Search arXiv⌕ Search

SEARCH · Search arXiv

Search Search arXiv

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

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,747 records · Page 97Linked to original sources
arXiv2026

Geometric characterization of non-Gaussian entanglement for finite stellar rank states

We introduce a general framework for the analysis of non-Gaussian entanglement in bosonic states of finite stellar rank. The central result is the full characterization of their entanglement structure through the atomic decomposition of their stellar polynomial and its associated structural graph, whose connected components determine the mode-intrinsic entanglement content of the state and all partitions compatible with passive separability. An essential ingredient in this construction is the concept of essential variables, which identify the minimal number of effective modes involved in a core state, in direct correspondence with the symplectic rank. This reduction provides the foundation for decomposing stellar polynomials into atomic factors and for revealing the underlying entanglement structure. Building on this, we derive complete separability criteria for two-mode states, expressed through hyperplane decompositions of zero sets, and for stellar-rank-2 states across an arbitrary number of modes. Applications to several example states illustrate how the method isolates genuinely non-Gaussian resources and quantifies preparation complexity.

quant-ph↗
arXiv2026

Four-dimensional supersymmetric massive QED

We study a four-dimensional $\mathcal N=1$ supersymmetric Proca model with an explicit mass term for a real vector superfield and a generalized transverse kinetic operator. Although it displays no gauge redundancy in Proca variables, for local kinetic operators the model is the exact classical unitary-gauge form of the minimal Abelian $\mathcal N=1$ Stueckelberg realization, with the SQED-like exponential as its minimal gauge-covariant matter coupling. We compute the Kählerian contribution to the one-loop effective action using both the background-field Hessian and an independent supergraph organization. In the minimal model the background-dependent divergence is local and is absorbed by the matter Kähler normalization, and the finite local terms left by the subtraction remain scheme dependent. At $α=0$, an anomaly-free minimal Stueckelberg formulation supports a conditional perturbative-renormalizability interpretation for the corresponding minimal theory. For $f(\Box)=\Box+α\Box^2$ we distinguish the exact polynomial model, defined over the full momentum range, from a truncated derivative expansion; in the latter reading, the additional pole and the local matching term are ultraviolet information determined by the completion rather than predictions of the truncated theory.

hep-th↗
arXiv2026

Deep Learning-Enhanced Real-Time Wi-Fi Sensing Through Single Transceiver Pair

The advancement of next-generation Wi-Fi technology heavily relies on sensing capabilities, which play a pivotal role in enabling sophisticated applications. In response to the growing demand for large-scale deployments, contemporary Wi-Fi sensing systems strive to achieve high-precision perception while maintaining minimal bandwidth consumption and antenna count requirements. Remarkably, various deep learning-driven perception technologies have demonstrated the ability to surpass conventional resolution limits. However, the theoretical underpinnings of this phenomenon have not been thoroughly investigated in existing research. We find that under hardware-constrained conditions, the performance gains of deep learning in Wi-Fi sensing primarily originate from two aspects: prior information and temporal correlation, which act as specific forms of side information that reduce the estimation error bound. We construct a deep learning-based Wi-Fi sensing system using only a single transceiver pair and design experiments to validate these gains. The system achieves an average human pose estimation error of 0.2189 m and an average localization error of 0.6124 m, while operating in real time at 42 fps on commodity hardware.

eess.SP↗
arXiv2026

Fair Division with Indivisible Goods, Chores, and Cake

We study the problem of fairly allocating indivisible items and a desirable heterogeneous divisible good (i.e., cake) to agents with additive utilities. In our paper, each indivisible item can be a good that yields non-negative utilities to some agents and a chore that yields negative utilities to the other agents. Given a fixed set of divisible and indivisible resources, we investigate almost envy-free allocations, captured by the natural fairness concept of envy-freeness for mixed resources (EFM). It requires that an agent $i$ not envy another agent $j$ if agent $j$'s bundle contains any piece of cake yielding positive utility to agent $i$ (i.e., envy-freeness), and that agent $i$ be envy-free up to one item (EF1) towards agent $j$ otherwise. We prove that with indivisible items and a cake, an EFM allocation always exists for any number of agents with additive utilities. We reduce the problem of showing the existence of EFM to the problem of proving the compatibility of EF1 and envy-freeability. An allocation of indivisible items is said to be envy-freeable if the allocation can be made envy-free by allocating a sufficient amount of money to each of the agents. We show that for indivisible items, an EF1 and envy-freeable allocation always exists. Moreover, such an allocation can be found in polynomial time. This result, in and of itself, is of independent interest.

cs.GT↗
arXiv2026

Forecasting implied volatility surface with generative diffusion models

We introduce a conditional Denoising Diffusion Probabilistic Model (DDPM) for generating one-day-ahead approximately arbitrage-free implied volatility surfaces. To capture the path-dependent nature of volatility dynamics, we condition our model on a set of market variables, including exponentially weighted moving averages (EWMAs) of historical vol-surfaces, returns and squared returns of the underlying asset, and scalar risk indicators associated with the underlying asset. A key challenge is that the historical data used for training often contain arbitrage opportunities, particularly in the earlier part of the sample, which conflicts with the goal of generating arbitrage-free surfaces. We address this by using a weighting scheme based on the signal-to-noise ratio (SNR) to incorporate the arbitrage penalty into the loss function. The scheme dynamically adjusts the penalty strength across the diffusion process. Through numerical experiments using market data on S\&P 500 options, we demonstrate the superior performance of our proposed model in volatility forecasting compared to existing modern and classical approaches.

q-fin.CP↗
arXiv2026

CENIC: Convex Error-controlled Numerical Integration for Contact

State-of-the-art robotics simulators operate in discrete time. This requires users to choose a time step, which is both critical and challenging: large steps can produce non-physical artifacts, while small steps force the simulation to run slowly. Continuous-time error-controlled integration avoids such issues by automatically adjusting the time step to achieve a desired accuracy. But existing error-controlled integrators struggle with the stiff dynamics of contact, and cannot meet the speed and scalability requirements of modern robotics workflows. We introduce CENIC, a new variable-step integrator that brings together recent advances in convex time-stepping and error-controlled integration, inheriting benefits from both continuous integration and discrete time-stepping. CENIC runs at fast real-time rates comparable to discrete-time robotics simulators like MuJoCo, Drake and Isaac Sim, while also providing guarantees on accuracy and convergence.

cs.RO↗
arXiv2026

Beyond the Monolithic Scheduler: Lightweight Adaptive Policy Switcher for Process Scheduling Optimization

Modern operating system process schedulers typically enforce a static, ``one-policy-fits-all'' design, which inevitably compromises performance under the highly dynamic and mixed workloads of contemporary edge systems. Transitioning to dynamic adaptation faces two fundamental challenges: scenario specific static policies lack robust runtime adaptability, whereas computationally intensive AI-assisted policies introduce unacceptable execution overhead for resource-constrained devices. To structurally resolve this dilemma, we propose the Lightweight Adaptive Policy Switcher (LAPS), a framework that dynamically switches between predefined expert policies, thus decoupling high frequency context switching from low-frequency policy transitions and amortizing transitional overheads. To eliminate prohibitive online computations, LAPS pre-resolves all heavy dynamic tradeoffs via an offline preparation phase, constructing a strictly overhead-aware Policy Mapping Table. During runtime, it executes an agentic Observe-Infer-Switch cycle that limits machine learning inference strictly to workload classification. Consequently, LAPS executes efficient dynamic adaptation simply by mapping the inferred workload state to optimal expert policies via low-cost table lookups. Experiments show that LAPS achieves an average performance gain of 8.91% and reduces response latency by up to 91.7% in latency-sensitive tasks over the default Linux scheduler EEVDF. Our results demonstrate that LAPS outperforms the baseline in a wide range of test cases, offering a practical path toward more responsive and intelligent operating systems. Our code is publicly available at https://github.com/lvmiao233/LAPS.

cs.DC↗
arXiv2026

Formal Foundations for Controlled Stochastic Activity Networks

We introduce Controlled Stochastic Activity Networks (Controlled SANs), extending classical SANs with explicit control actions and policies for stochastic discrete-event and real-time systems. The framework combines nondeterminism, probabilistic branching, and stochastic timing through controlled automata, controlled probabilistic automata, and controlled stochastic automata. At the nondeterministic level, we distinguish structurally available controls from behaviorally admissible transitions and establish a constructive modeling-power result: every decidable, effectively finitely branching controlled automaton with finite activity and control sets and a finite nonempty initial-state set admits a computable, well-behaved controlled activity network realization whose reachable restriction is effectively isomorphic to the original. We define memoryless, finite-memory, stack-augmented, and tape-augmented policies, investigate their language-theoretic expressiveness, and develop behavioral equivalences including bisimulation and stochastic isomorphism. At the probabilistic level, contingent activity-indexed control assignments and an activity-generation law yield a discrete-time Markov decision process (DTMDP) representation; conversely, every conventional DTMDP admits a single-activity representation. Under Markovian assumptions, controlled stochastic automata admit continuous-time Markov decision process (CTMDP) projections, and conventional CTMDPs admit single-activity representations. Randomized-policy correspondences follow the adopted product-form convention. These results provide a semantic basis for verification, reward optimization, policy synthesis, and potential applications to reinforcement learning, safety-critical control, and agentic AI.

cs.FL↗
arXiv2026

Specter: Fast MoE Inference via Quantization-Enabled Multi-Step Speculative Offloading

Mixture-of-experts (MoE) models often exceed a single accelerator's memory capacity. Offloading experts to host memory creates a severe PCIe I/O bottleneck because data-dependent routing keeps transfers on the critical path. We present Specter, an inference system that uses quantization to combine speculative decoding with expert offloading. Its lightweight quantized draft model runs efficiently on device while preserving high-fidelity expert activation patterns. This provides accurate multi-step lookahead over future expert usage, allowing the runtime to organize expert movement and execution ahead of verification and improve both communication and execution efficiency. Under single-request inference, Specter achieves an average 1.50x (up to 1.88x) speedup in time per output token (TPOT) over the state-of-the-art system. Specter makes memory-constrained, single-GPU MoE inference more practical on commodity hardware. Our code is available at https://github.com/wwwwwf/specter_asplos.

cs.LG↗
arXiv2026

BrainRotViT: Transformer-ResNet Hybrid for Explainable Modeling of Brain Aging from 3D sMRI

Accurate brain-age estimation from structural MRI provides a valuable biomarker for studying healthy aging and neurodegenerative disease. We propose BrainRotViT (Brain ResNet over trained Vision Transformer), a two-stage, slice-based (2.5D) framework that approximates volumetric context while combining the strengths of CNNs and transformers. A Vision Transformer is pretrained on an auxiliary age-sex composite classification task and subsequently frozen. Its per-slice embeddings are stacked into a 160 x 768 feature matrix and processed as a pseudo-image by a lightweight 2D residual CNN regressor, with subject sex incorporated at the final prediction layer. Trained and validated on 5,108 samples from 11 multi-site datasets and tested on four independent held-out cohorts comprising 948 samples, BrainRotViT achieves a validation mean absolute error (MAE) of 3.43 years (Pearson r = 0.98, $R^2$ = 0.96, Spearman $ρ$ = 0.97) and cross-cohort MAEs of 4.79-5.35 years, outperforming retrained 3D-ResNet, SFCN, 3D-ViT, Global-Local Transformer, Triamese-ViT, and TSAN baselines. After age-bias correction and covariate adjustment, brain-age gap is significantly associated with Alzheimer's disease, mild cognitive impairment, and autism spectrum disorder. An interpretability pipeline combining ViT attention with guided backpropagation, saliency sanity checks, and AAL3-based anatomical localization identifies aging-related regions including the occipital lobe, cerebellum, precentral and postcentral gyri, temporal lobes, and medial superior frontal gyrus. BrainRotViT thus provides an efficient and interpretable bridge between CNN and transformer paradigms, demonstrating cross-cohort generalization while yielding biologically meaningful insights into brain aging.

cs.CV↗
arXiv2026

Inverse reconstruction of Einstein cluster collapse and oscillations in tangential-pressure spacetimes

We formulate an inverse reconstruction procedure for spherically symmetric Datta--Bondi--Gair spacetimes with vanishing radial pressure and non-zero tangential pressure supported by angular momentum. Extending the logic of Krasiński--Hellaby reconstruction for Lemaître--Tolman--Bondi dust, we determine the full, exact space-time information from prescribed density, velocity or equilibrium state. Two specific constructions are developed. The first represents an initially expanding fluid with a finite turnaround, followed by an asymptotic collapse to an unstable Einstein cluster configuration. In the second, a static equilibrium density and a finite time density-velocity pair determine finite amplitude oscillations around a stable Einstein cluster. Numerical examples use a Jaffe-type final profile and an Einasto equilibrium with an approximate Navarro--Frenk--White finite time profile. The latter admits many oscillations before the first shell crossing is detected. The constructions provide diagnostics of the admissibility of prescribed configurations.

gr-qc↗
arXiv2026

On the Complexity of Bilevel Linear and Quadratic Programs in Fixed Dimensions

We study the computational complexity of bilevel linear programs (BLPs) and bilevel quadratic programs (BQPs) with continuous variables under restrictions on the numbers of variables or constraints. Our results identify new transitions between computational complexity regimes that are inherent to the hierarchical structure of bilevel optimization problems and have no direct counterparts in single-level optimization. First, we demonstrate that, under comparable assumptions, the pessimistic formulation can be computationally harder than its optimistic counterpart. Specifically, while optimistic BLPs are known to be polynomially solvable when the number of follower variables is fixed, we prove that the corresponding pessimistic problem is strongly NP-complete. We also show that this complexity transition is tied to the presence of a nonfixed number of coupling constraints. Second, we show that, in the pessimistic setting with a fixed number of follower constraints, BLPs are polynomially solvable, whereas replacing the follower's linear objective function with a convex quadratic one renders the problem NP-hard. Importantly, this complexity transition occurs even in the absence of coupling constraints. Finally, we demonstrate that replacing a convex quadratic follower objective with a nonconvex quadratic one renders the optimistic problem NP-hard, even when both follower dimensions are fixed and small.

cs.CC↗
arXiv2026

Reflection-Based Relative Localization for Cooperative UAV Teams Using Active Markers

Reflections of active markers in the environment are a common source of ambiguity in onboard visual relative localization. This work presents a novel approach that exploits these typically unwanted reflections for onboard relative localization in heterogeneous multi-UAV teams operating over specular, near-planar surfaces such as calm water and flat indoor floors. Using only a camera, active markers, and a single-beam rangefinder, the method requires no prior knowledge of robot size, marker configuration, or surface reflectance properties, and propagates the uncertainty introduced by surface irregularities, including those of dynamic water surfaces. We further characterize, analytically and experimentally, how surface tilt biases the estimate, and identify the slope range over which the level-surface assumption holds. Indoor multi-UAV and outdoor experiments validate the localization accuracy, demonstrating reliable operation across varying lighting conditions and greater effective range than state-of-the-art vision-based active-marker methods, with higher accuracy beyond 30 m. The project website with the video is available under the following link: https://timlakemann.github.io/reflection-based-relative-localization/.

cs.RO↗
arXiv2026

Weinstein Handlebodies for the Painlevé Betti Spaces

We prove that the generic fibre of the Betti moduli space associated to any of the ten Painlevé equations coincides with the result of attaching Weinstein handles along the Stokes Legendrian, and provide Weinstein handlebody diagrams for each of them in the process. We moreover extend the procedure for obtaining presentations in Gompf normal form of Weinstein manifolds obtained by attaching handles to Legendrian lifts of cooriented curves in surfaces to their unit cotangent bundle to the case where the surface may have nonempty boundary.

math.SG↗
arXiv2026

Can discrete-time analyses be trusted for stepped wedge trials with continuous recruitment?

In stepped wedge cluster randomized trials (SW-CRTs), the intervention is rolled out to clusters sequentially over multiple periods. It is common practice to analyze data from SW-CRTs using linear mixed models that treat time as discrete. However, a recent systematic review found that 95.1\% of cross-sectional SW-CRTs recruit individuals continuously over time. Despite the prevalence of continuous recruitment, there has been limited guidance on how to draw model-robust inference for such designs. In this article, we conduct simulation studies to examine the performance of discrete-time linear mixed models in the case of continuous recruitment designs with a continuous outcome. Specifically, we generate outcomes under a continuous-time exponential decay correlation structure with either discrete or continuous fixed period effects. We consider constant and exposure-time-dependent intervention effects, with and without random intervention effects across clusters. We then analyze the simulated data using linear mixed models under three discrete-time working correlation structures: simple exchangeable, nested exchangeable, and discrete-time exponential decay, with robust sandwich variance estimation. Our results demonstrate that discrete-time analysis often yields negligible bias and that the robust variance estimator with the Mancl and DeRouen correction consistently achieves nominal coverage and type I error rate. However, when recruitment patterns vary systematically between control and intervention periods, discrete-time analyses give slightly biased intervention effect estimates. Finally, we illustrate these findings by reanalyzing a completed SW-CRT.

stat.ME↗
arXiv2026

Testing Bump in the Cosmological Power Spectrum Using Dwarf Galaxies

We analyze the possibility of using observational data on nearby dwarf galaxies -- their luminosity functions and spatial distributions -- to constrain deviations of the cosmological power spectrum from the standard one. Specifically, we consider a cosmological model with a "bump" in the power spectrum at a wavelength of 1.3~Mpc and a dimensionless amplitude $\mathcal{A}=2.0$. Such a spectrum is motivated by observations of an excess number of galaxies at high redshifts. The bump leads to a noticeable increase in the luminosity function in the range $-13 > M_B > -17$ at $z=0$. Comparison with observations constrains the bump amplitude to $\mathcal{A} < 0.25$ at a 3-sigma significance level for a wavelength of 1.3~Mpc. For wavelengths smaller than 0.8~Mpc, the bump manifests only in the luminosity function of dwarfs with $M_B > -14$.

astro-ph.CO↗
arXiv2026

FLAWS: A Benchmark for Error Identification and Localization in Scientific Papers

The identification and localization of errors is a core task in peer review, yet the exponential growth of scientific output has made it increasingly difficult for human reviewers to reliably detect errors given the limited pool of experts. Recent advances in Large Language Models (LLMs) have sparked interest in their potential to support such evaluation tasks, from academic peer review to automated scientific assessment. However, despite the growing use of LLMs in review systems, their capabilities to pinpoint errors remain underexplored. In this work, we introduce Fault Localization Across Writing in Science (FLAWS), an automated benchmark consisting of 713 paper-error pairs designed to evaluate how effectively LLMs detect errors that undermine key claims in research papers. We construct the benchmark by systematically inserting claim-invalidating errors into peer-reviewed papers using LLMs, paired with an automated evaluation metric that measures whether models can identify and localize these errors. Developing such a benchmark presents unique challenges that we overcome: ensuring that the inserted errors are well-defined, challenging, and relevant to the content of the paper, avoiding artifacts that would make identification trivial, and designing a scalable, automated evaluation metric. On the resulting benchmark, we evaluate five frontier LLMs: Claude Sonnet 4.5, DeepSeek Reasoner v3.1, Gemini 2.5 Pro, GPT 5, and Grok 4. Among these, GPT 5 is the top-performing model, achieving 39.1% identification accuracy when k=10, where k is the number of top-ranked error text candidates generated by the LLM.

cs.CL↗