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 685 records · Page 38Linked to original sources
arXiv2026

Learning Meta-Skills for Agent Harness Design in Test-Time AI4AI

Agent performance depends on both reasoning ability and the environment in which it acts. We study test-time AI-for-AI, asking how a Builder can learn to construct better execution environments for a Target while both models' weights remain fixed. To make the Builder's experience reusable, we introduce Meta-Skill: principles specifying when support is needed and what resources to provide. The Builder learns these principles from Target's execution feedback on the development set, then uses the frozen skill bank to construct harnesses for unseen tasks. Across Harness-Bench and NewtonBench, full-bank meta-skills improve macro-average performance by 8.95 percentage points over no-skill construction, and 12.02 points over direct delivery of the same bank to the Target. These results highlight the value of translating experience into executable support. Gains when the same model serves both roles further suggest a path to system level self-improvement through learning to build better environments.

cs.AI↗
arXiv2026

Classification of topological phases of matter in stochastic systems

Topological phases of matter support edge states protected from noise and perturbations, and the classification of which symmetry and dimension support such phases was developed for quantum systems and analogous platforms. While topological phases have also been discovered in stochastic systems and posited as mechanisms for biochemical processes, a classification consistent with their Markovian constraints remains lacking, impeding generalization to new scenarios. These constraints alter the matrix space preventing the application of previous classification methods, while hosting new properties such as the necessity of non-Hermiticity for non-trivial topological phases. We introduce new methods including a new homotopy approach and redefinition of the point gap and find that only two symmetry classes remain robust: no symmetry and pseudo-Hermiticity. In these symmetry classes, we identify the dimensions with topologically non-trivial phases and their group structure, creating a rigorous framework for predicting robust behavior in active and living matter.

cond-mat.stat-mech↗
arXiv2026

Optimize, Learn, Refine: Whole-Body Grasping and Pick-and-Throw with a Spiral Soft Robot

Soft continuum robots can exploit distributed compliance for whole-body manipulation, but synthesizing behavior through changing contacts remains difficult. We address whole-body grasping and pick-and-throw from an initially ungrasped state through outcome-based actuation-space optimization. Grasping is quantified by tip angular sweep and body-object enclosure, while throwing further incorporates release-direction alignment and minimum release speed. These objectives allow grasping, acceleration, and release to emerge from compliant interaction without prescribing contact forces, contact locations, or body configurations. Because the resulting actuation-to-outcome mapping is nonsmooth, we utilize derivative-free CMA-ES within an optimize-learn-refine framework. CMA-ES generates solutions for sampled conditions, a task-conditioned predictor learns warm starts, and CMA-ES refines them for unseen conditions. In simulation, the method achieves 492/500 successful grasps (98.4%) and success rates of 98%, 97%, and 94% across three directional throwing trials. Learned initialization increases grasping success from 78.6% to 98.4% while reducing the median rollout count from 1184 to 816 in CMA-ES. Hardware experiments achieve a 100% grasping success rate across 50 executions and a 100% pick-and-throw success rate across 30 executions, with 10 repetitions per direction. Together, these simulation and hardware results demonstrate the effectiveness of the proposed framework across both simulated and physical whole-body manipulation tasks.

cs.RO↗
arXiv2026

XENONnT $pp$-Dominated Solar Neutrinos Constrain the $ν_1$ Pseudo-Dirac Splitting

XENONnT has measured a low-energy solar-neutrino signal dominated by $pp$ neutrinos. At these energies, an ultra-small pseudo-Dirac mass-squared splitting induces detectable active-sterile oscillations over the Sun--Earth baseline. We study pseudo-Dirac oscillations of the first mass eigenstate, compute the resulting modification of the $pp+{}^7$Be electron-recoil spectrum, and construct a likelihood from the published XENONnT energy spectra. We find that the XENONnT data exclude $5.9\times10^{-13}\,\mathrm{eV}^2\lesssim δm_1^2 \leq 10^{-11}\,\mathrm{eV}^2$ at the $2σ$ level, extending current solar-neutrino sensitivity to smaller mass splittings and approaching a regime previously discussed for future $pp$-sensitive detectors.

hep-ph↗
arXiv2026

SURE: Framework for Safety to Construct Trustworthy AI

Warning: This paper contains harmful and offensive text. Recently, large language models such as GPT-4, and Claude have revolutionized tasks in various domains. As the use of these large language models increases, people are increasingly concerned about AI safety and demand that large language models behave responsibly and safely. As a result, there has been growing global interest in developing methods to ensure AI safety. However, the detailed criteria for AI safety may vary depending on the country, culture, and policies of the company you serve. In this study, we propose SURE (A Safe and Unified AI Framework foR Everyone), which is designed as a framework for customizing the attributes of AI safety and ensuring the defined AI safety. Within SURE, we establish taxonomies for adversarial prompts that could threaten AI safety and construct prompts based on the taxonomies. We then define templates for desirable AI responses to these prompts and design an absolute safety scoring scheme. Finally, we conduct AI alignment using the datasets to gradually ensure AI safety. The effectiveness of SURE is demonstrated through experiments with various base models.

cs.CR↗
arXiv2026

Zero2Repo: Can Coding Agents Build Repositories from Scratch?

Coding agents are increasingly asked to build software rather than patch it, yet benchmarks for from-scratch repository construction are mostly limited to a single language and depend on manually curated tasks. We introduce Zero2Repo, a benchmark in which an agent receives a product requirements document, an interface contract, and an empty workspace, and must deliver a complete repository in the project's native ecosystem. Tasks are produced by a language-agnostic authoring pipeline that converts real, version-pinned open-source projects into behavioral specifications, reproducible environments, and hidden acceptance tests. Each task is validated by execution: a reference implementation derived from the upstream project must pass, and adversarial validation must show that the tests reject incorrect implementations. Evaluation runs production coding agents in isolated containers, withholds the acceptance tests until an explicit submission, and assigns a binary reward only when every test passes, with no LLM judge. The pipeline and harness make no language-specific assumptions and apply to mainstream programming ecosystems; the current release contains Python, TypeScript, Go, and C++ tasks. Even on 11 tasks drawn from repositories that frontier models have very likely seen during training, the strongest agent solves only 10, and every failing submission passes 90-99% of the hidden tests; for the two strongest agents, 67-100% of failed tests trace to a single omission or a low-frequency rule stated in the specification rather than to a missing subsystem, so each failure is a concrete target for improvement.

cs.SE↗
arXiv2026

Testing Offsets Between Cluster-Scale Halos and BCGs in Strong Lensing Models Using the Jackknife Method

Offsets between cluster-scale dark matter halos and brightest cluster galaxies (BCGs) inferred from strong lensing (SL) models have been used to investigate cluster dynamics and dark matter physics. However, it is necessary to test whether these offsets are required by SL data themselves. We investigate whether allowing these offsets improves predictive accuracy and examine the effects on magnification, time-delay predictions, and critical curve positions. We evaluate lens models of Abell 370, RX J1347.5-1145, and MACS J0416.1-2403 using the jackknife method, which assesses predictions for source systems excluded from the lens reconstruction. Models are constructed with MrMARTIAN, which combines analytic mass profiles with a regularized grid that represents mass structure not captured by the profiles. We compare fixed and free halo position models across regularization weights. The root mean square values averaged over excluded systems agree within bootstrap uncertainties, while the median paired differences remain close to zero across all three clusters and tested regularization weights. These results indicate comparable predictive accuracy between the fixed and free models. The total mass distributions remain broadly similar, although the magnitudes of fitted halo offsets vary with regularization, showing that the mass decomposition is not unique. For these three clusters, offsets between BCGs and cluster-scale halos in the lens models are not required by the SL data to improve predictive accuracy within the MrMARTIAN framework. The mass decomposition between the regularized grid and cluster-scale halos is not uniquely determined. These findings limit physical interpretations based on fitted halo positions alone. The model dependence of magnification and time-delay estimation can also affect the inferred intrinsic properties of high-redshift sources or the Hubble constant.

astro-ph.CO↗
arXiv2026

EVOKE: Eliciting World Knowledge in Agents for Transferable Decision-Making

Large language models (LLMs) are increasingly deployed as agents for multi-step decision-making, yet transfer poorly to unseen environments. World-model methods address this by training agents to predict future observations, at the cost of additional training and errors that compound when predictions are used for planning. However, for LLM agents operating in digital environments, much of this world knowledge is already internalized during pretraining, which shifts the problem from acquiring it to eliciting it. We argue that typical post-training provides little pressure for such elicitation, since supervision under a single goal at each visited state inadvertently drives policies to rely on superficial contextual habits. We introduce EVOKE, a post-training method that supplies this pressure through goal diversity at fixed states. Motivated by theory showing that an agent competent across diverse goals must encode a world model recoverable from its action preferences, EVOKE holds the environment state and interaction history fixed and ranks the same candidate actions under alternative goals, forcing action preferences to change, so that a policy relying on contextual habits or single-goal correlations cannot order them correctly. This implicitly elicits the policy's pretrained world knowledge to inform decisions. We evaluate EVOKE across diverse tasks in three backbones, demonstrating improved task performance, unseen environment generalization, and data efficiency. We further conduct controlled analyses to better understand what drives these gains. These findings offer a new perspective on eliciting internalized world knowledge for transferable action through direct decision supervision.

cs.CL↗
arXiv2026

TAGGRAPH: Tag-Augmented Graphs for Graph Retrieval of Agent Persistent Histories

Long-term memory lets LLM agents recall past interactions and remain consistent across sessions, but memory systems are hard to compare because they often vary in representation, indexing, retrieval, and evaluation. We present a controlled evaluation framework based on shared 5W-style conversational memories. Localized graph configurations traverse a common base graph; AdaptiveGraph adds chronological edges and Personalized PageRank diffusion. We also evaluate BM25 over the same extracted notes and OpenClaw as a raw-input external reference. Retrieval rankings vary across memory settings. On LongMemEval-S, AdaptiveGraph is the strongest graph configuration at 0.844 MRR, but BM25 reaches 0.867 and OpenClaw 0.880. On ATANT Core, localized graph traversal outperforms diffusion and BM25, whereas BM25 leads the stress rounds. Reducing LongMemEval-S within the tested range does not reproduce the ATANT diffusion penalty, but the smallest tested store remains larger than ATANT Core, so store size cannot be ruled out. The penalty also persists under a permissive content-match criterion. Vocabulary normalization and extraction quality substantially affect graph retrieval, and missing extraction tags are common among top-five misses. Retrieval strategies should therefore be evaluated jointly with the memory setting and against strong lexical baselines.

cs.IR↗
arXiv2026

Halluscoring 2026: The first shared task on llms hallucination detection and answer verification

We present HalluScoring 2026, a shared task for evaluating hallucination detection and factual verification in Arabic question answering under challenging generalization settings. Its four subtasks are organized into two tasks. Task~1 evaluates binary hallucination detection, considering generalization to unseen questions (Subtask 1.1) and responses generated by unseen LLMs (Subtask 1.2). Task 2 extends the evaluation beyond detection by requiring the systems to additionally identify the correct factual answer from six related candidates, covering Islamic knowledge (Subtask 2.1) and general knowledge (Subtask 2.2). The shared task is based on two Arabic datasets: HalluScore and HalluTruthQA. A total of 13 teams participated in the shared task, ten of which submitted system description papers. The results of Task 1 demonstrate that hallucination detection remains challenging. On the Task 1 test sets, the top-ranked systems achieved AUC-ROC scores of 0.7717 for Subtask 1.1 (REGLAT) and 0.7670 for Subtask 1.2 (NAMAA). Under assisted evaluation, the highest combined detection and answer-selection scores for Subtasks 2.1 and 2.2 were 0.8824 and 0.8565, respectively.

cs.CL↗
arXiv2026

Geometry and Mechanics of Ribbon Gridshells

Mechanical metamaterials exhibit anomalous properties induced from non-trivial mesoscopic constituents. Inspired from architectural and industrial structures, we introduce arrangements of long, narrow ribbons intersecting at prescribed angles as a model thin-sheet metamaterial. These ribbon gridshells are shown to display highly non-linear behavior driven by the geometric constraints, nonetheless unlike most complex mechanical systems, we are able to explicitly write out the coarse-grained governing equations and to classify their solutions in terms of the intersection patterns of the ribbons. It is shown that the structure can assume arbitrary, tunable Gaussian curvature distributions, allowing us to formulate an inverse design problem, solvable under a suitably defined local condition. An analysis of the soft modes, confirmed by experiment and numerics, reveals rich mechanics which may exhibit both a rigid and an anomalously soft behaviors.

cond-mat.soft↗
arXiv2026

MOBA-VL: Event-Localized Multi-Turn Reinforcement Learning for Real-Time MOBA Commentary

Real-time commentary for Multiplayer Online Battle Arena (MOBA) esports requires a vision-language model (VLM) to narrate a live match second by second, both fluently and accurately. Existing streaming VLMs sound natural but often miss key events such as kills and objectives. To address this limitation, we use game telemetry, which records exactly when each event occurs, as a supervision signal. We introduce MOBA-VL, a 9B-parameter model trained on this signal with event-localized multi-turn reinforcement learning, which rewards the turns that describe each event. We also collect MOBACast, 860 professional matches (about 460 hours) across three MOBA games with word-level timestamped commentary, and MOBACast-Bench, a benchmark from held-out tournaments. On MOBACast-Bench, MOBA-VL achieves the highest Overall score on full matches (63.25 vs. 55.12 for StreamingVLM) and clips (63.45 vs. 56.22 for DeepSeek-V4.1-Flash). Event-localized credit also raises event recall from 34.5 to 42.1 over supervised fine-tuning. Code and data will be released, and demos are available on an anonymous project page at https://moba-vl.github.io.

cs.CV↗
arXiv2026

On c-polynomial factorisations

c-Polynomials are polynomials whose (non-zero) coefficients are all equal to 1. Taking a theorem by Carlitz and Moser as a motivation, we prove a structure theorem for the factorisations of the c-polynomial $(x^n-1)/(x-1)$ into c-irreducible factors by relating c-torisations into c-polynomials to joint ordered factorisations arising from the integer n. The question of how many different joint ordered factorisations there are leads to the precise chromatic polynomial for path graphs, counting the ways of path graph colourings with $m$ colours under the condition that all colours are used. Moreover, we give similar counting formulae for joint ordered factorisations under the constraints that all factors are primes, or that all factors are square-free.

math.CO↗
arXiv2026

Retargeting Motions to Diverse Skeletons via Learnable Flattening

Cross-structural motion retargeting aims to transfer motion between different skeletal topologies. Despite recent progress, existing state-of-the-art models struggle with reliability in zero-shot settings, i.e. skeletons with different topologies which were unseen during training, and recent Transformer-based attempts have failed to outperform specialized geometric methods. We bridge this gap with a Transformer Autoencoder that learns a topology- and translation-invariant latent space. Our core contribution is a learnable flattening of skeletal graphs that captures both local dependencies and global structure. Unlike the standard transformer architecture, which adds positional information to token content, we integrate graph-based positional encodings multiplicatively, a design choice that follows directly from our flattening formulation. The resulting model handles diverse skeletal topologies within a single unified architecture and trains in a fully unsupervised manner, requiring no paired retargeting data. Ablation studies show, that the graph encodings, multiplicative formulation, and Transformer backbone is critical for the performance. In zero-shot evaluations, our method reduces global joint position error by $43-47\%$ over current benchmarks. A user study ($n = 37$), including expert animators, further ranks our approach highest in motion alignment and physical plausibility ($p < 0.05$). These results demonstrate that our model design is key to making transformer architectures effective for motion retargeting, outperforming existing approaches.

cs.CV↗
arXiv2026

Fraud Detection via Bayesian Positive-Unlabeled Learning with Gaussian Processes and Multilayer Networks

Tax fraud remains a central challenge for public revenue authorities worldwide, imposing fiscal losses estimated to reach up to 1 trillion euros annually in the EU alone. Fraudulent firms intentionally manipulate reported figures to conceal their activity. Business networks can provide complementary information and reveal fraud that covariates alone may miss. We propose a Bayesian positive-unlabeled (PU) classification framework that combines firm-level covariates with multilayer network information. As a policy relevant case study, we apply the framework to firms in a regulated segment of the Greek energy market, a sector exposed to excise tax evasion. Confirmed fraud labels exist only for audited cases, while the remaining observations are unlabeled rather than verified compliant. We develop a Bayesian Gaussian process (GP) classifier integrating covariates with multilayer network information through a Product of Experts (PoE) construction and incorporating a nondetection probability for missed positives. We derive identification bounds for the nondetection rate under an anchor condition and show that, for a fixed latent risk function, a common nondetection rate affects calibration but not ranking. Simulations show strong ranking performance relative to PU and network-based alternatives, while illustrating the difficulty of estimating the nondetection rate. In real audit data, the method identifies high risk firms with posterior uncertainty, estimates the number of undetected fraudulent cases, and identifies whether risk is driven by covariates, networks, or both. Of the six highest ranked firms, the two that had already been reinspected independently were both confirmed as noncompliant, providing an independent validation of these cases. The remaining four were selected by the tax authority for follow-up inspection.

stat.ME↗
arXiv2026

StreamDecisionBench: Evaluating Decisions in Force on Evolving Language Streams

Language models increasingly make real-time decisions in applications that apply the latest answer until a newer one arrives. A late answer can prolong an outdated decision, such as a call recorder still running while a customer reads out card details, an error offline accuracy misses. We make three contributions. First, we release StreamDecisionBench (SDB), a dataset of eight streaming scenarios in four application families, with executable reference decisions derived from public rules. Second, we propose an evaluation protocol and a metric, in-force accuracy: the share of time the applied decision is correct across update intervals of 0.5-8 s. It reflects accuracy and latency jointly, attributing each error to judgment, latency or both. Third, we evaluate thirteen single-model settings, and this attribution separates speed-limited from judgment-limited models: slower, more accurate models lose 42-51% of the time to outdated answers, a fast model 34% to wrong ones. We therefore test hybrids in which a slow model corrects a fast one; with the right pairing and configuration, a hybrid outperforms every single model. However, even the best evaluated system keeps a correct decision in force only about two-thirds of the time, leaving a substantial gap for real-time use.

cs.CL↗
arXiv2026

Marking Contour Tones in Yorùbá: A Typographic and Computational Proposal

Yorùbá is a tonal language in which contour tones pose persistent orthographic challenges. These are especially notable for personal names and lexical items whose conventional spellings avoid vowel lengthening that would otherwise provide a host syllable for the second tone. A particular concern is a class of names in which the conventional spelling does not just omit tonal information but inverts the meaning of said name, sometimes asserting the opposite of what the name intends. This paper describes the problem, illustrates the inadequacy of current solutions, and proposes the adoption of the caron and circumflex marks. These are symbols with precedent in Yorùbá phonological scholarship since Olmsted (1951), used as orthographic conventions on single vowels to encode rising and falling contour tones, making them accessible for the first time through standard keyboard input and computational text processing. The proposal is supported by an implementation in the WriteYoruba keyboard and the TTSYoruba speech synthesizer, whose architecture and listener evaluation are reported separately (Tubosun et al., 2026).

cs.CL↗
arXiv2026

Alignment via Training Against Probes Without Losing Monitorability

Models are usually aligned based on their observed outputs, using demonstrations, preference data, or reward signals. These objectives reward responses that look aligned. More capable models may learn to satisfy them without internalizing the intended behavior, for example by faking compliance during training. Such superficial compliance could be harder when the objective is defined on model internals rather than outputs. Therefore, we study probe-guided fine-tuning, using probes that detect undesired properties in model activations as a direct training signal. We evaluate linear and non-linear probes with different numbers of probes per layer across two alignment objectives: harmlessness and honesty. We find that training against probes that do not update during training is an easily exploitable objective, while continuously updated probes substantially reduce harmfulness and improve honesty while preserving utility. Probe-guided fine-tuning achieves better safety-utility trade-offs than DPO and inference-time steering, while being substantially more robust against jailbreak and abliteration attacks. Moreover, the concepts stay linearly encoded after fine-tuning, meaning oversight is not lost by our method. Training against probes thus offers a way to shape what models represent rather than only what they output, which may become increasingly important as models get better at making their outputs look aligned.

cs.LG↗