跳到正文
研踪PaperTrail

论文库

检索计算机领域论文,按研究方向、作者或关键词筛选。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+197新增论文
+151其中 AI 相关
+16已有论文修订
+91新增 AI 导读

arXiv 最近完成扫描:2026-11-22 06:41 · 全站最近入库:2026-11-21 12:19各信源进度 →

统计口径与数据新鲜度

统计 2026-11-21 08:18 — 2026-11-22 08:18(北京时间) · 点击数字查看对应简报;AI 相关包含在新增论文内。

最近完成检查:2026-11-22 06:41最近发现新增版本:2026-11-21 12:19库内论文最新更新时间:2026-11-20 02:59

每分钟刷新已保存的数据。空检查不代表采集异常;新收录可能包含历史论文。AI 导读按新增版本计数。

29,158 已收录论文24,181 AI 相关8,992 有修订112 HF 社区发现

浏览论文库

时间范围跟随排序:版本更新、本站收录或首次发表。最新收录可能包含历史论文。部分信源异常,查看状态 →

找到 21,490 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-22 06:39(北京时间)
版本更新 · v2arXiv:2605.06481

OA-WAM: Object-Addressable World Action Model for Robust Robot Manipulation

Yushan Liu, Peibo Sun, Shoujie Li, Yifan Xie, Lingfeng Zhang, Xintao Chao, Shiyuan Dong, Fang Chen 等 10 位作者

原文摘要

World action models (WAMs) learn manipulation by predicting actions and future observations, yet their success drops sharply when the camera viewpoint, object layout or robot initial state changes. Such changes keep the instructed object the same but alter the appearance and pose features the policy observes, which in standard attention also form the keys that locate the object. We introduce OA-WAM, an object-addressable WAM that finds each object by an identity address fixed for the episode and reads its changing state. In every layer of a pretrained multimodal trunk, object-slot keys depend only on the address and frame position, and values carry the current state; the same addresses key the action pointer and index per-object forecasts, which can trigger re-grounding. Fine-tuned on standard LIBERO demonstrations, OA-WAM transfers zero-shot to LIBERO-Plus with 90.5% success, outperforming all ten published baselines. It also leads on LIBERO (99.0%) and, fine-tuned on BridgeData V2, on SimplerEnv WidowX (88.9%). On five real-robot tasks, it succeeds in 89 of 100 nominal trials, versus 73 for the strongest baseline. Ablations trace its robustness to addressing rather than to object slots: with standard attention, slots do no better than a slot-free policy, and addressing them by identity raises geometric success by 10.7 points. Forecasting the addressed objects is likewise a stronger training signal than forecasting image codes.

cs.RO

新论文 · v1arXiv:2611.20137

Cleanly-DINO: Efficiently Learning Visual Task Spaces for Robotics

Adrian Röfer, Melika Shekarriz, José Arce y de la Borbolla, Iman Nematollahi, Abhinav Valada

原文摘要

Learning manipulation tasks from few demonstrations requires abstractions that capture the changes relevant to task completion, such as object poses in rigid-body manipulation. However, tasks such as cleaning are defined by changes in visual state rather than object pose, making the choice of task representation less clear. In this work, we propose Cleanly-DINO to learn visual task spaces from few demonstrations. Given a mask of a task-relevant region and frozen ViT features, Cleanly-DINO fits a projection to isolate a task space and fit a progression metric to it. We train the model on diverse cleaning examples and evaluate its ability to localize incomplete task regions and transfer across visual domains. We also study its utility for robot learning by using it as an observation encoder and reward signal in a reinforcement learning setting. Finally, we deploy our fitted model within a real-world robotic wiping pipeline to evaluate its use in guiding task execution.

cs.RO · cs.LG

版本更新 · v2arXiv:2608.26739

Simulation-Based Evaluation of Residual DDPG-Augmented Computed Torque Control for a Cable-Driven Lower-Limb Rehabilitation Robot Model under Disturbances and Parametric Uncertainties

Mohammad-Hossein Fakouri, Ali Keymasi-Khalaji

原文摘要

Nominal computed torque control (CTC) relies on accurate inverse dynamics; parameter mismatch and disturbances can leave tracking error, while joint limits and unilateral cable forces constrain motion. This numerical study evaluates a bounded Deep Deterministic Policy Gradient (DDPG) residual correction around nominal CTC for a planar three-degree-of-freedom, three-cable lower-limb rehabilitation robot model. In four simulation cases, we examine tracking, joint limits, workspace, cable-Jacobian conditioning and inferred cable demands. In a representative combined uncertainty/disturbance case, Cartesian RMS error was 0.0221419 m with CTC and 0.0128462 m with residual DDPG-CTC (approximately 42% lower). On a different circular reference, the residual branch nevertheless increased joint-limit-violating channel samples from 44 to 101. Separate supplemental experiments analyze training variability and checkpoints, alternative compensators, reward/action sensitivity, independently varied parameters and disturbances, trajectory transfer, constrained cable-force allocation and force lag. These studies show both improvements and failures, including early terminations and force-capacity-dependent degradation. The verified original algebraic force-backprojection path and later bounded-force simulations are distinguished; numerical allocations do not establish physical pretension or motor performance. Simscape and MSC ADAMS checks are numerical cross-model consistency assessments. No hardware, hardware-in-the-loop or patient tests were performed, and no clinical efficacy or physical safety guarantee is claimed.

cs.RO · eess.SY

版本更新 · v3arXiv:2607.12995

One Shot, Twenty-One Balls: A Computer-Assisted Proof of a Single-Stroke Total Clearance in an Idealized Model of Snooker

Avner Kantor

原文摘要

Snooker folklore holds that no single stroke can pocket all twenty-one object balls. We examine the claim in an explicit model: twenty-two equal disks on a rectangle with six pocket points, under constant rolling deceleration, frictionless collisions with normal restitution, and lossy specular cushions. We exhibit a configuration and a stroke, and prove in interval arithmetic that every stroke in an explicit box around it pockets all twenty-one object balls, so the set of successful strokes has positive measure. That set is a ribbon: every speed from 9.3214 to 12 m/s works, while the certified window of directions is 9.481e-43 radians wide, a number explained by a first-order analysis of twenty near-grazing collisions. A second configuration, a near-frozen cluster, has a certified window 9.4e-4 radians wide, thirty-nine orders of magnitude wider; its certificate uses Taylor models of degree six. The strokes with a given list of events form a semialgebraic set, and the endpoints of the windows are algebraic points of the circle. For the regulation opening position we state the corresponding conjecture, and show that if it holds with a regular witness then a finite certified search establishes it.

cs.CG · math.DS · nlin.CD

版本更新 · v2arXiv:2603.29921

Quantale-Enriched Co-Design: Toward a Framework for Quantitative Heterogeneous System Design

Hans Riess, Yujun Huang, Matthew Klawonn, Gioele Zardini, Matthew Hale

原文摘要

Monotone co-design enables compositional engineering design by modeling components as feasibility relations between required resources and provided functionalities. However, its boolean formulation cannot natively represent quantitative criteria such as cost, confidence, or implementation choice. In practice, these considerations are introduced through ad-hoc scalarization or by augmenting the resource space, which obscures system structure and increases computational complexity. Therefore, we develop a quantale-enriched theory of co-design in which resources and functionalities are quantale-enriched categories and design problems are quantale-enriched profunctors, which lifts co-design from boolean feasibility to quantitative evaluation. Series, parallel, and feedback composition remain valid over any commutative quantale, and change-of-base maps between quantales yield heterogeneous composition, so that subsystems evaluated in different local semantics can be composed in a common framework. Numerical examples on a target-tracking system and a UAV delivery problem demonstrate this framework and show how it avoids the drawbacks of boolean-only formulations.

eess.SY · cs.LO · math.CT · math.OC

版本更新 · v2arXiv:2607.05626

Population-Level Profiling of DSM-5 Depressive Symptom Language Among Twitter Users With Self-Reported ADHD or ASD: Exploratory Study Using Natural Language Processing and Statistical Analysis

Muhammad Rizwan, David Nabergoj, Jure Demšar

原文摘要

Depression frequently co-occurs with attention-deficit/hyperactivity disorder (ADHD) and autism spectrum disorder (ASD), but how depressive symptoms are expressed by people with these conditions is underexplored. We examined whether Twitter users self-reporting ADHD or ASD differ in language related to the nine DSM-5 depressive symptoms. We analysed 1,282,437 tweets (2020-2021) from 792 users. Tweets were screened for depressive relevance by a zero-shot classifier combined with a locally run large language model (Qwen3-8B) and classified into the nine symptoms by MentalRoBERTa fine-tuned on the ReDSM5 corpus. The pipeline was tuned on two stratified samples of 500 tweets and internally validated against human-adjudicated labels on 150 of these tweets. User-level profiles (582 users) were compared with L1-penalised logistic regression, nested cross-validation, bootstrap stability analysis, per-symptom effect sizes, and co-occurrence analysis. A baseline pipeline with a natural language inference filter reached a precision of only 0.09-0.13, whereas the final pipeline reached precision 0.48-0.59 and F1 0.55-0.68. Discrimination between groups was modest (nested ROC-AUC 0.60-0.62). Language related to suicidal ideation was more frequent among ASD users and language related to cognitive issues among ADHD users; both differences were small but robust across sensitivity analyses. These exploratory findings describe how depressive symptoms are expressed online, not their clinical prevalence.

cs.CL

新论文 · v1arXiv:2611.20129

Can decision models extract clinical data from cancer pathology reports? An evaluation of accuracy, calibration, and failure modes

Indranil Mallick, Paromita Roy

原文摘要

Decision models are a new class of language model that, instead of generating text, choose one option per question from a fixed list and return a probability for every option. They have not yet been evaluated on clinical text. We evaluated two open-weights decision models, Nimble 9B and Tev1 4B, for extracting 25 fields from 1,113 oral cavity cancer pathology reports from a single centre, against a prospectively maintained clinical database. We compared them with two generative models: Qwen3.5 9B, run locally, and GLM-5.3-flash, run through a cloud API. Tev1 reached a mean field accuracy of 0.917 (95% CI 0.913 to 0.920), the same as Qwen3.5 9B (0.918; p = 0.26), and was four times faster (15 s against 61 s per report). Nimble reached 0.862 and GLM-5.3-flash 0.986. Both decision models matched GLM-5.3-flash on simple dichotomous categories. They were weak on measurements binned into categories, on fields whose correct answer was not reported, and on some complex decisions (e.g. involved margins). We found that the main cause of the inferior performance of Nimble's came from asking 25 questions in one request: with two questions per request, its accuracy improved. Pooled over fields, the returned probabilities were well calibrated (expected calibration error 0.022 and 0.026), but they did not flag systematic errors within single fields. A simulated strategy of sending answers with a probability below 0.94 to GLM-5.3-flash raised Tev1's accuracy to 0.982, with 38% of answers escalated. Decision models can extract explicitly stated findings from pathology reports quickly on local hardware and return useful probabilities. They should be validated in the configuration in which they will be used, restricted to the fields in which they are reliable, and combined with other methods for numeric values.

cs.CL · cs.AI · cs.LG

新论文 · v1arXiv:2611.20127

Aerial Person Detection under Edge Constraints: Held-Out Evaluation and Edge-Ground Verification

Raghu Venkat, Tricha Anjali

原文摘要

A search drone sees a missing person as a few pixels, carries a small computer, and shares a narrow radio link. We study person detection under those limits; the primary evaluations were declared before they ran, and later analyses are marked exploratory. In an exploratory comparison, holding out recording blocks inferred from thermal frame identifiers cut recall at a fixed false-alarm target from 32.7% to 12.3%. With nothing tuned on test blocks, a 46,000-parameter detector gated by a contrast cue finds 10.4% of people against 5.7% for the same network applied everywhere. A standard detector, YOLO11n, trained under the same folds finds 54.5% at a similar false-alarm rate for about eleven times the arithmetic, and on our 4K field footage at 117 m has a walking person among its top 16 detections in up to 92% of frames against 74% for our board. Run alone, YOLO11n's thresholds set at one location overshot their false-alarm target 1.7 to 3.4 times at new locations, and its best configuration at altitude cost 4 to 24 standard passes per frame. Re-scoring its 16 best crops (154 KB per frame) with an 842,000-parameter ground model, thresholds frozen on a validation location, raised recall from 44.9% to 53.7% against the same 16 crops with YOLO's own scores; in a post hoc paired analysis over 15 test locations the gain was +8.8 points [+2.4, +12.5]. Pooled false detections fell from 0.86 to 0.64 per frame, but that paired change was not distinguishable from zero. Coverage of a still person in earth tones was very low for both detectors in the clips we flew.

cs.CV · cs.RO

新论文 · v1arXiv:2611.20126

A Corpus-Driven SentencePiece Unigram Tokenizer for Sindhi: Design and Benchmark Against Fourteen Existing Tokenizers

Ali Nawaz, Muqadas Fareed, Soonh Sadaf

原文摘要

Sindhi is a low-resource Indo-Aryan language written mainly in an extended Perso-Arabic script. Several Sindhi tokenizers have been released on public model hubs, but they have not been compared with one another on shared data. We train SentencePiece Unigram tokenizers with byte fallback at four vocabulary sizes (32k to 150k) on about 1.10 million Sindhi sentences and a smaller, repeated Urdu, Persian and Arabic slice, from four public sources, and compare them with fourteen existing tokenizers on four evaluation sets: FLORES+, Sindhi Wikipedia, a news-derived sentiment corpus, and a contemporary news corpus. The metrics are fertility, single-token retention, UNK rate, byte-fallback rate, and round-trip character loss. At 64k, our tokenizer averages 1.147 tokens per word, against 1.208 for the best existing tokenizer and 1.512 for XLM-RoBERTa, and keeps 88.6% of words as single tokens. On fertility, single-token retention and UNK rate it is best or tied for best in all four evaluation sets at 64k and above; at 32k it trails one baseline on FLORES+. Margins over the best Sindhi-adapted baselines are small (0.002 fertility on FLORES+ at 64k), and we report no significance tests. A round-trip check finds no character loss for our tokenizer and silent character deletion (up to 0.85% of characters) in two existing ones; seven others lose less than 0.1%. We evaluate intrinsic metrics only, so effects on downstream tasks remain untested. Code and tokenizer files are public (DOI: 10.5281/zenodo.23962474).

cs.CL

新论文 · v1arXiv:2611.20124

QA-CMIF:Accurate and Lightweight Cross-Modal Latent Imputation and Fusion for Fault-Tolerant sEMG-Inertial Hand Gesture Recognition

Xinyu Tao, Shuang Huang, Yeyun Cai, Fang Deng

原文摘要

Fusing surface electromyography (sEMG) with inertial measurements is widely used for myoelectric prosthesis control and wearable human-machine interfaces. In daily wear, however, socket micro-displacement, electrode lift-off, and wire failures often corrupt an unknown subset of sEMG channels through burst artifacts, lift-off shocks, or open-circuit dropouts. Existing fusion networks mix all channels in the first convolutional layer, so a few corrupted electrodes contaminate the entire latent representation, whereas gate-only attenuation of corrupted channels discards the co-activation cues needed to separate fine-grained gestures. We propose QA-CMIF, a quality-aware cross-modal latent imputation and fusion framework that encodes each sEMG channel independently, scores its reliability with a soft gate, and protects intact channels via a confidence-weighted residual bypass, while a cross-attention imputer repairs degraded channels in latent space using the inertial stream and remaining reliable sEMG channels under a bad-channel alignment loss. Across within-subject evaluations on 29 participants from Ninapro DB3, DB2, and DB5 spanning 17-52 gesture classes, QA-CMIF outperformed the gate-only baseline under 50% channel corruption by 1.29-2.65 percentage points, denoted pp, across all three 17-class cohorts with paired-t Holm-adjusted p<0.002, widened this margin to 2.59-4.15 pp on the 40- and 52-class tasks, and surpassed corruption-augmented conventional fusion by 4.61-19.85 pp. Requiring only 0.175-0.185M parameters and 2.45-2.63M multiply-accumulate operations across the 2000 Hz and 200 Hz setups, QA-CMIF saves 46%-55% in parameters and 72%-81% in compute over recent baselines with a 1.44 ms 99th-percentile single-thread CPU latency.

cs.LG

版本更新 · v2arXiv:2602.23363

MediX-R1: Open Ended Medical Reinforcement Learning

Sahal Shaji Mullappilly, Mohammed Irfan Kurpath, Jaseel Muhammad, Omair Mohamed, Mohamed Zidan, Fahad Khan, Salman Khan, Rao Anwer 等 9 位作者

原文摘要

In this work, we introduce MediX-R1, an open-ended Reinforcement Learning (RL) framework for medical vision language models (VLMs) that enables free-form answers beyond multiple-choice formats. MediX-R1 fine-tunes a baseline vision-language backbone with Group Based RL and a composite reward tailored for medical reasoning: an LLM-based accuracy reward that judges semantic correctness with a strict YES/NO decision, a medical embedding-based semantic reward to capture paraphrases and terminology variants, and lightweight format and modality rewards that enforce interpretable reasoning and modality recognition. This multi-signal design provides stable, informative feedback for open-ended outputs where traditional verifiable or MCQ-only rewards fall short. To evaluate performance, we employ a unified evaluation framework for both text-only and multimodal (image+text) tasks that uses a Reference-based LLM-as-judge in place of brittle string-overlap metrics, capturing semantic correctness, clinical accuracy, and contextual alignment. Our results show that open-ended RL with comprehensive reward signals can effectively strengthen medical reasoning in multimodal models. Despite using only $\sim51$K instruction examples, MediX-R1 achieves strong performance across standard medical LLM (text-only) and VLM (image + text) benchmarks, outperforming existing open-weight medical VLMs, with large gains particularly on open-ended clinical tasks. These improvements remain consistent across multiple RL algorithms and VLM backbones and are further supported by blinded evaluations with medical experts. Our trained models, curated datasets and source code are available at https://medix.cvmbzuai.com

cs.CV

版本更新 · v2arXiv:2610.41263

Yesterday's Code, Today's Dependencies: How Dependency Selection Shapes PHP Migration Benchmarks

Afridi Mahmud, Md. As-Aid Rahman, Rifat Shahriyar

原文摘要

Migration benchmarks need projects whose tests pass before migration begins. Yet an old project can behave differently without any change to its code. Reinstalling it today may introduce newer libraries, changing both its initial test results and the failures observed during migration. We investigate this effect in PHP 7.4-to-8.3 migration. Using unchanged project source and saved package records, we compare two dependency policies. Modern selection allows all versions in the snapshot, while release-date selection allows only versions published by the project's release date. With Composer 2 fixed, release-date selection improves the PHP 7.4 pass rate by 7.1 pp across 3,660 releases. A separate comparison against committed locks shows that it also recovers recorded dependency versions more closely than modern selection. The choice matters after migration as well. Among projects that pass on PHP 7.4 under both policies, PHP 8.3 test outcomes differ in 15.0% of eligible pairs. We use a shared test runner and bypass Composer's platform check to isolate test behavior. The resulting candidate migration task sets overlap by only 58.5%. Most differences arise in Laravel projects, where older dependencies trigger deprecations that newer versions avoid. Targeted interventions show that removing these initial failures can restore passing tests or expose further incompatibilities. Thus, dependency selection changes not just whether an old project passes, but also which migration failures a benchmark can observe. Preserved locks provide the strongest available version reference; without them, release-date selection offers a historically grounded alternative. Migration benchmarks should make this choice explicit and report its effect on their results.

cs.SE

新论文 · v1arXiv:2611.20121

Matching two-dimensional statistics does not certify three-dimensional transport in battery electrodes

Maziar Raissi, Devesh Upadhyay

原文摘要

Three-dimensional battery electrode structures reconstructed from two-dimensional micrographs are routinely used to predict transport, assuming that matching the measured 2D statistics reproduces the relevant 3D properties. We test this premise on X-ray tomograms of three all-solid-state cathodes. With a differentiable conductivity solver, we optimise voxel structures that match the real electrode in phase fractions, boundary density, two- and three-point correlations and lineal-path functions, on all section planes, slab by slab and layer by layer, within measurement noise. Every sample has statistical twins conducting 10-88-fold less than the real electrode, robust to 5^3 majority filtering, whereas the most conductive are at most 3.3-fold above it. The twins rearrange rather than remove material, severing or funnelling through-plane paths. A classifier on 2D sections detects them, as it does current 2D-to-3D generators: the information is in the images but not in the statistics. Matching standard 2D statistics therefore does not certify 3D transport.

cs.CE · cond-mat.mtrl-sci

版本更新 · v2arXiv:2611.13015

From Questions to Decisions: Benchmarking How LLM Agents Acquire Decision-Relevant Information from Human Beings

Yuhan Tang, Ke Fan, Xinkai Zou, Baichuan Mo, Kangxin Cui, Ao Qu, Ou Zheng, Jiangbo Yu 等 12 位作者

原文摘要

LLM agents are increasingly asked to make plans for people, such as scheduling drivers or assigning nurses. However, people's requests are often incomplete, and the supporting information for decision-making may be outdated, requiring agents to ask for clarification. Existing benchmarks typically provide fully specified problems or rely on uncalibrated LLM-simulated users, leaving this ability insufficiently tested. We introduce Ask2Decide, a benchmark of 43 planning tasks that evaluates whether agents ask for missing information that matters for their decisions. An operator holds facts absent from the agent's request and records, and optimization solvers determine which facts change the optimal plan. We also include similarly worded facts that leave the optimal plan unchanged to test whether agents distinguish useful information from irrelevant details. We collect 79 interaction sessions from 19 participants and find that, after an incorrect plan, human replies omit the correcting fact far more often than replies from an uncalibrated simulated user (43% versus 7%). Using these interactions, we calibrate a simulated operator that more closely matches human reply patterns. Evaluating four frontier LLMs, we find that their initial plans are infeasible on 58-98% of tasks when information is missing, compared with 5-28% when all facts are provided upfront. Agents uncover facts that change the optimal plan only slightly more often than facts that leave it unchanged (39-52% versus 32-41%). These results show that obtaining decision-relevant information through interaction remains a major bottleneck for LLM planning. This is ongoing work and will be updated. The date of this version is shown on the first page.

cs.AI

版本更新 · v2arXiv:2606.13934

Adversarial Concept Search: Predicting Compositional Errors From Feature Geometry

Jennifer Meng Lu, Ruochen Zhang, Isabelle Lee, David Alvarez-Melis, Ellie Pavlick, Naomi Saphra

原文摘要

Humans cannot always intuit what scenarios are most challenging to LLMs. Hoping to capture challenging edge cases, developers either design problems to be difficult for humans or curate extensive benchmarks. What if we could instead anticipate which scenarios a model will fail on? In this paper, we use an LLM's representational geometry to predict which concept combinations it will fail on. We attribute this compositional failure to interference between salient features. In tasks that require systematic composition - toy programmatic settings, multihop reasoning, multilingual factual recall - we find that when a pair of concepts is encoded near-orthogonally, the model reliably composes them. When their linear encodings are close, producing interference, the model fails to compose them. Our method reliably anticipates failure modes across different compositional tasks, without evaluating specific inputs. These results lay the groundwork to use representational geometry to identify high-risk examples, construct targeted stress tests, and provide a scalable foundation for active learning in real-world deployment.

cs.AI

新论文 · v1arXiv:2611.20119

CECAR: Cache & Expert Co-Aware Routing Accelerates On-Device Inference of MoE LLMs

Byeongju Kim, Hoonki Lee, Byungjun Kim, Inyul Ra, Yongjin Kim, Yongseong Lee, Sangyeob Kim

原文摘要

Despite growing interest in on-device LLM deployment, large-scale Mixture-of-Experts (MoE) models remain impractical on resource-constrained devices, as sparse expert activation still requires all expert weights to be memory-resident. To address this, we identify three key observations for on-device MoE inference: (a) cache hit rates are fundamentally bounded even under Belady's optimal policy; (b) these bounds are insufficient for low-latency on-device inference; and (c) prior cache-aware routing based on binary cache presence can induce routing instability. Based on these insights, we propose CECAR (Cache & Expert Co-Aware Routing), a MoE inference system that jointly considers expert routing and cache management. Unlike prior cache-aware routing methods, CECAR uses an ML-based cache policy to predict the expected reuse distance of each expert and prioritizes routing and execution toward experts with smaller predicted reuse distances. On the Qwen3-30B-A3B model, CECAR achieves 12.5 tokens/s on a consumer-grade GPU with only 16 GB of VRAM, where non-resident experts are fetched from SSD, delivering a 4.5x speedup over conventional MoE inference and exceeding the decode speed of baselines that keep the entire model fully resident in DRAM. Code is available at https://github.com/hoonki2002/CECAR.

cs.LG · cs.AI

版本更新 · v2arXiv:2606.25787

Cited Web Sources in LLM Answers about Brands: A Multilingual Audit

Dmitrij Żatuchin

原文摘要

Displayed citations make part of generative search observable, but they do not reveal its complete retrieval process. This audit covers three archived datasets containing 167,551 URL-bearing citation entries across 128 brands, 13 languages and 12 home markets. In the principal Nordic-Baltic component, 3,863 of 21,033 registrable domains account for 80% of 131,514 citations. Wikipedia leads in 11 of 12 query languages. A recovered, separate brand-associated website classification retains 131,512 entries: 108,410 (82.4%) fall outside its brand-associated category and 23,102 (17.6%) inside. This classification does not establish legal ownership. In the full 57-brand Polish aggregate, four recruitment hosts receive 637 citations against 297 for Polish Wikipedia. The results show why normalization, classification rules and denominators belong in citation audits. They describe displayed links without estimating causal effects on brand reputation.

cs.IR · cs.CL · cs.CY

新论文 · v1arXiv:2611.20114

A plug-in energy-balance freezing module for transformer frost nowcasting without leaf-sensor inputs

Juyeon Kim, Chaeyeon Yi, Yun Am Seo

原文摘要

Frost protection requires local warnings where leaf-sensor inputs are unavailable. In the frost literature we surveyed, physics entered machine-learning models through inputs, numerical-model coupling, or losses, rather than model structure. We develop a transformer-compatible module that embeds leaf energy-balance physics in freezing-probability forecasts: an energy-balance temperature reference and a learned residual predict leaf temperature, which drives a multiplicative gate imposing a temperature-dependent probability cap. We mathematically prove conditional monotonicity and an alarm boundary, and test whether placing the same information in the output pathway or in attention yields different benefits. We evaluated 12 hourly leads at 19 Korean stations over five frost seasons, with and without leaf-sensor inputs. In the 2025-2026 hold-out season without these inputs, our time-token transformer (PET-E) achieved a lead-mean critical success index (CSI) of 0.516 versus 0.478 for fixed-configuration, joint-lead LightGBM, with significant gains at all 12 leads, and the fraction of alarms at observed leaf temperatures of at least 2 degrees C fell from 0.087 to 0.051 relative to the unmodified backbone. Across the hold-out and a retrospective 2024-2025 evaluation, the module reduced Brier scores and warm-surface alarm fractions in both the time-token transformer and iTransformer, with no significant CSI decrease at any lead. At held-out stations without static or leaf-sensor inputs, both module-equipped backbones exceeded LightGBM at all 12 leads (CSI 0.468 and 0.467 versus 0.437). In the time-token backbone, the temperature pathway reduced this alarm fraction in all four settings, whereas the attention prior increased CSI in three. These findings link an explicit temperature-dependent probability cap to improved forecasts and distinguish output-side from attention-based embedding.

stat.AP · cs.LG · physics.ao-ph

版本更新 · v2arXiv:2510.00650

Threats to the sustainability of Community Notes on X

Zahra Arjmandi-Lari, Alexios Mantzarlis, Tom Stafford

原文摘要

Community Notes are emerging as an important option for content moderation. The Community Notes system pioneered by Twitter, now known as X, uses a bridging algorithm to identify user-generated context with upvotes across political divides, supposedly spinning consensual gold from partisan straw. It is important to understand the nature of the community behind Community Notes, especially as the feature has now been imitated by several billion-user platforms. We look for signs of stability and disruption in the X Community Notes community and interrogate the motivations which may be driving users to contribute. We show substantial churn in the population of users who author Notes. We note that the majority of notes considered "helpful" (and therefore shown to users on X) is low, 10%, and declining. We conduct a novel analysis of the impact of having a note published, which requires being considered "helpful" by the bridging algorithm, utilising a regression discontinuity design. This allows stronger causal inference than conventional methods used with observational data. Our analysis shows than any positive effect on future note authoring of having a note published is unclear. This analysis has implications for the future of Community Notes on X and the extension of this approach to other platforms.

cs.SI

新论文 · v1arXiv:2611.20112

Washout-free synchronization, parameter estimation and uncertainty quantification of echo state networks

Andrea Nóvoa

原文摘要

Echo state networks (ESNs) forecast chaotic systems for several Lyapunov times. Like all recurrent networks, ESNs require a washout before forecasting, i.e., a phase in which the full state drives the network until the reservoir state synchronizes with the system. At deployment, however, sensors measure only part of the state, with noise and at a lower sampling rate than the training data. We propose a washout-free deployment of ESNs via ensemble data assimilation. In closed loop, the ESN is a Markovian dynamical system; hence, a sequential filter can correct its reservoir state whenever data are available. The filter synchronizes an ensemble of reservoir states, drawn from the training data, with the system, and quantifies the forecast uncertainty. We extend the method to parametric ESNs, to infer physical parameters, and to parallel ESNs with localization, to forecast high-dimensional systems. We test the method on the Lorenz-63 and Lorenz-96 systems. First, the filter synchronizes the ESN with a fraction of the data of a washout. Second, the filter deploys one full-state ESN from a subset of its components, which a washout cannot do. Third, the filter estimates a forcing that is not in the training data, and the estimated forcing lengthens the forecast. Fourth, localization synchronizes a parallel ESN with eight times fewer members than the global filter. Fifth, the ensemble spread signals, without the true state, when the ESN synchronizes and when the filter fails. This work enables the deployment of ESNs from partial, sparse and noisy observations.

nlin.CD · cs.LG

新论文 · v1arXiv:2611.20111

When Valid Decisions Are Not Enough: Evaluating Structural Constraints in Knowledge-Guided Generative Systems

Josef Pavlíček, Petra Pavlíčková, Martin Molhanec

原文摘要

Knowledge-guided generative systems use explicit domain constraints to improve decision consistency and alignment with human-created solutions. However, locally valid decisions do not necessarily preserve contextual relationships in human decision-making. This paper examines when structural constraints improve or degrade agreement with human reference solutions. We introduce HARMONIX, an experimental framework using automated musical harmonization as a controlled testbed for knowledge-guided decision-making. Four configurations share the same Viterbi-based optimization mechanism: a baseline, repetition constraints, functional constraints, and their combination. Experiments cover 19,745 records and 143,634 motif pairs across training and validation partitions. Combined constraints improve reference root-interval agreement by approximately five percentage points in stable structural contexts but reduce it by nearly nine percentage points in contexts involving transposition tracking. These opposing effects are consistent across both partitions, while overall exact chord agreement changes only marginally. Contexts are identified retrospectively using reference annotations; no adaptive constraint-selection mechanism is evaluated. The findings demonstrate that interpretable constraints can introduce systematic context-specific failures concealed by aggregate accuracy metrics. They highlight the need to evaluate relationships between decisions and the contextual applicability of constraints in knowledge-guided generative systems.

cs.AI · cs.CY

新论文 · v1arXiv:2611.20110

Who Observes Whom: Network Rationalizability and Its Complexity

Mojtaba Fayaz-Bakhsh, MohammadAmin Fazli

原文摘要

Nash equilibrium and rationalizability sit at opposite informational extremes: the former presumes that each player knows all opponents' strategies, the latter that players reason from common knowledge of rationality alone. When visibility is partial and asymmetric, we study the solution concepts induced by a directed \emph{observation network} $G$ in which an edge $i\to j$ means that $i$ observes $j$'s strategy. A set of profiles is \emph{self-sustaining} if every profile in it can be justified, for every player, by conjectures from the same set that agree with everything the player observes. We consider \emph{point} conjectures (single profiles) and \emph{belief} conjectures (distributions over profiles). Both range from Nash equilibrium on the complete network to point- and correlated rationalizability on the empty network, and shrink monotonically as links are added. Network topology governs both behavior and complexity. A predicted profile can be found in polynomial time on the empty network, whereas a single link makes search $\mathsf{PPAD}$-hard: under one-sided two-player monitoring, the observed player's strategies are exactly her Nash equilibrium strategies. On hierarchical networks (transitive tournaments), membership reduces recursively to smaller games and admits polynomial-size certificates, placing verification in $\exists\mathbb{R}$, and in $\mathsf{P}$ for two players. On the directed three-cycle, however, iterated elimination may never terminate under either variant, and under point conjectures interior predictions may admit only infinite witness sets, a pathology that belief conjectures avoid in the same game. Finally, when only realized pure actions are observed, the solution set is guaranteed to be non-empty exactly when the network has no edges, and is computable in time $O(n|A|^2)$.

cs.GT · cs.CC · cs.MA · econ.TH

版本更新 · v2arXiv:2611.15260

Driving World Models Move the Camera, Not the Scene

Yuqiao Feng, Xiaoyu Ao

原文摘要

Plan-conditioned driving world models are used to compare candidate plans, so their forecast should change with the plan as the real scene would, yet widely used control metrics check only whether the camera follows the command. Recorded drives already contain real outcomes of plans that were never driven: retiming a log gives pose-matched images and LiDAR for other speeds, and a stereo rig records a laterally shifted view. Against these references, released world models move their camera but barely move the scene: their scene response, the share of the real change in distance that the forecast reproduces, reaches at most about a third of real video's, and the trajectory score cannot tell an executed speed command from an ignored one. Interventions inside the generator trace the failure to the plan's channel. The forecast geometry follows the context frames; once they carry the planned geometry, plan tokens add little. ReTime-Compose therefore delivers the plan as geometry, without training. It transports the frames of one logged-speed rollout to the planned cameras, and the frozen model completes only what no frame covers. Frozen generators then follow a 0.54 m lateral shift nearly as the real scene does on KITTI, nuScenes and Argoverse 2, and respond more than re-sampling to slow-down plans in three generators and to speed-up plans in two.

cs.CV

版本更新 · v2arXiv:2608.25773

Drift-Aware Multimodal User Representation Learning via Multi-Scale Temporal Modeling and Sparse Mixture-of-Experts

Ziqing Qian, Haohang Chen, Shengqi Dang, Yuhan Xiong, Canyu Shen, Jiaying Lei, Nan Cao

原文摘要

Understanding user preferences from noisy and temporally evolving social media behaviors is fundamentally challenging due to interest drift, where user preferences shift across time and exhibit both multi-scale temporal patterns and diverse co-existing interests. To address this, we propose DUMoE, a unified framework for drift-aware multimodal user representation learning. Our model consists of (i) a temporal dynamics-aware backbone that captures and integrates static profiles, short-term behavioral signals, and long-term dependencies into a coherent representation, and (ii) a sparse mixture-of-experts (MoE) interest adapter that disentangles multiple latent interests via expert specialization and adaptive routing. Each expert models a distinct interest subspace, while a gating network dynamically selects and aggregates a sparse subset of relevant experts for each user. To enable stable and effective optimization, we further introduce a three-stage training strategy that decouples backbone learning, expert specialization, and gating optimization. Extensive experiments on real-world social media datasets show that DUMoE consistently outperforms state-of-the-art methods on both user interest prediction and interaction prediction tasks.Code is available at https://github.com/idvxlab/DUMoE.

cs.LG

版本更新 · v2arXiv:2609.29067

TileBench: A Controlled Benchmark for Performance Evaluation and Bottleneck Diagnosis of Tile-Based Programming Models

Bowen Cui, Zhongchun Zhou, Hao Wu, Tejas Ramesh, Junyu Yin, Jialiang Gu, Keren Zhou

原文摘要

Tile-based programming models, such as Triton and cuTile, aim to simplify high-performance kernel development, but their practical performance, tuning behavior, and usability remain difficult to compare systematically. We present TileBench, a controlled benchmark for evaluating Triton and cuTile on NVIDIA B200 GPUs under matched operator semantics and comparable implementation structures. TileBench contains 45 operators covering diverse AI-kernel patterns and memory/computation behaviors. Each task provides a PyTorch reference, verified Triton and cuTile implementations, standardized data-types (dtype) and input-size sweeps, default and autotuned configurations, roofline-based metrics, and profiling-guided diagnosis. Our evaluation shows that performance gaps are workload-dependent: cuTile excels on a small cluster of Tensor-Core/TMA-friendly kernels, while Triton is stronger on many irregular, streaming, and bandwidth-bound operators. We further evaluate LLM-generated cuTile and Triton kernels and find that Triton is consistently more token-efficient than cuTile under the same iterative refinement protocol. TileBench is publicly available at https://github.com/Deep-Learning-Profiling-Tools/Tilebench.

cs.PF · cs.PL

新论文 · v1arXiv:2611.20107

SAFER: Navigating the Exploration-Exploitation Dilemma in World-Model-Based Planning for Autonomous Driving

Yansong Qu, Zilin Huang, Zihao Sheng, Sikai Chen, Samuel Labi

原文摘要

An autonomous vehicle must foresee the consequences of its actions, especially in the rare, time-critical situations where crashes happen. A world model offers this foresight, but only within the coverage of its training data, which poses an exploration-exploitation dilemma. Within this coverage a planner can hardly improve on the data, and beyond it the model's errors mislead it. Logs of competent driving rarely contain the failures that would widen the coverage, and agents that learn from interaction gain them only by failing themselves, without being shown what to do instead. We propose SAFER, a framework that learns offline from failures that others make, judged by an expert, and keeps learning online from its own driving. Offline, partially trained agents and scripted, recoverable perturbations Stage failures in simulation, and an expert Annotates every visited state with its action and value, so that much of what the planner explores is covered before it drives. In planning, a chunked latent world model Foresees the consequences fast enough for real-time use, and a risk-aware sampling-based planner Evaluates candidate maneuvers by their return minus a bounded collision penalty. Online, a loop continuously Refines the model wherever this driving leaves the coverage of the offline data. In closed-loop evaluation against the compared methods, SAFER reaches the highest route completion and reward per step on the CARLA Leaderboard routes, 0.95 and 0.77 from a privileged bird's-eye view, and the highest episode reward in MetaDrive, 53.4 in regular and 49.0 in safety-critical traffic. Project page: https://ys-qu.github.io/safer-website.

cs.RO · cs.AI

新论文 · v1arXiv:2611.20106

Making Preemption Productive in Long-Output LLM Serving

Ziji Shi, Xinyi Wan, Chaoyi Ruan, Guangxing Huang, Min Lin, Jialin Li

原文摘要

Reasoning LLMs generate long outputs whose key--value (KV) caches occupy GPU memory for minutes. When memory fills, the server preempts requests, and a preempted request can repeatedly lose and recover its state while producing little new output. We call this behavior KV thrashing. In certain cases, existing systems leave more than a fifth of requests stalled for over a minute. Because preemption cannot be avoided without delaying admission, we ask how to make it productive. We present Stint, a serving system that preserves a preempted request's work, bounds how long it waits, and gives it fresh output after it returns. Stint stores fixed-length KV snapshots in a contiguity-first CPU pool with an exact-extent allocator, and moves them with fused K/V gather/scatter kernels and double buffering. Age-based readmission brings stalled requests back, and protection gives them a turn of fresh decoding before another preemption. Remaining-length estimates from decoding hidden states guide victim selection. On olympiad-math replays with Qwen3-32B on H200 GPUs, no request pauses for more than 50 seconds under Stint and p99 time per output token falls by 50.7\% relative to recomputation in the same engine. The replay as a whole finishes 10.7\% sooner (12.0\% higher throughput), because the requests that baselines starve no longer finish last.

cs.DC · cs.AI · cs.OS

新论文 · v1arXiv:2611.20103

Trading Parameters for Accuracy in MambaVision for Chest Radiograph Classification

Qaiser Farooq, Tehseen Zia

原文摘要

MambaVision, a hybrid state-space and attention backbone, places most of its parameters in its widest, fourth stage. We quantified what that stage contributes to multi-label chest radiograph classification. The fourth stage was replaced with lighter single-scan state-space blocks adapted from Spatial-Mamba, with multi-scale dilated depthwise convolutions applied to the scan output; the first three stages were retained. Tiny, Small, and Base models were evaluated across nine dataset-scale combinations drawn from two source cohorts: the National Institutes of Health ChestX-ray14 archive (full ChestMNIST and a high-resolution subset) and Indiana University Open-i. At the Tiny scale, replacement reduced parameter count by 35.7% and increased graphics processing unit throughput by 19.3%; the savings came from the stage's dense projections, not from removing quadratic attention. Without distillation, the mean area under the receiver operating characteristic curve on ChestMNIST was 0.004 lower than the baseline (95% bootstrap confidence interval, 0.002 to 0.006), and 0.003 to 0.006 lower across all nine combinations. Knowledge distillation from an unreplaced teacher narrowed the gap to the undistilled baseline to 0.001-0.002 and met the equivalence criterion on ChestMNIST, but distilling the baseline with the same recipe improved it by 0.001-0.003, so a gap of 0.002-0.005 persisted under symmetric training. At the Base scale, the unreplaced Small model was both smaller and more accurate. Fourth-stage replacement is a useful trade-off only for Tiny and Small models. These measurements let developers of chest radiograph screening tools weigh model size and speed against a quantified loss of diagnostic discrimination.

cs.CV

新论文 · v1arXiv:2611.20100

EgoSwarm: Learning Decentralized Robot Swarm Intelligence from Human Swarm Demonstrations

Ziqiang Liu, Xitong Chen, Min Li, Yiqun Li, Zhuoping Yin

原文摘要

Advancing from individual embodied intelligence to swarm embodied intelligence is a key step toward scalable, general-purpose embodied intelligence. Complex tasks often require multiple agents to divide responsibilities, coordinate handovers, and manipulate shared objects collaboratively. However, existing robot learning methods rely primarily on individual demonstrations, which provide limited supervision for modeling dependencies among agent actions, while collecting multi-robot demonstrations at scale remains difficult. Human egocentric demonstrations offer a scalable alternative, yet distributed observations from multiple participants cannot directly supervise decentralized policies. We introduce EgoSwarm, the first robot learning framework for learning decentralized multi-robot collaboration from human swarm demonstrations. EgoSwarm collects more than 50 hours of collaborative experience using a robot-free multi-person egocentric system. We introduce the Swarm Interaction Packet (SIP) and Swarm Interaction Token (SIT) to recover cross-agent interaction relations from distributed observations and transform this experience into local supervision consistent with deployment-time information constraints. We further propose SwarmVLA, which transfers knowledge of swarm behavior to decentralized policies conditioned on local observations, local history, and limited neighbor messages. SwarmVLA achieves a mean success rate of 72.2% across three categories of multi-robot collaborative tasks. Project page:https://xitong-c.github.io/EgoSwarm_Page/.

cs.RO

新论文 · v1arXiv:2611.20095

High-Throughput Serving of Nemotron 3.5 Transcribe with Regime-Aware Megakernels and Kernel Fusion

Aditya Chatterjee, Sanjay Singh Chauhan, Mayank Jain, Arushi Dhingra, Sahin Chowdhury, Rahul Mittal

原文摘要

Nemotron 3.5 Transcribe ranks first among open-weight models on the Hugging Face Open ASR leaderboards. Serving it efficiently takes more than fast operators: continuous batching changes the amount and mix of work at every iteration, and the best fusion granularity changes with it. We present a serving engine that, for every pass, chooses between persistent megakernels and chains of fused kernels at crossovers measured offline. On an H100 SXM GPU, throughput of our accuracy-preserving serving peaks at 4,345x real time, 1.39x the peak of tuned vLLM, and is up to 4.39x faster at low concurrency; on B200 it peaks at 6,367x, 1.64x vLLM at the highest concurrency we measured. Word error rate (WER) is 1.91 to 1.93% against 1.88 to 1.93% for vLLM. Within a 400 ms budget for p95 end-to-end completion time, it serves 8x as many concurrent requests. Choosing per pass beats either level alone: 23% higher throughput than always using megakernels at 128 concurrent requests, and 9% higher than always using fused kernels at 8. Opt-in switches, including greedy MoE layer skipping, add 13 to 16% throughput on B200 for offline jobs that can accept higher WER.

cs.PF · cs.DC · cs.LG · eess.AS