times CA-MAMBA's interactions on the median SMAC configuration, and CA-MAMBA earns $2.8$ to $12$ times the baselines' MAMuJoCo reward at equal budget. In controlled tests on one task per benchmark, pricing the model's cost advantages at parity more than halves a cost-blind learner's cost and meets a tight continuous-control limit the learned multiplier misses, suggesting the model's cost estimates are not the bottleneck.

cs.AI↗
arXiv2026

ZoomCap: Evidence-Routed Visual Zooming for Recoverable Visual Information Transfer

Detailed captions transfer visual information into vision-language model supervision, but quality is easily conflated with length: long single-pass outputs still miss small objects, text, and spatial relations or promote uncertain observations into confident claims. We formulate fine-grained captioning as recoverable visual information transfer, in which a teacher acquires missing evidence under a finite observation budget and converts it into reusable supervision. ZoomCap uses a draft caption to route conditional knowledge retrieval, OCR-sensitive reading, coverage-gap recovery, and adaptive visual zooming, allocating deeper crops only to the most complex eligible region, deriving spatial relations from grounded geometry, and fusing non-duplicate evidence. We also introduce FactCycle, which scores caption-derived atomic claims on two blind anchors--source image (truthfulness) and reconstruction (realizability)--separating fabrication from generator loss claim by claim; human and multi-judge comparisons complement this fact-level view. Against Qwen3.5-397B-A17B captions, ZoomCap achieves an 81.9% human tie-adjusted preference score with a teacher over 10x smaller, and FactCycle counts 52.4 transferred claims per image vs. 31.5 for ScaleCap. The evidence also transfers: in continual pre-training, ZoomCap-450K improves the 11-benchmark average by 1.31-2.38 points over the strongest competing corpus and Ling-Tiny-VL-v3 gains 1.4 points on a production document/OCR monitor suite (1.5 on perception) at 60B tokens; in medical pre-training, 2.64M ZoomCap-captioned images improve five VQA benchmarks by 3.04 points (vs. 1.76 for same-image Gemini-3.1-Pro captions).

cs.CV↗
arXiv2026

CoViG: Coarsened Vision GNN for Image Classification

Although Graph Neural Networks are widely used in various domains, they remained mostly unexplored in computer vision. Convolutional Neural Networks and Vision Transformers were leading the field, until the quite recent introduction of Vision Graphs (ViG) that treat images as a set of nodes and apply graph convolution on their k-NN graphs. ViGs show impressive accuracy on image classification but their computational cost remains relatively high. Inspired by optimization methods for GNNs, we present CoViG, a novel approach combining graph coarsening and ViGs. Our method introduces an additional MinCUT pooling step within the Grapher component of intermediate ViG blocks. By simplifying the intermediate graphs and thus decreasing the number of nodes of specific blocks, our approach considerably reduces the computational cost of ViG during inference. We evaluated our approach on ImageNet with several ViG architectures, and we obtained accuracies equivalent to the original ViG models with only about 51\% of FLOPs.

cs.CV↗
arXiv2026

Time Delay of KPP Fronts Across a Semipermeable Barrier

We study a KPP reaction--diffusion equation on the real line with a semipermeable barrier at the origin, where the flux is continuous and the density jump is proportional to the spatial derivative. With diffusion and linear growth normalized to one, let $β$ denote the barrier resistance. For initial data equal to one sufficiently far to the left and zero near the barrier and on the receiving side, we prove convergence on the receiving side to a minimal-speed wave with a unique phase in the Bramson frame. The difference between distant first-arrival times with and without the barrier therefore has a finite limit $τ_β$, independent of the arrival level and continuous in $β$. For concave KPP reactions, this delay is strictly increasing in $β$ and satisfies $τ_β=Kβ+o(β)$ as $β\downarrow0$, with $K>1/4$. Without the concavity assumption, the strong-barrier expansion is $τ_β=\frac12\logβ-\frac12\log\logβ+C+o(1)$, where $C$ is determined by the homogeneous reference amplitude and the closed donor-side Neumann evolution. The analysis relates the sustained interface input to the front phase through leading-edge estimates, interior matching, and parameter-uniform moment bounds.

math.AP↗
arXiv2026

AdaLoop: Thinking with the Vision Encoder

Most vision-language models (VLMs) treat the vision encoder as a static feature extractor: it encodes an image once, independently of the question, and leaves subsequent reasoning to the language model. Yet different questions may require different visual operations and processing depths. We present AdaLoop, a framework for question-conditioned visual recurrence in VLMs. AdaLoop adapts visual computation to the question and reuses the encoder's transformations across multiple iterations, with the pretrained vision weights frozen. We also introduce VisRO, a procedural dataset and generation pipeline for training and evaluating vision-centric reasoning. On six VisRO tasks, AdaLoop outperforms standard fine-tuning and question-conditioned single-pass baselines with matched training data and optimizer updates. These gains hold across three VLM families on both in-distribution and out-of-distribution examples. Our findings suggest that pretrained vision encoders can contribute to reasoning through task-directed, iterative visual computation.

cs.CV↗
arXiv2026

Multilayer Evaluation of Vertiport Departure Directions Under Urban Obstacles and Wind

Vertiports must accommodate electric vertical takeoff and landing (eVTOL) aircraft with different climb capabilities and crosswind tolerances. However, conventional runway based departure criteria do not exploit the directional freedom of vertical takeoff. This paper proposes a framework that evaluates every departure azimuth and identifies usable and preferred departure directions for each climb capability. Each azimuth is evaluated by its obstacle clearance from surrounding urban buildings and its wind operability based on ten years of wind observations. Three climb angle layers reflect differences in climb capability among aircraft. Four weighting scenarios between obstacles and wind reveal how operational priority affects the preferred departure direction. In an application to Suseo station and the Jamsil heliport in Seoul, steeper climbs widened the usable azimuths from 8.9 to 34.4% and from 66.1 to 83.6%, respectively. By relaxing obstacle constraints, steeper climbs also shifted the preferred direction toward the prevailing wind at lower wind weights. These results show that vertiport departure paths should be planned as a set of candidate directions matched to the obstacles and wind of the site and the climb capability of the aircraft, rather than as a single runway style path.

eess.SY↗
arXiv2026

CertEx: Exact Visual Retrieval at Scale by Certified Exclusion

Exact $k$-nearest-neighbor (kNN) search over image descriptors and embeddings underlies visual retrieval, near-duplicate detection, the evaluation of learned representations, and increasingly the memory of LLM agents. However, exhaustive search computes every distance at full precision, whereas approximate indexes return recall below one. We observe that exact search only needs a proof that each excluded vector cannot be a neighbor. Because such a proof only has to be conservative, it can be checked in low precision, much as the retina examines the whole scene at low resolution. We call this principle certified exclusion and realize it in an exact kNN method for GPUs, which we call \certex. Specifically, \certex\ stores a lower-bounding projection of each vector in half precision and folds a rounding certificate into the same tensor-core dot product, so that only the few vectors that the test keeps need exact distances. We prove that \certex\ returns the exact kNN of every query. On sixteen datasets with up to 100M vectors, \certex\ is exact for every query at 7.1$\times$ to 18.0$\times$ the throughput of the fastest cuVS search on one H100 and outperforms quantized and graph indexes whenever every query must be exact.

cs.CV↗
arXiv2026

Critical Ising Model on the Periodic Cubic Lattice $(\mathbb{Z}/L\mathbb{Z})^{3}$

We investigate the critical behaviour of the nearest-neighbour Ising model on the three-dimensional periodic cubic lattice $(\mathbb{Z}/L\mathbb{Z})^{3}$, and we present a companion set of Monte Carlo simulations. After recalling the rigorous results on the continuity of the spontaneous magnetization in $d=3$ and the triviality of the scaling limit in $d\ge 4$, we compile modern high-precision estimates of the 3D Ising critical exponents and verify the scaling relations that connect them. We then describe numerical measurements of the Binder cumulant, the magnetic susceptibility, the specific heat, the two-point correlation function, and the finite-size scaling of the magnetization. We discuss the systematic uncertainties that affect the extraction of the exponent ratios $γ/ν$ and $α/ν$, and we emphasize that an exponential fit of the correlation function is not appropriate at the critical point, where the decay is algebraic.

cond-mat.stat-mech↗
arXiv2026

DocPilot: Interactive Agent Orchestration and Training for Long Visual Document Understanding

Multimodal long-document question answering requires models to seek sparse evidence across pages and modalities, understand heterogeneous document content, and reason over the collected evidence. Even advanced vision-language models (VLMs) can overlook sparse evidence and struggle with cross-modal reasoning in a single pass. Existing document agents further introduce multi-step interaction, but some still rely on predefined workflows or coarse-grained actions, limiting the flexibility and granularity of document reasoning. We introduce DocPilot, an agentic document framework that supports expressive orchestration and makes complex document solving learnable by small language models (SLMs). DocPilot orchestrates evidence seeking through page-level navigation and fine-grained evidence localization, and integrates local visual inspection, code-based numerical reasoning, and cross-modal reasoning for multimodal understanding. To transfer these capabilities to small models, we propose curriculum-based agentic training that progresses from single-ability learning to adaptive orchestration through supervision from data-grounded reasoning traces and executable interaction trajectories. Across MMLongBench-Doc, DocBench, FinRAGBench-V, and MADQA, DocPilot achieves strong gains with frontier VLM backbones, while DocPilot-4B delivers competitive performance against substantially larger models, demonstrating strong effectiveness and generalization across diverse long-document reasoning settings. The code is available at https://github.com/HKUSTDial/DocPilot.

cs.AI↗
arXiv2026

Directional delays and singular strong-barrier limits for Fisher--KPP propagation across a semipermeable interface

We study Fisher--KPP propagation from compactly supported data across a planar semipermeable interface in $N\ge2$ dimensions, with unit diffusion and $f'(0)=1$. For every finite resistance, the receiving front has speed $2$ and logarithmic correction $-(N+2)\log t/2$. Its phase is determined by a positive directional amplitude on the closed receiving hemisphere. This amplitude is jointly continuous in resistance and direction and is Hölder continuous in direction with exponent $1/3$. It also determines a threshold-independent signed arrival-time difference. Under concavity, the amplitude decreases strictly with resistance. As the resistance $β$ tends to infinity, the amplitude is asymptotic to $2(4π)^{-N/2}I_N(θ)(\logβ)/β$ uniformly on each fixed receiving cone bounded away from the equator. Here $I_N$ is a weighted input moment of the closed Neumann donor problem. Although this moment diverges logarithmically near the equator, the equatorial amplitude tends to half the amplitude of the even Neumann donor solution. The strong-resistance and equatorial limits therefore do not commute. The proof combines critical derivative moments, an integrated Robin trace estimate, Cartesian leading-edge limits, and matching through the wave displacement. The phase and asymptotic formulas require only the smooth KPP assumptions, without the additional restriction $f'\le f'(0)$. Concavity is used for strict resistance ordering and the strict bounds on the limiting equatorial delay.

math.AP↗
arXiv2026

Hybrid Generative-Tagging Decoding for Efficient Product Attribute-Value Extraction

Attribute-value extraction is important for building high-quality product catalogs in e-commerce. Existing systems either rely on generative or tagging strategies for extracting the attribute names and their corresponding values. Tagging models directly label spans in the source, making them fast and grounded, but they cannot produce values that require reformatting or generate rare attribute names outside their label space. Generative models avoid these drawbacks by generating both attribute names and values token by token, but this auto-regressive nature makes them slow and limits attribute coverage. We introduce hybrid generative-tagging decoding, a novel model architecture that combines tagging source spans and open-ended generation. For frequent attributes whose values are copyable, a tagging head extracts values directly from the source. This lets the generative head focus on the remainder - rare attributes outside the tagger's label space and non-copyable values, including implicit or normalized values. Integrated into vLLM inference, the hybrid decoder raises attribute coverage by 10% and extracts 63% more correct attribute-value pairs at no precision cost, while reducing latency by 44%.

cs.CL↗
arXiv2026

VoteGEO-Bench: Measuring the Impact of Generative Engine Optimization on Electoral Visibility

Large language models (LLMs) are increasingly used for election information, yet little is known about how external web content shapes the political information they present. Generative engine optimization (GEO) rewrites web content to improve its visibility in generated responses. To investigate its potential risks in voting scenarios (VoteGEO), we introduce VoteGEO-Bench, a benchmark covering the 2024 U.S. presidential election and the 2026 Japanese House of Representatives election, combining election-related queries, historical webpages, and voter personas. We first analyze sources cited by generative engines and find that many citations come from lower-cost owned and open-platform sources, while some webpages remain visible across days. We then evaluate 11 rewriting strategies in 4 categories, examining target specification, effectiveness across engines and elections, and the role of voter personas. Rewriting without specifying a target rarely increases target exposure, whereas target-specified rewriting more often does. Strategies that improve page organization or strengthen a position through argumentation yield greater gains than adding supporting evidence or changing wording. For the two strongest strategies, gains remain broadly positive across evaluated U.S. personas, but persona conditioning shows no consistent effect on these gains. Explicit partisan identities can create baseline exposure gaps larger than the gain from rewriting one webpage. When partisanship is implicit in the voter profile, baseline shifts depend on how strongly the engine responds to that signal.

cs.IR↗
arXiv2026

Unveiling star formation at sub-solar metallicity with ALMA2040

A fundamental question in astrophysics is whether the processes governing star formation are universal or depend on the physical conditions of their local galactic environment. The Milky Way and the Magellanic clouds offer ideal laboratories to explore this question, as the gas conditions in them span up to orders of magnitude in metallicity, gas surface density, ambient pressure, dust abundance, radiation field, and cosmic-ray flux. In particular, both the outer Milky Way and the Magellanic clouds provide the closest low-metallicity environments with conditions similar to those of metal-poor galaxies, while remaining sufficiently close to soon allow spatial resolution of individual star-forming cores and disks. Recent observations with ALMA and other facilities have begun to unveil the physical and chemical properties of star-forming regions in low-metallicity environments. Yet, their distances from the Sun still require transformational gains in current ALMA capabilities to address fundamental open questions: (1) Are the properties of prestellar cores universal? (2) Does cloud fragmentation, and the resulting stellar IMF, depend critically on metallicity? (3) How does dust evolution depend on metallicity, and what is the impact in the survival of protoplanetary disks? (4) Is the reduced amount of metals and dust shielding at low metallicity critical for the formation and survival of molecules? ALMA2040 will identify similarities and differences between solar- and sub-solar-metallicity star-forming regions, and answer the question: how do different environmental conditions influence the physical and chemical pathways leading to star formation?

astro-ph.IM↗
arXiv2026

Semantic Latent Alignment for Flow Matching with Invertible Bi-Lipschitz Adapters

Pretrained latent spaces trade compression, reconstruction, and the learnability of the coordinates a downstream generator has to work in, which complicates adapting these spaces to different domains or tasks: continued training is expensive, and without a new alignment objective it often fails to improve downstream generation. We instead leave the encoder and decoder fixed and change only the coordinates between them, using an invertible bi-Lipschitz adapter aligned offline to a semantic teacher and frozen during class-conditional flow matching. Invertibility preserves compatibility with the pretrained autoencoder, while the bi-Lipschitz constraint limits how far alignment can distort its latent geometry. This provides a simple way to improve semantic organization while retaining the compression and reconstruction properties of the original latent space. On two medical imaging datasets, constrained latent adaptation improves feature metrics, seed stability and counterfactual edit strength, while remaining competitive on downstream classification. In contrast, unconstrained adaptation can become highly anisotropic and sensitive despite remaining invertible, highlighting the importance of controlling geometric distortion. Our results suggest that carefully constrained post-hoc adaptation offers a parameter-efficient and tuning-free alternative to jointly modifying the representation during generative training.

cs.CV↗
arXiv2026

QuasiLeak-Bench: Evaluating LLM-Based Privacy Detectors on Policy-Conditioned Quasi-Evidence Group Recovery

Natural language messages often contain quasi-identifiers that are weak in isolation but can jointly reveal an individual's private attributes. We call such a jointly revealing set of cues a quasi-evidence group. Existing privacy benchmarks focus on explicit identifiers, attribute inference, or task-driven disclosures, leaving inference from quasi-evidence groups largely unmeasured. In practice, whether exposing such a quasi-evidence group violates privacy also depends on sender-specified disclosure policies. We introduce QuasiLeak-Bench, a benchmark for policy-conditioned quasi-evidence group recovery, which evaluates whether detectors recover complete quasi-evidence groups and flag the disclosures the given policy prohibits. To obtain reliable ground truth at scale, a reverse-construction pipeline first selects weak quasi-identifiers, composes them into quasi-evidence groups verified to jointly reveal a target attribute, and generates documents around this evidence, followed by automated quality control. Experiments show that existing privacy detectors transfer poorly to QuasiLeak-Bench, and locally deployable LLM-based detectors fall far behind proprietary models, mainly because they miss most quasi-evidence groups altogether. However, a 4B model fine-tuned on our training split surpasses the strongest prompted model on held-out documents, showing that the task is learnable.

cs.CL↗
arXiv2026

Monte Carlo Framework for Physics-Based Design and Optimization of X-Ray Compressive Ghost Imaging Systems

X-ray ghost imaging has emerged as a promising computational imaging modality capable of forming images from spatially integrated measurements. Unlike optical implementations, image formation is governed by multiple energy-dependent photon-matter interactions: photoelectric absorption, Compton scattering, and fluorescence. The relative contributions of these interactions depend on the source spectrum, sample geometry and composition, and detector configuration. We present a FLUKA-based Monte Carlo framework that models photon transport and predicts bucket signals generated by binary illumination patterns. The framework couples a physical scaling analysis, which converts the simulated bucket signals into photon numbers, tube loading and absorbed dose, with a ghost imaging reconstruction analysis of image quality and convergence. Our framework helps assess whether a given configuration is limited by the source output, the mask modulation or the reconstruction before any measurement is made. We evaluate the reconstruction for a centimeter-scale imaging system for an iron crown using one transmission detector and three off-axis scattering detectors over a range of source spectra. The results reveal how source characteristics and detector geometry influence reconstruction quality and radiation dose, providing quantitative guidelines for future X-ray ghost imaging system design. More broadly, this work establishes a predictive design framework for application-specific X-ray ghost imaging systems.

physics.optics↗