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 1,171 records · Page 65Linked to original sources

arXiv2026

RCSD: Reinforcement and Continual Learning for Domain-Specific Speculative Decoding

Speculative decoding accelerates large language model inference by employing a compact draft model to propose candidate tokens that are verified in parallel by a target model. Conventional training relies on soft-label distillation, which forces the draft model to mimic the target's full output distribution, including high-entropy regions that degrade acceptance rates and increase early rejection risk. We propose RCSD (Reinforcement and Continual Learning for Speculative Decoding), a training framework that aligns draft model optimization with the system-level objective of maximizing acceptance length. RCSD leverages naturally occurring accepted and rejected sequences during speculative inference to construct implicit preference pairs, with masked loss to avoid penalizing target-aligned tokens. Evaluated across multiple benchmarks, RCSD improves average acceptance length by 35% over HASS and achieves 1.35x faster end-to-end inference than HASS on practical workloads, without compromising generation quality.

cs.CL↗

arXiv2026

Are Joint-Embedding Predictive Objectives Effective for Longitudinal EHRs?

Masked language modelling (MLM) is the dominant pretraining objective for transformer encoders applied to longitudinal electronic health records (EHRs). However, reconstructing individual events may overemphasise noisy or weakly informative observations and does not explicitly encourage learning higher-level patient states. Joint-Embedding Predictive Architectures (JEPAs) instead learn by predicting representations in latent space and have shown strong performance in a variety of domains. In this paper, we investigate to what extent JEPA can be adapted to EHR sequence representation. Our empirical investigation using MIMIC-IV shows that although all three of our JEPA formulations train stably without representational collapse, they consistently underperform standard MLM. We find that target size, not target heterogeneity, increases prediction error, yet reducing the target size does not recover downstream performance. This suggests that stable latent prediction alone is insufficient for competitive EHR representation learning and highlights the challenge of defining clinically meaningful latent targets for EHR.

cs.LG↗

arXiv2026

Impact of Torque Ripple Induced Angular Oscillations on Maxwell Pressure Harmonics in Permanent-Magnet Synchronous Machines

Torque ripple in electrical machines is a wellknown source of noise and vibration. While numerous studies focus on control strategies to mitigate this ripple, the feedback mechanism by which the resulting mechanical oscillations influence the electromagnetic forces remains less explored. This paper presents a detailed investigation into the impact of torque ripple on the rotor angular dynamics and the subsequent generation of new harmonics in the Maxwell pressure distribution. We first develop an analytical model for a rigid rotor subjected to a single torque harmonic, revealing the inherently non-linear nature of the angular response. This analysis demonstrates the creation of new harmonics in the rotor motion and a slight reduction in the mean rotational speed. The study is then extended to a flexible rotor model incorporating torsional modes, solved using the Finite Element Method. Finally, we introduce a strongly coupled magneto-mechanical model, based on the subdomain method, to quantify how these rotor oscillations modulate the air-gap magnetic field. The results show a new modulated spatio-temporal harmonic distribution in the Maxwell pressure spectrum, which is not predicted by conventional uncoupled models. These findings are crucial for high-fidelity Noise, Vibration, and Harshness (NVH) analysis, as these additional pressure waves can excite stator structural modes.

physics.class-ph↗

arXiv2026

Residual-Based Causal Discovery via Pretrained In-Context Quantile Regression

Residual-based causal discovery assumes that each variable is generated from its parents and an independent noise source. It estimates this noise and tests its independence from candidate causes. For identifiable functions such as additive-noise models (ANMs) and the broader class of location-scale noise models (LSNMs), this principle identifies the causal direction, as only the causal model admits this independence. Existing LSNM methods rely on parametric assumptions to obtain standardized residuals, risking model misspecification. In contrast, the RESIT-framework allows for nonparametric estimation of residuals to iteratively identify sink nodes, but was proposed for ANMs. To close this gap, we propose Loscar, an extension of RESIT to LSNMs that estimates the conditional mean and scale via quantile regression, enabling pretrained tabular foundation models as plug-in estimators. We compare it with a feasible generalized least squares (FGLS) variant, which is consistent given suitable basis functions. Across bivariate and multivariate causal-discovery benchmarks, in-context quantile regression substantially outperforms FGLS with flexible function classes, particularly for multivariate causes.

stat.ML↗

arXiv2026

Phase Transitions in Learning Driven by Extremal Data

We study an extremal regression problem, an example of multiple instance learning (MIL), where the predictors are given as multisets, called bags. The bag elements (called instances) possess hidden labels which are pooled to produce an observed bag-level label. Max pooling is common in MIL applications, but gives rise to vanishing gradients. To handle this, modern MIL pipelines approximate the max with a softmax, treating the sharpness as a learnable parameter, denoted $λ$. When the max pooling model is the true data generating process, learning will be successful when $λ$ is increased during gradient descent, thereby driving the dynamics into regions of parameter space with vanishing gradients. To understand the conditions under which gradient descent breaks down, we study a minimal model and derive a low dimensional projection of the gradient flow in the joint limit of large bags $n$ and high dimension $d$ with $α=\ln n/d$ fixed. The flow has two distinct phase transitions: an intrinsic freezing transition related to the random energy model of disordered systems, and a retrieval transition occurring in modern Hopfield networks. Numerical simulation confirms both the phase diagram and the convergence of the finite-size training gradient to its predicted asymptotic form.

cs.LG↗

arXiv2026

Minimax Rates for imperfect Physics-Informed Regression

Physics-informed machine learning incorporates physical knowledge into data-driven models, often through a partial differential equation (PDE) that the target is assumed to satisfy. However, how much this knowledge improves statistical performance remains unclear, especially when the physical model is imperfect. Since satisfying a PDE implies being differentiable, a natural benchmark for physics-informed regression is the best achievable performance when smoothness is the only prior information about the target function. We show that there is a characteristic threshold for the modeling error above which the physical prior cannot improve the convergence rate. The form of this threshold is common to every linear PDE and certain nonlinear PDEs. Below it, the gain depends on the equation type. For classes of elliptic and hyperbolic PDEs with constant coefficients, we establish convergence rates that improve continuously as the modeling error decreases, and prove their minimax optimality up to logarithmic factors. With exact physics, the rates correspond to learning in dimension d-1 rather than d, reflecting the freedom left by unknown boundary or initial data. We also give conditions for computing the estimator with near-linear complexity and show how to tune its hyperparameters without knowing the modeling error.

math.ST↗

arXiv2026

Block Diffusion for Longitudinal EHR: Accuracy, Efficiency, and Representation Trade-offs

Autoregressive (AR) models have proved effective for zero-shot prediction from longitudinal electronic health records, but impose strictly sequential generation on trajectories containing temporally clustered and interdependent clinical events. We investigate block-based masked diffusion (BD) as an alternative on MIMIC-IV under matched tokenisation, context length, model capacity, and evaluation conditions. Across block sizes, BD exhibits a clear quality--efficiency trade-off, with small-block configurations retaining much of the predictive performance of AR while enabling substantially faster inference. A representative $B{=}8$ configuration achieves approximately $1.43\times$ faster evaluation. Although generative rollout remains below AR in most settings, frozen linear probes on BD and AR representations achieve near-identical performance across clinical prediction tasks, indicating that BD learns comparably informative clinical representations. Increasing the denoising budget yields diminishing returns, while alternative decoding strategies provide limited gains. Overall, our results show that block diffusion is a viable and computationally attractive approach to longitudinal EHR modelling, with further improvements in iterative decoding offering a promising route toward closing the remaining generative performance gap.

cs.LG↗

arXiv2026

Safe World Model for Joint Trajectory and Beamforming Design of Mobile Base Stations

Mobile base stations (MBSs), including aerial and ground-based platforms, provide flexible wireless coverage through adaptive positioning. However, jointly optimizing their trajectories and digital beamforming is challenging because movement changes channel conditions and interference, while beamforming determines the resulting signal and interference powers. These coupled decisions also account for how current movements affect future positions and communication performance under mobility and safety constraints. We formulate this problem as a constrained Markov decision process that maximizes the time-averaged sum signal-to-interference-plus-noise ratio (SINR), subject to mobility, restricted-zone avoidance, and transmit-power constraints. To support long-term planning and reduce the extensive environment interactions required by model-free reinforcement learning, we develop a SafeDreamer-based world model framework that learns latent system dynamics from experience and optimizes a joint trajectory and beamforming policy through imagined trajectories. Feasibility-preserving action mappings and safety cost signals incorporate operational constraints into policy learning. Simulation results demonstrate that the proposed framework improves the time-averaged sum SINR by an average of 7\% and 9\% compared with proximal policy and deep Q-network baselines, respectively, while achieving higher sample efficiency and more reliable constraint satisfaction.

eess.SY↗

arXiv2026

Large-size multivariate scale-free texture generation from small-size learning sets with self-attention Wasserstein-GAN

Multivariate scale-free texture images are nowadays central in numerous modern applications very different in nature. A common challenging issue consists of being able to generate such multivariate textures at any arbitrary size, from a learning set of (possibly much) smaller size. Parametric models often turn out to be too restrictive for the generation of realistic textures, thus prompting neural network training-based generation. While several neural network approaches propose generation at sizes larger than training, they do not explicitly assess the quality of the generation by quantifying how well multiscale or scale-free statistics in textures are reproduced and rather rely on perceptual assessment. The present work combines convolutional Wasserstein-GAN architectures with self-attention layers to generate textures of arbitrary size while focusing on reproducing multiscale statistics, without coding a model into the loss function. The relevance of using a self-attention layer is studied. Quantitative assessment conducted on multifractal textures, representative of a broad class of multiscale textures, over ten independent trainings shows that the proposed architecture is able to generate textures with accurate statistics for sizes $8 \times 8$ larger than the originals. The results show that placing self-attention in the critic provides the most consistent performance across output sizes.

eess.SP↗

arXiv2026

A recovery based linear finite element method for the Monge--Ampère equation

In this paper, we study finite element approximations of a fourth-order perturbed semilinear partial differential equation that arises naturally in the computation of viscosity solutions to the fully nonlinear Monge--Ampère equation. We develop a novel recovery-based $C^0$ linear finite element method for discretizing this model problem. The key analytical tool is a Miranda--Talenti-type estimate for the recovery operator, which guides the formulation of the discrete scheme and guarantees its well-posedness. We rigorously establish convergence in the $H^2$ norm by employing a fixed-point argument. Furthermore, optimal $H^1$-error estimates are derived using a duality argument. Numerical experiments confirm the efficiency of the proposed method and validate the theoretical convergence results.

math.NA↗

arXiv2026

Melange: A Synthetic Visual Bias Benchmark for Multi-modal Large Language Models

Multi-modal Large Language Models (MLLMs) have seen broad adoption, but their biases are still far less examined than those of standard LLMs, creating possible social harms when deployed in real-world settings. Prior efforts to extend bias analyses from LLMs into the multi-modal space either cover only a few bias categories or do not ensure that the groups being compared are clearly addressed by visual cues in the images. To overcome these limitations, we introduce Melange, a wide-ranging synthetic visual benchmark covering seven categories for assessing social biases in MLLMs. It is grounded in the widely used language-only BBQ dataset (Bias Benchmark for Question Answering) from BigBench, carefully extending its constructed contexts via controlled generation of visual scenarios so the target attributes appear visually and clearly. To maximize reliability, our generated images pass through two stages of validation: automated screening using visual question answering (VQA) and subsequent manual inspection by human annotators. By systematically evaluating 23 open-source MLLMs across three matched evaluation settings---text-only, image-only, and image-with-context---Melange isolates the fairness effect of the input modality. Switching from text to vision reduces fairness by 0.291 on average, a gap that varies by model family rather than by scale alone. A perception control confirms that this gap is not an artifact of visual comprehension: models identify the depicted attributes at 90--95\% accuracy yet still fail to abstain. When models do commit to a biased answer, polarity-aware stereotype alignment shows they choose the human-stereotypical response roughly three-quarters of the time.

cs.CV↗

arXiv2026

Stochastic Pruning of the DPO Computation Graph: 1.4x Faster Training and Lower Peak Memory

Direct preference optimization (DPO) incurs training costs in output scoring, sequence execution, and backpropagation, so reducing supervision alone does not necessarily save computation. We introduce a unified framework for stochastic pruning of the DPO computation graph that independently samples response positions for scoring, response prefixes for execution, and preference branches for backpropagation. By making these selections before the corresponding operations, the framework omits computation at each stage of a training step. Our analysis characterizes the objective changes induced by position and sequence pruning and uses inverse-probability weighting to obtain a conditionally unbiased single-branch gradient estimator for the sampled objective. Controlled measurements show that position and sequence pruning primarily reduce peak memory, with most output-memory savings attributable to an implementation that retains full response supervision. Branch pruning yields a $1.426\times$ step-time speedup over a concatenated DPO baseline and reduces training time by about 30\% relative to stock DPO across two model families, with slightly lower mean accuracy and benchmark-dependent changes in open-ended scores. These findings relate the resource benefits of stochastic graph pruning to the computation omitted and the resulting changes in the training objective.

cs.LG↗

arXiv2026

Cartan Subalgebras of Dimension Drop Interval Algebras and Their Inductive Limits

We describe the conjugacy of non-degenerate Cartan subalgebras of the dimension drop interval algebras $M_m(\widetilde{\mathbb I}_p)$ in terms of double cosets of permutation groups, and describe their spectra by finite bi-partite multigraphs. All Cartan subalgebras are non-degenerate precisely when the two parameters are coprime. We also give splitting interval examples with prescribed finite numbers of Cartan conjugacy classes. Passing to inductive limits, we construct a unital simple AD algebra $\mathcal{A}$ of real rank zero and stable rank one, with $K_0(\mathcal{A})\cong\mathbb Z[1/3]$ and $K_1(\mathcal{A})\cong\mathbb Z/2\mathbb Z$. It contains infinitely many mutually isomorphic but pairwise nonconjugate $\mathrm{C}^*$-diagonals. Their spectra are all homeomorphic to the dense amalgam of circles, while orbit multiplicities in the Weyl groupoids distinguish their embeddings. The same AD algebra also contains infinitely many pairwise nonisomorphic Cartan subalgebras.

math.OA↗

arXiv2026

A compact ultracold strontium experiment

We describe a compact and simplified experimental setup for the production of ultracold strontium atoms. Thanks to its low-power oven and coil-free Zeeman slower design, the vacuum system has an overall length of only 40cm including oven, Zeeman slower and all-glass science cell. The oven has a power consumption of 14W at 460°C and produces a collimated flux of 3 10^10at/(s cm^2) at this temperature. The laser system employs master-slave laser architecture throughout and requires neither second-harmonic generation nor tapered amplifiers, reducing cost and complexity. The system has been in continuous use for more than a year.

physics.atom-ph↗

arXiv2026

Domain models of turbulent magnetic fields and the spectral shape of ensemble averaged axion-photon conversion

Axion-like particles are light pseudoscalars that appear in many extensions of the Standard Model and are candidates for dark matter. Through their mixing with photons, they imprint energy dependent modulations on the spectra of astrophysical sources shining through magnetised media. When a distant source shines through the turbulent magnetic field of an intervening galaxy cluster, these modulations are stochastic. It has recently been shown that averaging over an ensemble of such sources instead produces a smooth absorption-like feature. In this work we derive the analytic form of this ensemble averaged signal, previously known only numerically. We further relate the index $a$ of the magnetic field energy spectrum to the index $β$ of the domain size distribution that simulations use to describe the turbulent field. For $1<a<2$ the two indices obey the simple analytic relation $β=3-a$. Any distribution with $β<1$ instead mimics the spectral index $a=2$, which the sharp domain walls impose. For Kolmogorov turbulence, relevant for galaxy clusters, the relation gives $β=4/3$.

astro-ph.HE↗

arXiv2026

Broadcasting in Centimeter-Wave Gigantic MIMO Systems: Beam Shaping is All You Need

Beamforming is a key enabling technology for centimeter-wave Gigantic MIMO (G-MIMO) systems, using antenna arrays to generate wide beams for cell-wide broadcasting or narrow beams for data transmission. This paper presents a novel \emph{beam shaping technology} for designing environment-aware broadcast beams (i.e., beams for common control information delivery, as in 5G SSB). As a case study, we shape a \emph{prostrate beam} in the vertical plane to approximate uniform electromagnetic power distribution across the ground, thereby enhancing broadcast coverage quality. We first propose a geometric broadcasting model that, through coordinate transformation, relates base station height, beam spread, maximum service distance, and downtilt angle. Building on this model and the signal propagation model, we formulate the beam shaping problem as a non-convex optimization problem that aims to maximize ground-level coverage. To solve it, we develop a generalized optimization framework. We reformulate the problem into a matrix representation, equivalently transforming it into a target pursuit optimization (TPO) problem. Using variable substitution, the objective is convexified into a quadratically constrained quadratic program, and the non-convex rank constraint is handled via successive convex approximation, decomposing the problem into iteratively solved subproblems. We provide complexity and convergence analyses for the proposed algorithm. Numerical simulations confirm that the optimization effectively shapes the intended beam pattern and that the proposed prostrate beam yields significant performance gains, including substantially extended coverage distance, over conventional wide-beam benchmarks.

cs.IT↗

arXiv2026

Detectable Dark Matter Freeze-in via Low Reheating

This work focuses on freeze-in production of dark matter (DM) during low reheating, where the latter arises from the decay of inflatons leading to a non-adiabatic matter-dominated era. Under freeze-in, DM comoving yield is inversely proportional to Hubble, and during matter domination, as Hubble increases, the coupling, keeping the DM interaction with SM, should be enhanced to have relic density $Ω_{DM}\simeq0.12$, making DM potentially vulnerable to direct detection searches. A renormalizable Yukawa-like Lagrangian is considered for DM interaction with a singlet electron $e$, and a BSM parent particle $P$. With the decay being the dominant freeze-in production channel, the required coupling enhancement is calculated with the aid of freeze-in Boltzmann equation. Dealing with freeze-in, the higher coupling could thermalize the feebly interacting particles, and this places an upper bound on coupling. The coupling has been found to be sensitive to the vacuum energy of the inflaton during the inflationary phase. Keeping this vacuum energy at the upper bound from {\tt planck}, the model fails to feature any detectable direct detection signals even at a reheating temperature of BBN. The findings of this thesis then show that, only under low reheating and low inflationary vacuum energy, the DM-electron cross sections become high enough to be detectable in near-future direct detection searches. However, the proposed model becomes non-perturbative near the direct detection exclusion limits.

hep-ph↗

arXiv2026

RISK-KV: Quality-Aware Selective KV Cache Reconstruction for Distributed LLM Inference

Request migration and interruption recovery in distributed large language model (LLM) inference rely on historical key-value (KV) caches. Retaining or transferring a complete cache incurs substantial state overhead, whereas replaying historical tokens requires additional computation. This paper proposes RISK-KV, a quality-aware method for selective historical KV reconstruction in pretrained models. RISK-KV uses exact KV states on both sides of a target layer and cross-layer ridge regression to predict the missing layer's attention-projection input; it then folds in the original model's KV projection matrices to reduce predictor storage. Reconstruction risk is measured by the change in negative log-likelihood (NLL) caused by actual reconstruction, and low-overhead topology search with quality gating selects the layer groups to reconstruct. Independent tests on three language models show that representative recovery configurations omit 18.2 percent to 33.3 percent of historical KV, with perplexity changes below 0.5 percent. In a local request-recovery experiment with a 512-token history, RISK-KV achieves a 1.51 to 3.31 times speedup over full Token Replay. These results indicate that selective KV reconstruction can reduce historical cache overhead with small changes in predictive quality and improve request-recovery efficiency under the evaluated deployment conditions.

cs.DC↗