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 1,783 records · Page 99Linked to original sources

arXiv2026

SPECOLA: A Tool for Specification Driven Robot Evaluation

Simulation benchmarks are a useful way to measure progress in robotics; typically they provide success rate as the dominant measure of performance. However, success rate reports only whether a rollout completed the task, hiding by what margin it succeeded, how sensitive the success is to goal parameter changes, requirements it may have violated, and whether success persisted. Signal Temporal Logic (STL) provides a formal language for capturing specifications; it can be used to define requirements, and measure how close a robot behavior is to satisfying or violating them. Using STL monitors in a simulator, however, requires exposing simulator state and mapping it to specification-relevant signals. We present Specola, a tool for evaluating robot behaviors in simulation with respect to STL specifications. Specola provides an interface for writing specifications over simulator signals, and creates and manages the relevant monitors while a policy executes. We provide a Python API and command-line interface for integrating monitoring into existing policy execution loops. We provide an adapter interface for connecting new simulators, and provide adapters for nine simulation environments and benchmarks, including LIBERO, RoboCasa, Isaac Lab, and RoboDojo. We provide case studies that illustrate the nuanced evaluation that Specola enables. Code and adapters are available at https://github.com/specola-eval/specola.

cs.RO↗

arXiv2026

Whittle--Matérn metrics and isotropic Gaussian processes on compact metric graphs

Isotropic Gaussian processes on networks have a covariance that is a function of a distance between locations, but valid constructions have so far required graphs with Euclidean edges and the resistance metric. We introduce a one-parameter family of metrics $d_κ$, $κ\ge0$, on arbitrary compact metric graphs, including those with loops and parallel edges, defined as the variograms of first-order Whittle--Matérn fields, where $κ=0$ corresponds to an intrinsic limiting field. We show that these Whittle--Matérn metrics are invariant under insertion and removal of vertices of degree two, that they decrease monotonically in $κ$, and that they can be evaluated by sparse linear solves. The metric $d_0$ coincides with the resistance metric on graphs with Euclidean edges, and thus extends it to general compact metric graphs without any modification of the graph. For every completely monotone function $r$ and every $κ\ge0$, we show that $r(d_κ)$ is a positive semidefinite function. This yields isotropic Gaussian processes on general compact metric graphs. For $κ>0$ the metric levels off beyond the screening length $κ^{-1}$, which can be used to induce long-range dependence in short-range covariance models, and $κ$ can be estimated from data together with the parameters of $r$. The models are compared on daily traffic maps from the street network of Paris and the motorway network around London. For both, $κ$ is estimated to be positive, $d_κ$ increases the maximised log-likelihood relative to $d_0$ for every covariance function, and it gives the best predictions at randomly held-out locations.

math.ST↗

arXiv2026

Multi-Objective Reinforcement Learning Using Routed Ensembles and Trajectory Attribution

Multi-objective reinforcement learning is a fundamental problem for modern AI, with applications spanning post-training, reasoning models, and robotic control. A central challenge is to learn policies that balance competing objectives, such as helpfulness versus honesty in safety and alignment, or speed versus energy in robotics. The optimal policy can vary sharply with objective-weighting combinations, and the extent of this variation depends crucially on the degree of conflict among objectives, making per-preference policy learning sample-inefficient. In this paper, we introduce objective-conflict coefficients and derive near-optimal regret bounds for online settings using this measure, thereby quantifying the impact of misalignment on learning efficiency. Then, we design routed ensemble policy learning, an algorithm that maintains a small set of basis policies and dynamically routes trajectories to the policy with the highest potential reward. Finally, we consider trajectory selection as another approach to mitigate objective conflict, and design a kernel surrogate trajectory attribution algorithm that estimates how feedback signals from a set of trajectories influence policy learning on other trajectories. The kernel can capture nonlinear interactions such as synergy and antagonism between trajectories. We evaluate our approach across diverse environments and benchmarks: routed ensembles improve average performance by 4.3% over baselines for online environments from Meta-World and MO-Gymnasium, and kernel surrogate attribution improves correlation with leave-one-out analysis over existing baselines by 50%. When applied to trajectory selection, kernel-based attribution yields a 27% improvement in reward over baseline attribution methods.

cs.LG↗

arXiv2026

Parametrically Enhanced Ergotropy and Charging Power in a Three-Mode Bosonic Quantum Battery

Quantum batteries are quantum systems that potentially serve as fuel for other quantum devices by receiving and supplying energy. In this work, we propose a theoretical scheme to enhance ergotropy and charging power of quantum Battery in the system under consideration through squeezed light. Moreover, The system comprises a coherently driven charger cavity, a strongly dissipative auxiliary cavity acting as a mediator, and an energy--storage battery cavity. Quantum nonreciprocity and directional energy flow are established via destructive quantum interference between the direct coupling path and the loss--engineered indirect path. A degenerate optical parametric amplifier integrated into the charger cavity acts as an active, phase--sensitive energy source that injects two--photon quantum excitations and anomalous correlations. Using a strictly open--quantum--system framework derived directly from the Lindblad master equation, we obtain the full expectation--value dynamics and stationary states. We show that active parametric driving drastically increases the stored energy and peak charging power while maintaining nonreciprocal energy trapping in the battery. We achieve a multi-order-of-magnitude boost in the battery's stored energy, charging power, and ergotropy. We also analyze the effects of pure dephasing on the charging power, ergotropy, and extractable work quality.. Although two--photon pumping introduces photon--number fluctuations that slightly lower the fractional ergotropy ratio ${\cal W}_{\rm B}/{\cal E}_{\rm B}$, the absolute extractable work remains substantially elevated compared to unassisted configurations.

quant-ph↗

arXiv2026

Decoding with alternating weighted graph sparsification for quantum low-density parity-check codes

Real-time decoding is a fundamental necessity for building fault-tolerant quantum computers. In this work, we introduce decoding with alternating weighted graph sparsification (DAWG), an ensemble based approach that combines Tanner graph sparsification with ensembling belief-propagation (BP) decoders. The DAWG decoder runs BP decoders over the full detector error model Tanner graph for certain iterations to then transfer the posteriors to a sparsified graph in which additional rounds of BP are run. Each of the decoders in the ensemble uses a different transfer matrix to map the posterior information to the sparsified graph. We use the serial schedule for the minimum-sum BP decoders and, thus, employ Vibe decoding ideas by randomizing the schedules of each of them. The performance of the DAWG decoder is then numerically studied for bivariate bicycle codes and superdense color code circuits to be compared with other state-of-the-art decoding approaches. Importantly, the DAWG decoder shows a similar performance to Relay belief-propagation and Vibe decoding when combined with an order-$0$ Localized Statistics Decoding (LSD-$0$) post-processor. Crucially, the post-processor is called an order of magnitude less times than for Vibe with the same ensemble size, indicating that the introduced latency tail will be minimal. Furthermore, LSD-$0$ operates on a sparsified decoding graph of a fraction of the size of the full detector error model, which simplifies the resources required by specialized hardware to support it. Overall, the DAWG+LSD-$0$ decoder represents a viable low-cost BP-based decoding algorithm for quantum error correction codes.

quant-ph↗

arXiv2026

Preference-Guided Cooperation in Multi-Objective Multi-Agent Reinforcement Learning

Multi-objective multi-agent reinforcement learning (MOMARL) addresses agent coordination and objective trade-offs simultaneously while optimising overall system performance. Existing work tends to treat multi-agent cooperation and multi-objective optimisation as largely independent components of a learning problem, which amounts to an exorbitant complexity except in trivial cases. We study here whether the structure provided by multiple objectives can itself be exploited to facilitate cooperation, rather than being treated as an additional optimisation challenge. We formulate MOMARL at both the team and individual levels and introduce objective-wise counterfactual credit assignment to evaluate each agent's contribution across the team objectives. We propose Meta-Gradient Informed Multi-Objective Multi-Agent Reinforcement Learning (MIRL), with static (MIRL-S) and dynamic (MIRL-D) variants. By incorporating meta-gradient information, MIRL enables agents to learn how to prioritise objectives individually while collectively optimising the system-level objective, allowing the multi-objective structure to directly guide cooperation. Experiments on a multi-objective multiple particle environment show that our methods improve Pareto-front quality over baselines, with the dynamic variant achieving the strongest performance.

cs.MA↗

arXiv2026

Local Broadband Gigahertz Spectroscopy via SQUID-on-Tip DC Readout

Superconducting microwave circuits are shaped by their local electromagnetic environment, yet standard microwave measurements do not resolve the local fields and device features that govern their behavior. Here, we combine microwave spectroscopy, nanoscale magnetic sensing, and topographic imaging in a single tapping-mode SQUID-on-tip microscope. The nonlinear Josephson response converts local GHz excitations into Shapiro steps, enabling quantitative measurements of microwave frequency signals and their excitation strength using only DC readout. Applied to a superconducting coplanar waveguide resonator, local spectroscopy reproduces the power-dependent resonance and nonlinear bifurcation observed in microwave transmission measurements. Spatially resolved measurements map the microwave response while simultaneously recording topography and static magnetic features, including vortices and Meissner screening currents. Electromagnetic simulations reveal how coupling through the superconducting probe leads limits spatial resolution and guides localized detector designs. Together, these capabilities establish a platform for correlating microwave circuit dynamics with the local structural and magnetic environment of superconducting devices.

cond-mat.supr-con↗

arXiv2026

Fuji: The Shape, Composition, and Spread of Demand Bursts on Robinhood Chain

Demand on blockchain-based marketplaces arrives in bursts and at a high rate. On these platforms the data are completely visible, which makes it feasible to compare the bursty demand against the usual load. The congestion manifests as an increased demand for gas, the primary resource a blockchain sells. Existing studies consider how to price the gas. The demand itself, and what it is made of during bursts, has not been measured. We measure it on Robinhood Chain, a layer-2 blockchain that quickly became a venue for trading tokenized equities and memecoins. We collected the data from every block of its first three active months and more than 50 million Uniswap swaps that occurred during bursty periods. We find that a burst trades about as many pools as a quiet period, mostly different ones, and trades them harder, on top of elevated demand that accumulated for hours. With the bursts aligned on their peaks, the median peak is 2.5 times the usual load. Both the rise and the fall of the demand are described by the sum of two exponential functions with different timescales, a slow one of hours and a fast one of minutes. Against a matched quiet period, a burst has 1.05 times the pools, 1.40 times the trades, and 1.21 times the senders. We use two models to explain the arrival of bursts: externally triggered cascades and self-excitation. The former fits the data relatively better. Inside a burst, the entry of each trader into a pool is followed for about 24 seconds by more entries than the existing traders predict. Whether this is because of imitation or because of a shared signal is left open.

econ.EM↗

arXiv2026

From skew Combinatorial Nullstellensatz to sharp Ax-Katz congruences and zero counts

We give a unified account of zero counting in finite-field polynomial rings with commuting variables twisted by a common automorphism. A decomposition into fixed-field slices connects skew evaluation with ordinary polynomial evaluation. First, we derive a slice coefficient formula from the skew Combinatorial Nullstellensatz and use it to deduce skew Chevalley-Warning, including the sufficient condition $n>h\sum_i d_i$. Next, classical Ax-Katz gives the corrected integer congruence $Q^{\max\{0,\lceil(n-h\sum_i d_i)/D\rceil\}}\mid N+(M-1)δ$. We construct homogeneous examples proving optimality for every positive degree table and every number of variables. We also establish a sharp lower bound with a prescribed zero, an input-sensitive coefficient-filtration refinement, and transfers of known bounds for nonlinear zero sets. Detailed calculations explain the origin correction, incomplete norm blocks, and the distinction between characteristic-$p$ summation and higher integer divisibility. Existing geometric, interpolation, and counting theorems are explicitly attributed.

math.CO↗

arXiv2026

LUCID: Local Understanding of Covariate Importance with Conditional Density Estimation

Computing local-level variable importance (VI) is critical for understanding machine learning models. Local permutation-based VI metrics offer a simple yet powerful approach, measuring how much a model's output changes when perturbing its inputs. However, existing perturbation strategies often ignore feature dependence or account for it only through restricted approximations. Perturbing a feature without fully respecting this dependence can generate unrealistic counterfactuals, thereby evaluating the model in regions unsupported by the observed data distribution. We address this gap by introducing a new model-agnostic local VI metric and estimator, Local Understanding of Covariate Importance with conditional Density estimation (LUCID). LUCID perturbs features of interest using plausible values drawn from the full conditional distribution given the remaining observed features. We establish theoretical connections between LUCID and existing VI measures, and provide an estimator with convergence guarantees and uncertainty quantification. Finally, we demonstrate LUCID's stable performance and practical utility across simulated and real-world examples.

cs.LG↗

arXiv2026

Graph Neural Network for Wide-Area 6DoF Ambisonic Rendering

Wide-area 6DoF rendering from sparse Ambisonic recordings is commonly addressed by interpolating parametric spatial descriptors rather than the waveforms themselves. This preserves the fidelity of a nearby measured signal and avoids phase artifacts caused by mixing distant arrays. However, existing DirAC-based approaches rely on geometric interpolation weights that do not depend on the acoustic scene observed by the microphones. We propose a lightweight Graph Neural Network (GNN) that refines only this interpolation stage. The known Ambisonic arrays and the query position are represented as a graph whose nodes contain local DirAC features and whose edges encode relative distance and directional relations. The network predicts positive query-conditioned weights over the measured arrays, while the target energy, diffuseness, and direction of arrival are reconstructed by an analytical DirAC layer. Experiments on unseen rooms, unseen speech signals, and unseen array configurations show that the proposed weighting improves energy reconstruction and radiation-pattern similarity with respect to baseline interpolation.

cs.SD↗

arXiv2026

Yang-Mills-Higgs Limits of Reflexive Higgs Sheaves

For stable Higgs bundles, Hermitian-Einstein metrics can be obtained as limits of the Yang-Mills-Higgs (Y-M-H) flow. We study the algebraic structure of subsequential limits of this flow for reflexive Higgs sheaves over compact Kähler manifolds, without assuming stability. We prove that the limiting reflexive Higgs sheaf is isomorphic to the double dual of the graded sheaf associated with the Harder-Narasimhan-Seshadri filtration of the initial sheaf.

math.DG↗

arXiv2026

SEMA-Auto: Sighted and Efficient Multi-Turn Attack with Autonomy

Existing jailbreak attackers fall into two classes: static planners fix their single- or multi-turn adversarial plans in advance, remaining simple but losing sight of the victim's replies, wasting turns and limiting transferability; interactive generators read each victim reply to generate the next-turn attack but require in-loop evaluators for success signals or hints, incurring substantial costs and inflated performance. Bridging this gap, we propose SEMA-Auto, a sighted and efficient multi-turn attacker with autonomy, interacting with the victim model and terminating the session by itself upon success. To stabilize its behaviors and boost its capabilities, SEMA-Auto is trained in two stages. Synthetic bootstrap tuning installs the notion of attack targets and conforming decisions by fine-tuning the base model with bootstrapped trajectories. Decision-supervised reinforcement learning then optimizes two objectives jointly yet separately, reinforcing the attack with higher jailbreak reward while supervising the decision at each turn against an absolute target. Our method achieves state-of-the-art (SOTA) attack success rates (ASR) across multiple datasets, victims, and judges, outperforming all static and interactive baselines, within almost the fewest turns, demonstrating both effectiveness and efficiency. For example, SEMA-Auto performs a $76.0\%$ ASR@1 within an average of $3.0$ turns against DeepSeek-V4-Flash on AdvBench, while the best static baseline is $54.8\%$ with fixed $7$ turns and the best interactive baseline is $43.8\%$ in $3.7$ turns. Against safety-hardened frontier models, GPT-oss-120B, GPT-5.4, Claude-Haiku-4.5, and GPT-6-Astra, our counter-trained attacker achieves an average $59.6\%$ ASR@1 on HarmBench, $13.8\%$ over prior SOTA. Our code is available at: https://github.com/microsoft/SEMA

cs.CL↗

arXiv2026

The Weyl Problem for Planar CBB$(-1)$ Surfaces in Hyperbolic $3$-Space

In this paper, we study the Weyl problem for noncompact convex surfaces in hyperbolic $3$-space $\mathbb{H}^3$. We prove that any planar CBB$(-1)$ surface with countably many ends can be isometrically embedded as the boundary of a closed convex subset $X\subset\mathbb{H}^3$ such that each connected component of the ideal boundary of $X$ is either a round disk or a single point. By known rigidity results, such an embedding is unique up to isometries of $\mathbb{H}^3$. Moreover, we show that the assumption of countably many ends can be dropped if Koebe's circle domain conjecture holds.

math.DG↗

arXiv2026

Minimum Contagious Sets in Trees with Degree-Proportional Thresholds

We investigate how large the minimum size of a contagious set can be among trees of a fixed order. Given a graph $G$ and $ρ\in[0,1]$, an inactive vertex $u$ becomes active when at least $\lceil{ρd_G(u)\rceil}$ of its neighbors are active. An initial set of active vertices is contagious if the process eventually activates every vertex. We obtain exact values for the minimum size of a contagious set in perfect $d$-ary trees. We then determine the exact maximum of this minimum size over all trees of order $n$. We establish this result by bounding the number of pairwise disjoint resistant sets, each of which must intersect every contagious set, and constructing trees that attain the resulting bounds. Finally, by bounding the change in the minimum size of a contagious set under a leaf exchange, we show that, for $0<ρ\leq1$ and $n\geq2$, every integer between $1$ and this maximum occurs as the minimum size of a contagious set in a tree of order $n$.

math.CO↗

arXiv2026

Daylight-Referenced Efficiency (DRE): A Radiometric Measure of Electric Light Source Efficiency

Photopic luminous efficacy (lm/W) weights optical power strictly against human eye sensitivity V (lambda), creating a systemic metrological penalty for electric light sources engineered with broad spectral output in the ultraviolet (280-380 nm) or infrared (780-4000 nm) bands. Similarly, metrics evaluating chromaticity, color quality, or melanopic pathways assess photoreceptor responses strictly within the visual band, offering no measure of broad-spectrum physical radiant power conversion. To resolve this limitation, we introduce Daylight-Referenced Efficiency (eta_DRE), a radiometric framework that evaluates how efficiently an electric light source converts electrical power into an optical spectrum matching natural terrestrial daylight across domain Omega = [280 nm, 4000 nm]. The framework factorizes performance into primary physical quantities: domain-restricted Wall-Plug Efficiency (eta_wall [W_optical/W_elec]), measuring unweighted electrical-to-optical conversion, and Daylight Spectral Fidelity (eta_spec [unitless, 0-1]), quantifying geometric shape match to the ASTM G173 AM1.5G solar reference via total variation distance (1 - TVD). Multiplying total optical power by spectral fidelity yields the absolute Daylight-Referenced Output Power (P_dr [W]), establishing a direct physical flux quantity for lighting specification. Dividing by electrical drive power gives the summary metric Daylight-Referenced Efficiency (eta_DRE = eta_wall * eta_spec [W/W_elec]), measuring net reference-conforming radiant output per electrical Watt consumed. The mathematical architecture is reference-agnostic, domain-extensible, and maintains provable lower-bound monotonicity for eta_wall and eta_DRE under spectral truncation. Open-source reference software is provided to facilitate deployment and standardized rating of broad-spectrum light sources.

physics.optics↗

arXiv2026

Correspondence Is Not Transfer: A Pairing-Controlled Audit of Clinical Audio-Metadata Alignment

Clinical audio models increasingly align recordings with participant metadata or metadata-derived text. This supervision is useful but ambiguous: the paired target can contain symptoms, demographics, missingness, and cohort structure, not just disease information. We study this ambiguity by making the audio-metadata pairing itself the intervention. Our Pairing-Controlled Transfer Audit trains matched contrastive trajectories under correct pairing, global shuffling, disease-label-preserving shuffling, and label-and-recorded-sex-preserving shuffling. Each endpoint is then frozen and tested with three readouts: profile correspondence, attribute decodability, and matched disease prediction. On UKCOVID, correct improves retrieval and sex decodability over label-only shuffling, showing that the pairing intervention worked. The sex-probe contrast is largely absorbed by the sex-preserving control, while smaller retrieval differences remain. Disease-transfer contrasts, however, remain uncertain. After UKCOVID discovery fixed the pipeline, CODA TB, Cambridge COVID-19 Sounds, and Coswara were rerun as a locked external panel with the same arms, backbones, matching, readout, calibration, primary contrast, and primary metric. The external panel preserves the descriptive separation without proving a universal null. Successful profile correspondence and improved disease transfer are therefore different empirical claims.

cs.SD↗

arXiv2026

Model Selection for Off-Policy Evaluation via Pairwise Bellman Validation

Model selection and hyperparameter tuning pose a central challenge in off-policy evaluation (OPE) because errors of fitted Q-functions and occupancy-ratio estimates cannot be directly evaluated from held-out transitions. Bellman-residual and minimax validation methods often require auxiliary regressions or optimization over separate critic classes, adding computational and tuning costs. We introduce pairwise Bellman validation (PBV), a tuning-free minimax procedure for selecting among finite classes of fitted Q-functions and occupancy-ratio estimators using held-out transitions. PBV constructs critics from pairwise candidate differences, requiring only finitely many empirical moment calculations without auxiliary validation models or separate critic optimization. We establish finite-sample oracle inequalities relative to the best available candidate, even under candidate-class misspecification. Our key insight is that Q-function errors and log ratios between candidate and true occupancy ratios provide oracle critics whose Bellman and adjoint Bellman moments control estimation error. Pairwise candidate differences approximate these unknown critics, while penalties account for comparator approximation error. For Q-function selection, we first develop unweighted PBV, which evaluates pairwise Bellman moments under the offline distribution and admits oracle guarantees under a pairwise coercivity condition relating the offline distribution to the target-policy dynamics. We then develop target-aligned PBV (TA-PBV), which uses marginalized importance weighting to provide error control without pairwise coercivity. For occupancy-ratio selection, we develop adjoint PBV (A-PBV), which evaluates adjoint Bellman moments against log-ratio critics. On Hopper and Deep OPE, unweighted PBV performs competitively, while target-occupancy weighting can improve performance in settings with controlled overlap.

stat.ML↗