跳到正文
研踪PaperTrail

YOUR RESEARCH, IN SIGHT

让下一篇重要论文，不再错过。

覆盖计算机全学科，以 AI 为快捷入口。关注方向、作者和关键词，把发现变成持续的研究积累。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+1072新增论文
+845其中 AI 相关
+94已有论文修订
+100新增 AI 导读

arXiv 最近完成扫描：2026-11-09 19:37 · 全站最近入库：2026-11-09 13:22各信源进度 →

统计口径与数据新鲜度

统计 2026-11-08 20:23 — 2026-11-09 20:23（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：2026-11-09 19:37最近发现新增版本：2026-11-09 13:22库内论文最新更新时间：2026-11-07 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。

18,172 已收录论文15,506 AI 相关4,527 有修订112 HF 社区发现

浏览论文库

研究日报 / 周报 →按信源浏览 →

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 18,068 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-09 19:35（北京时间）
新论文 · v1arXiv:2611.09601

A Fast and Accurate Algorithm for the $p \to q$ Norm of Sparse Non-Negative Matrices

Marco Pirazzini, Daniel A. Spielman

原文摘要

We present an algorithm that computes the $p \to q$ norm of a non-negative matrix with $s$ nonzero entries to accuracy $\varepsilon$ in time $\widetilde{O}(s^{3/2} \log (1/\varepsilon))$, for any $p \ge q \ge 1$. The fastest algorithm previously designed for this problem has running time $\widetilde{O}(s / \varepsilon)$ (Objois and Vladu, 2026), which is slower when high accuracy solutions are desired. Our algorithm employs a surprising application of Laplacian solvers inside an Interior Point Method (IPM). We formulate the problem as a linear optimization problem over a product of power cones, and show that the Newton updates of the IPM can be computed by solving a Laplacian system in a Schur complement after subtracting a rank-2 matrix. We also discuss an implementation of our algorithm, showing that it is practical and it outperforms optimized solvers such as MOSEK by factors of up to $20\times$ on a variety of tasks.

cs.DS

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09600

Coresets for Capacitated Clustering via Dual Concentration

Milind Prabhu, Chris Schwiegelshohn, Sudarshan Shyam, Erik Waingarten

原文摘要

We study coresets for clustering problems, focusing on the classical $k$-median and $k$-means objectives. A coreset is a small weighted summary of the input data that approximately preserves the clustering cost for every choice of centers. Such summaries are a central tool for scaling clustering algorithms to large datasets. In this work, we consider the capacitated setting, where the clustering is subject not only to choosing centers but also to respecting prescribed cluster-size constraints. This model is a natural generalization of standard clustering and has been widely studied, but its coreset complexity is less well understood. For finite metric spaces, we obtain optimal-size coresets; surprisingly, these match the best-known bounds for unconstrained clustering. We also give improved coreset bounds in Euclidean space, notably the first coresets with near-linear dependence in $k$ while being independent of the dimension and input size. The main new technical ingredients are concentration bounds for the dual solutions of min-cost flow problems, which allow for both significantly simpler and tighter analyses of the underlying sampling distribution used in the coreset construction.

cs.DS · cs.CG · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09599

Dual-Region Slice Aggregation for CT-Based Renal Tumor Classification

Yuan Liang, Sourav Bhattacharjee, Abraham Campbell

原文摘要

Clear cell renal cell carcinoma (ccRCC) is the most common kidney cancer in adults. We present a region-aware architecture for CT-based ccRCC classification. Tumor and 0--10 mm peritumoral context are pooled separately from shared frozen DINOv3 feature maps. A shared position-aware slice aggregator precedes fixed training-only principal component analysis (PCA). The resulting representations are fused with radiomics using two scalar feature gates. Evaluation on KiTS23 and 111 external cases yields internal test and external AUCs of 0.8078\pm0.0154 and 0.9094\pm0.0015. Compared with matched single-tumor readout, mean AUC increases by 1.50 and 2.34 percentage points, respectively. Early-PCA and reconstruction controls yield lower internal mean AUCs, while external AUC is similar for early and late PCA. These findings highlight the joint role of regional readout and slice aggregation in adapting frozen visual features to renal CT.

cs.CV

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09598

Transformer-FSQ: Discrete Latents for Extreme KV-Cache Compression

Andras Szecsenyi, Thomas L. Lee, Vaisakh Shaj, Elliot J. Crowley, Amos Storkey

原文摘要

Private, personalised language models must run on local consumer devices, where large, frontier models cannot fit, and retaining long contexts is non-trivial. Sequential memory---the key-value (KV) cache---grows linearly with context length, quickly dwarfing the model itself: for Qwen3-1.7B-Base, 200,000 retained entries require 21 GiB, 6.6 times more than the model weights. Yet the KV-cache compression methods we measured fail beyond moderate compression ratios. We introduce Transformer-FSQ, addressing extreme KV-cache compression by replacing standard key and value projections with a finite scalar quantisation bottleneck. Our cache stores a packed vector of discrete integer codes per token, with no per-token quantisation metadata. Converting a pretrained transformer trains only these projections on 0.25 tokens per original parameter. We evaluate Transformer-FSQ against five baseline families on models from 0.6B to 8B parameters. On long-context retrieval (RULER), from 12$\times$ to 97.5$\times$ Transformer-FSQ is the most accurate method we evaluate, across scales. A fused Triton kernel decodes directly from the integer codes; with baselines given their authors' kernels where available, it decodes at $1.01\times$ (95% interval $0.98$--$1.05$) the throughput of native SDPA attention on the uncompressed Qwen3-8B-Base at $32\times$ and $1.54\times$ at $97.5\times$. Extreme KV-cache compression is therefore a practical way to extend how much context a fixed device can hold.

cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09597

RuACT: Recovering Inactive Rubrics as Adaptive Critics and Teachers

Yijun Li, Zhengxi Lu, Zhipiao Liu, Jinzheng Yu, Xutao Weng, Jian Shao, Yongliang Shen

原文摘要

Rubric-based reinforcement learning (RL) extends RL with verifiable rewards to open-ended tasks by scoring each response against query-specific rubrics. Under group-relative optimization, however, a rubric that every sampled response fails (unreachable) or passes (saturated) gives all responses the same outcome and no learning signal; in Qwen3-4B medical training, 60.88% of query-rubric observations are inactive. We find that the two states lack different supervision: stating an unreachable rubric in the prompt raises its satisfaction from 1.13% to 91.25%, whereas responses that all pass a saturated rubric still differ in quality. We therefore propose RuACT (Rubrics as Adaptive Critics and Teachers), which assigns each inactive rubric the role that supplies what it lacks. Rubric as Teacher uses unreachable rubrics as privileged information for reward-gated on-policy self-distillation, and Rubric as Critic decomposes saturated rubrics into child rubrics that separate current responses while sharing the parent's weight. RuACT outperforms GRPO, rubric reweighting, and rubric-guided exploration on all four medical and scientific benchmarks with Qwen3-4B and achieves the best average with Qwen3-1.7B. On initially unreachable rubrics, unguided recovery rises from 35.02% to 42.86%, while child refinement improves reward-quality alignment on independent rollouts from 57.74% to 60.77%. Code is available at https://github.com/ZJU-REAL/RuACT.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09596

Coordinatewise Epsilon for Adam under Heavy-Tailed Noise

Jeonseong Kim, Chenglin Fan

原文摘要

Adam estimates second moments coordinatewise, but typically adds the same scalar $\varepsilon$ to every denominator. We study how this shared $\varepsilon$ affects convergence when noise and curvature vary across coordinates. Under coordinatewise finite-$p$ noise moments, $p\in(1,2]$, choosing $\varepsilon_i$ from the noise and curvature bounds for each coordinate attains the heavy-tailed $\ell_1$-stationarity benchmark $O(T^{-\frac{p-1}{3p-2}})$, matching the known optimal time exponent for smooth nonconvex stochastic optimization. For $1<p<2$, Adam with a shared $\varepsilon_t$ admits a lower bound of order $Ω(T^{-\frac{p-1}{2p}})$. This lower bound holds for every predictable scalar choice of $η_t$, $β_{1,t}$, $β_{2,t}$, and $\varepsilon_t$, precluding a dimension-uniform benchmark rate for this class.

cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09589

Dual-Number Associative Arrays for One-Pass Provenance and Sensitivity

Timothy Habashy, Hayden Jananthan, Jeremy Kepner

原文摘要

Associative arrays and GraphBLAS express sparse computations as matrix operations over a chosen semiring, but the result normally contains only the ordinary value. We define a dual-number extension $D_k(S)=S\ltimes S^k$ that carries this value together with $k$ independent auxiliary channels in the same computation. We prove that the construction is a split square-zero extension of any commutative semiring. Over min-plus, the auxiliary value is the weight of the shortest walk constrained to use the corresponding edge, excluding the weight of that edge. This gives an exact shortest-path membership test for that edge, although ties return the union of the edges on all optimal paths. We implement the min-plus construction using sparse \texttt{python-graphblas} operations, with a fixpoint test that checks the value and auxiliary components together. We validate the implementation across 18,692 checks, including 2,825 tied target pairs. On random graphs with 100,000 vertices and 32 seed edges, the dual method is $3.39\times$ faster than the sequential algorithm but $1.50\times$ slower than a min-plus-specific batched method. This difference is largely explained by the dual method requiring $1.55\times$ as many sweeps. Thus, the main contribution is a general algebraic approach with competitive sparse performance, rather than information that specialized min-plus algorithms cannot already compute.

cs.DS

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2611.00375

MoRE-GS: Motion-Regime Experts for Future Gaussian Splatting

Shima Salehi, Joshua Peeples

原文摘要

Future dynamic novel-view synthesis requires extrapolating multiple motion regimes within the same scene. A single dynamics model may not simultaneously represent stationary regions, coordinated motion, local interactions, and articulated trajectories. We introduce Motion Regime Expert Routing for Gaussian Splatting (MoRE-GS), a causal, render-aware framework for forecasting dynamic 3D Gaussian scenes. MoRE-GS freezes a Gaussian representation reconstructed exclusively from the observed temporal prefix and forecasts its control-keypoint motion using four complementary experts: persistence, a coefficient-space Transformer, a graph convolutional network, and a structured Transformer with rigid and non-rigid motion constraints. A compact routing table assigns expert mixtures across soft motion groups and forecast-horizon bins, while a persistence gate suppresses unreliable extrapolation. Across D-NeRF and HyperNeRF scenes, MoRE-GS achieves improved performance against other methods, and these results demonstrate that renderer-calibrated composition of complementary motion predictors improves causal dynamic Gaussian forecasting across heterogeneous motion regimes. Project page: https://advanced-vision-and-learning-lab.github.io/MoRE-GS-website/

cs.CV

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2604.22235

Learning-augmented robotic automation for real-world manufacturing

Yunho Kim, Quan Nguyen, Taewhan Kim, Young Jin Heo, Joonho Lee

原文摘要

Industrial robots are widely used in manufacturing, yet most manipulation still depends on fixed waypoint scripts that are brittle to environmental changes. Learning-based control offers a more adaptive alternative, but it remains unclear whether such methods, still mostly confined to laboratory demonstrations, can sustain hours of reliable operation, deliver consistent quality, and behave safely around people on a live production line. Here we present Learning-Augmented Robotic Automation, a hybrid system that integrates learned task controllers and a neural 3D safety monitor into conventional industrial workflows. It retains the core industrial automation backbone while introducing learning only where adaptability is needed. We deployed the system on an electric-motor production line to automate deformable cable insertion and soldering under real manufacturing constraints, a step previously performed manually by human workers. With less than 20 minutes of real-world data per task, the system operated continuously for 5 hours 10 minutes, producing 108 motors without physical fencing and achieving a 99.4% pass rate on product-level quality-control tests. It achieved near-human takt time with reduced cycle-time variability while consistently producing high-quality solder joints comparable to those of experienced operators. These results establish a practical pathway for extending industrial automation with learning-based methods.

cs.RO · cs.AI · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2310.07801

Stay on the Manifold: Hierarchical Bayesian Trajectory-Family Transfer for Sparse-Target Reconstruction and Sampling

Elvis Han Cui, Yuhan Pan, Xin-Ya Zhang, Bingbin Li, Yanan Li, Donghui Wang, Weng Kee Wong

原文摘要

Where should reconstructions and new samples lie when a few observations reveal only fragments of a trajectory? We propose hierarchical Bayesian trajectory principal-manifold sampling (HB-TPM), which learns a family of smooth curves from ordered source tasks and adapts it to sparse targets. For unordered vectors, alternating progression assignments and coefficient updates estimate trajectory structure between and beyond the observed shots under the learned source family. Observation-aware extensions connect the trajectory prior to photon arrivals and image-encoded cardiac cycles. Our analysis characterizes the coefficient directions supplied by the prior and bounds sampling error conditional on curve recovery. Across 30 held-out vector tasks, HB-TPM reduces mean sliced-Wasserstein distance by 21.0-41.8% relative to a source-adapted VAE. Prior-strength and sampling ablations identify coefficient shrinkage as the main supported mechanism. On 100 simulated photon records, adaptation lowers mean relative trajectory RMSE by 50.2% versus target-only fitting and 23.6% versus source-mean prediction. Across three training seeds and five folds on 98 cardiac sequences, HB-TPM with latent-MSE training achieves the best mean reconstruction among the tested configurations. Together, the results demonstrate trajectory-family transfer for sparse-target reconstruction and trajectory-centered sampling when source and target tasks share comparable progression structure.

cs.CV · cs.AI · stat.ME

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09585

Scalable Full-Horizon SFC Deployment over Predictable SAGIN Topologies via Configurable Column Generation Framework

Hang Li, Hao Huang, Yihu Xu, Zixuan Kong, Yanling Jin

原文摘要

Space-air-ground integrated networks (SAGINs) integrate heterogeneous communication and computing resources to support ubiquitous services. Service function chains (SFCs) enable complex services to be composed of ordered functions and executed collaboratively across these resources. Mobility continuously reshapes connectivity throughout service execution, yet much of the resulting topology evolution is predictable. Decisions based only on isolated topology snapshots may therefore overlook valuable future connectivity. We formulate an admission-first, cost-aware model on a time-expanded graph for SFC deployment. Our key observation is that temporal feasibility is request-specific, whereas global coupling arises mainly from shared-resource competition. We therefore develop a configurable column-generation framework in which complete request deployments form columns and a restricted master problem coordinates admission and resources. SAGINFE, a fast constructive heuristic built on per-flow embedding (PFE), supplies a standalone feasible solution and initial columns. Configurable pricing schedules combine early-terminated and stricter searches, supplemented by selective PFE candidates. Candidates undergo feasibility and exact reduced-cost checks. Request-wise pricing can run in parallel, followed by finite-pool integer selection. The default configuration improves the admission-first objective over initialization on all 120 main instances. Four pricing workers achieve up to a 2.69-fold median pricing-stage speedup while preserving admitted request counts in the tested cases.

cs.NI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09584

Extended Model-Based Learned Inertial Odometry

Sasanka Kuruppu Arachchige, Joni-Kristian Kämäräinen

原文摘要

Inertial odometry is a compelling approach to state estimation for agile quadrotor flight due to its affordability, low weight, and robustness to perceptual degradation. However, relying only on integrated inertial measurements is impractical, as sensor errors and time-varying biases lead to significant pose drift. Although recent advances have enabled inertial odometry for drone racing, existing approaches show limited generalization to trajectories not seen during training. This work improves generalization by adopting a body-frame representation and incorporating body-frame torque dynamics into the learning pipeline. The learned model predicts short-horizon relative displacements, which are fused with IMU measurements in an Extended Kalman Filter. Experimental results show that the proposed method outperforms the previous state-of-the-art learned inertial odometry approach for quadrotor pose estimation on unseen trajectories. An additional evaluation on four public quadrotor datasets, with whole trajectory shapes held out of training, confirms a lower velocity error on every held-out set and shows that the torque input is most informative when it is derived from the rotor signals. Code, data pipeline and trained models are publicly available.

cs.RO

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09583

MSR-ABC: Complementary Feature Fusion for Clear Cell Renal Cell Carcinoma Classification

Yuan Liang, Sourav Bhattacharjee, Abraham Campbell

原文摘要

Clear cell renal cell carcinoma (ccRCC) classification from computed tomography (CT) may benefit from representations that capture different aspects of tumour phenotype. However, simply concatenating reconstruction-derived features, image--text embeddings, and radiomic descriptors may not fully exploit their complementary information. We propose MSR-ABC, a global--local framework integrating MedVAE, MedSigLIP, and radiomics. Spatially matched local image descriptors interact through cross-representation attention, while radiomic auxiliary supervision encourages the pooled local representation to retain quantitative tumour attributes. A gated residual mechanism integrates local information with global summaries. Area-aware pooling and visual and soft-text prompting complete the framework. We evaluate discrimination using a patient-level split of 396 KiTS23 cases and an external cohort of 111 TCGA/TCIA cases, reporting mean AUROC over three training seeds with 95% patient-bootstrap confidence intervals. MSR-ABC achieves mean AUROCs of 0.806 internally and 0.879 externally. Internally, it exceeds global MedVAE--radiomics and three-source concatenation by 3.35 and 3.61 AUROC percentage points, respectively. Removing any source lowers mean AUROC in both cohorts, with the largest decreases following radiomics removal. Individual removal of cross-representation attention, auxiliary supervision, or gated fusion also lowers internal mean AUROC. These findings are consistent with complementary contributions within the integrated architecture and support further investigation of structured representation fusion for non-invasive renal tumour characterisation.

cs.CV

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2605.07297

Spectrum-Adaptive Generalization Bounds for Trained Deep Transformers

Mana Sakai, Masaaki Imaizumi

原文摘要

Understanding why trained Transformers generalize well is a fundamental problem in modern machine learning theory, and complexity-based generalization bounds provide a principled way to study this question. While existing norm-based bounds for Transformers remove the explicit polynomial dependence on the hidden dimension, they typically impose fixed norm constraints specified a priori and can exhibit unfavorable exponential dependence on depth. In this paper, we derive spectrum-adaptive post hoc generalization bounds for multi-layer Transformers. Under layerwise spectral norm control, the bounds are expressed in terms of layerwise Schatten quantities of the query-key, value, and feedforward weight matrices. Since the Schatten indices need not be fixed a priori and can instead be selected after training, separately for each matrix type and layer, the bounds adaptively trade off spectral complexity against the dimension- and depth-dependent factors according to the learned singular-value profiles. Empirical comparisons of BERT-adapted proxies for the leading complexity factors suggest that the proxies induced by our bounds grow more slowly with depth and hidden dimension than the corresponding norm-based proxies. Overall, our results provide a complexity-based perspective on how the spectral structure of trained Transformers is reflected in generalization analyses.

stat.ML · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2609.00219

A Dichotomy for Complex Boolean Holant with Binary Disequality

Chenghua Liu, Boning Meng

原文摘要

We classify Boolean Holant problems with a supplied binary disequality signature and an arbitrary fixed finite set of algebraic complex-valued signatures. Each problem is polynomial-time computable or $\#\mathsf{P}$-hard, and the tractability criterion is decidable. The proof combines arity reduction with the growth of the available binary transfer group and uses stable pairs to recover the structure of higher-arity signatures. We also prove a dichotomy for the disequality-edge model when the signature set contains an endpoint-nondegenerate irreducible eight-vertex signature, resolving the corresponding arity-four cases in the main proof.

cs.CC

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09579

Locally Adaptive Conformal Prediction for Polynomial Chaos Expansions

Dimitrios Loukrezis, Arthur Hatstatt, Bruno Sudret, Dimitris G. Giovanis

原文摘要

This work develops a locally adaptive jackknife+ conformal prediction framework for polynomial chaos expansions. Therein, the non-conformity scores are normalized with an auxiliary polynomial chaos expansion trained on the log-transformed leave-one-out residuals of the base model. We show that the coverage guarantee of jackknife+ conformal prediction carries over to the adaptive intervals if the leave-one-out auxiliary models are trained on nested leave-one-out residuals. For fixed-basis polynomial chaos expansions fitted by ordinary least squares, all required quantities are derived in closed form and the guarantee-preserving adaptive intervals are obtained from a single model fit, without any model retraining. For sparse polynomial chaos expansions, where multiple model fits are necessary, we propose a heuristic simplification that limits the computational overhead of adaptivity at the cost of the formal coverage guarantee. Numerical experiments on an academic benchmark with known heteroscedastic noise and on an engineering simulation dataset show that the suggested method attains the target coverage while providing intervals that adapt to local prediction uncertainty, therefore being considerably more statistically efficient. These advantages are particularly prominent in high-uncertainty regions of the input space, where the performance of non-adaptive methods degrades substantially.

stat.CO · cs.CE · stat.ME · stat.ML

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2610.00081

A Full Complexity Dichotomy for Complex-Valued Boolean Holant Problems

Chenghua Liu, Boning Meng, Juqiu Wang

原文摘要

A Boolean Holant problem evaluates a graph partition function specified by local signatures at its vertices. We prove a complexity dichotomy for arbitrary finite sets of algebraic complex-valued signatures on general graphs: an explicit, decidable criterion characterizes the polynomial-time cases, and every remaining case is $\#\mathsf{P}$-hard under polynomial-time Turing reductions. The proof resolves the all-even case by constructing signatures with nonzero endpoints and all remaining support on the central layer, then classifying them according to half their arity. Binary transfer groups and matching-product reconstruction provide the main tools in the small- and large-height cases, respectively.

cs.CC

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2605.00072

XekRung Technical Report

Jiutian Zeng, Junjie Li, Chengwei Dai, Jie Liang, Zhaoyu Hu, Yiliang Zhang, Ziang Weng, Longtao Huang 等 16 位作者

原文摘要

We present XekRung, a frontier large language model for cybersecurity, designed to provide comprehensive security capabilities. To achieve this, we develop diverse data synthesis pipelines tailored to the cybersecurity domain, enabling the scalable construction of high-quality training data and providing a strong foundation for cybersecurity knowledge and understanding. Building on this foundation, we establish a complete training pipeline spanning continued pre-training (CPT), supervised fine-tuning (SFT), and reinforcement learning (RL) to further extend the model's capabilities. We further introduce a multi-dimensional evaluation system to guide the iterative improvement of both domain-specific and general-purpose abilities. Extensive experiments demonstrate that XekRung achieves state-of-the-art performance on cybersecurity-specific benchmarks among models of the same scale, while maintaining strong performance on general benchmarks.

cs.CR · cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09575

An Empirical Evaluation of a Weighted PALU Model for Internet Traffic Networks

Wood Moore, Hayden Jananthan, Jeremy Kepner

原文摘要

The PALU model is a hybrid power law generative framework that decomposes Internet traffic graphs into three structural components: a preferential attachment core, degree one leaf nodes, and disconnected satellite clusters. Although the original PALU formulation provided a theoretical basis for this decomposition, it was never validated against real traffic data and was restricted to unweighted networks. This work presents the first empirical evaluation of PALU against real Internet traffic, using packet traces from the MAWI WIDE samplepoint F dataset. We introduce a weighted extension of the model incorporating strength degree scaling, and develop a packet-to-graph pipeline for parameter estimation across multiple observation window sizes. Results show that the power law degree exponent $α\approx 2.0$ remains stable across observation scales from $10^5$ to $3 \times 10^6$ packets, consistent with the PALU structural prediction. However, the satellite cluster parameter $λ$ and the weighted scaling parameter $θ$ exhibit systematic dependence on observation scale, and the Poisson star assumption for unattached components is formally rejected at all tested window sizes. These results indicate that PALU captures important aspects of Internet traffic structure but requires an explicit observation model to fully account for finite sampling effects.

cs.NI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09573

Half-precision inference of AI weather prediction models: negligible loss of skill at a fraction of the cost

Stefan Siegert

原文摘要

AI weather prediction models are released with 32-bit (fp32) weights and run in fp32 by default, although current GPUs compute much faster in 16-bit formats. We run five publicly available models -- Aurora, Pangu-Weather, FourCastNet, SFNO and FastNet -- on a single 16 GB GPU from 52 initial dates in 2024, in fp32 and with their networks in half precision (fp16), and verify the forecasts against ERA5. The physical state must stay in fp32: converting the whole model rounds it in physical units, which in fp16 overflows and in bf16 discards meaningful information. With the state, its normalisation and denormalisation in fp32, running the network in fp16 changes the global RMSE by at most 1.4%, and by about 0.5% or less for four of the five models, equivalent to less than one hour of lead time up to day 8 and at most two hours at day 10. The differences show no systematic pattern by region, scale or for tropical and extratropical storms, apart from a small tropical penalty in FourCastNet, and fp16 leaves the CRPS, spread and calibration of Aurora's ensemble almost unchanged. No value overflowed fp16's numeric range (+/-65,504); the largest stay 38 to 544 times below it. A 10-day forecast in fp16 is 1.3-3.5 times faster than in fp32 and uses 1.4-3.7 times less energy. Sensitivity experiments show that rules of thumb from other applications of reduced precision carry over only partly: the first and last layers dominate for some models but not for others, and no type of layer is sensitive on its own. For running AI weather models on a single GPU, fp16 with the state in fp32 is practically lossless at a fraction of the cost. bf16, often preferred for its wider range, is as fast but less accurate than fp16 for every model, and simple 8-bit quantisation costs far more skill.

physics.ao-ph · cs.LG · cs.PF

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09570

Video Stitching from Multiple Moving Cameras

Shira Ifergane, Roy Amoyal, Shahar Benishay, Oren Freifeld

原文摘要

We address geometric video alignment for the same scene captured by multiple freely moving cameras. Existing methods, which are based on image stitching or video stabilization, are typically limited to two cameras or must assume the restrictive setting of a fixed linear ordering of the cameras. We propose MViS, a lightweight test-time optimization framework for joint multi-view alignment with temporal consistency. MViS uses a spatial-transformer-inspired net to jointly process frames from all views within a short temporal window and predict per-view, per-time homographies. Optimization proceeds in two stages: geometry-guided alignment followed by appearance-based refinement, with both stages enforcing multi-view and temporal consistency across the sequence. The method requires no scene-specific hyperparameter tuning, handles arbitrary and time-varying view-intersection graphs, and remains robust under rapid camera motion. On a standard two-view video benchmark, MViS matches or surpasses the state of the art. To facilitate evaluation in more complex scenarios, we introduce an annotated dataset with more than two freely moving cameras and establish the first benchmark for this setting. Project page: https://bgu-cs-vil.github.io/MViS-project/ .

cs.CV

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09569

Performance Bottlenecks in Computing ZK Proofs on a GPU

Victor Carolino, Caleb Geren, Michael Kaufman, Maximilian Nussbaum, William Scott, Henry F. Korth

原文摘要

Zero-knowledge proofs allow one party to prove the correctness of a statement or execution of a computation while revealing no more than the statement of correctness itself. Recent use cases, such as AI datasets, suggest the need for faster proof generation over increasingly large constraint sets. This paper explores efforts in optimizing zero-knowledge proof generation, specifically PlonK, using GPUs, for applications that are not amenable to solution via incrementally verifiable computing (IVC). Our intention with this work is to improve ZK performance by introducing fundamental changes in the low-level architectural structure of proof generation. These changes affect performance curves for most current proof generation algorithms. In particular, this paper explores the trade-offs between increased parallelization on GPUs and the resultant memory management problems that occur due to this parallelization. We implement a parallelized PlonK in order to analyze these trade-offs and make arguments concerning the nature of data flows within the proving protocol. Our results lead us to consider multi-GPU environments to maximize performance for large circuit sizes. Such environments introduce issues of communication, efficient hardware management, data partition, and data transfers. This paper further discusses the use cases that motivate multi-GPU implementations, their cost models, and the use of remote direct memory access (RDMA) for data transfer among GPUs.

cs.CR

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09568

Alignment by Confident Debate: Preference Learning from Multi-Agent Debate

Nikhil Verma, Dingxian Wang, Mohammad Reza Rezaei, Andrew Rabinovich, Zhao Chen

原文摘要

Post-training language models for reasoning typically rely on gold answers, task-specific verifiers, or learned reward models, limiting scalability when such signals are scarce. Multi-agent debate offers a self-supervised alternative, but converting debate into reliable training supervision remains challenging: majority voting weights agreeing responses equally, lacks a principled tie-breaking rule, and is ill-defined for open-ended tasks without a canonical answer; likelihood-based scoring can conflate fluency with correctness; and homogeneous committees may reinforce correlated errors. We introduce Alignment by Confident Debate (AbCD), which converts debate traces into preference data without gold answer labels. A sampling-diverse committee debates each problem, after which final answers are clustered by task-level semantic equivalence. We use verifiable tasks (mathematical reasoning, multiple-choice quantitative QA, and executable code-generation) as a controlled test bed: their answer keys or executable checkers let us evaluate the quality of debate-derived supervision against ground truth, even though the oracle itself never uses gold labels. Post-training on the resulting preferences outperforms inference-only ensembling and debate, as well as SFT on the same traces, across five model families spanning 1B to 8B parameters and four datasets, improving consensus accuracy, with a maximum gain of $21.5_{pp}$. Ablations show that the approach transfers to a larger model and to code generation, while also revealing limitations of heterogeneous committees and agreement-based weighting. Overall, the results suggest that debate traces can yield useful preference supervision when answer equivalence is combined with confidence-weighted disagreement.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09567

TurtLES: A Large-Scale Benchmark for Turbulent 3D Neural PDE Surrogates

Jarno Platenburg, Armand Kassaï Koupaï, Paola Cinnella, Patrick Gallinari

原文摘要

Machine-learning surrogates are increasingly used to accelerate computational fluid dynamics, yet progress is limited by the lack of benchmarks capturing realistic, time-dependent turbulent flows. We introduce TurtLES, a 14 TB dataset of 3D turbulent cylinder wakes in the shear-layer transition regime, computed via high-fidelity implicit large-eddy simulations. The dataset contains 400 trajectories of 400 time steps each, with 3-11 million points per frame, spanning variations in geometry, Reynolds number, and angle of attack. Unlike existing datasets, TurtLES combines (i) physically realistic turbulence with an active energy cascade driven by vortex stretching, (ii) dense irregular meshes, and (iii) long-horizon temporal dynamics. We benchmark state-of-the-art neural operators on autoregressive rollouts and identify key failure modes, including temporal instability, loss of small-scale energy, and challenges in learning on large irregular domains. TurtLES provides a new benchmark for evaluating surrogate models on 3D turbulent wakes at industrially relevant mesh resolution.

physics.flu-dyn · cs.LG

查看论文
arXiv ↗PDF ↗
版本更新 · v4arXiv:2507.08038

AblationBench: Evaluating Automated Planning of Ablations in Empirical AI Research

Talor Abramovich, Gal Chechik

原文摘要

Language model agents are increasingly used to automate scientific research, yet evaluating their scientific contributions remains a challenge. A key mechanism to obtain such insights is through ablation experiments. To this end, we introduce AblationBench, a benchmark suite for evaluating agents on ablation planning tasks in empirical AI research. It includes two tasks: AuthorAblation, which helps authors propose ablation experiments based on a method section and contains 83 instances, and ReviewerAblation, which helps reviewers find missing ablations in a full paper and contains 350 instances. For both tasks, we develop LM-based judges that serve as an automatic evaluation framework. Our experiments with frontier LMs show that these tasks remain challenging, with the best-performing LM system identifying only 45% of the original ablations on average, below human-level performance. We observe an inverse performance trend between the author and reviewer tasks, which we attribute to differences in model grounding. Lastly, we analyze the limitations of current LMs on these tasks, and find that chain-of-thought prompting outperforms an agent-based approach. Our data is available on https://huggingface.co/collections/ai-coscientist/ablationbench, and our code is available on https://github.com/ai-scientist-bench/ablation-bench .

cs.CL · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v4arXiv:2607.01730

Resilient Liquid Democracy: Mitigating Voting Power Imbalances via Secure Delegation Networks

Zhuolun Li, Evangelos Pournaras

原文摘要

Liquid democracy lets voters either vote directly or delegate their voting power to another trusted participant. It is often deployed in political parties, municipal participation platforms, and the governance of decentralized autonomous organizations. Existing approaches make delegations publicly visible as they form, which results in herding driven by popularity, makes coercion verifiable, and leaves the election fragile when popular delegates abstain. We propose a novel liquid democracy mechanism that removes these vulnerabilities, while keeping the tally fully auditable. Delegation choices are sealed with decentralized timed-release encryption, while they are formed and before revealed in full for tallying. Each voter determines ranked backup delegates together with a personal fallback ballot so that delegation failures do not result in abstaining. Experiments grounded in real-world voting data show that the mechanism keeps voting power dispersed, whereas transparent formation concentrates it into few hands. Strikingly, vote loss under targeted delegate failures falls from 26% to 3%.

cs.CR · cs.CY

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2608.14845

From Block Orthogonality to Decidability in Complex-Weighted Counting CSP

Chenghua Liu, Boning Meng

原文摘要

In a landmark JACM paper recognized with the 2021 Gödel Prize, Cai and Chen established a complexity dichotomy for counting CSPs over arbitrary finite domains with algebraic complex weights. Their tractability criterion consists of three conditions on the infinite family of functions obtained from CSP instances by summing out some variables: the Block Orthogonality and Type Partition conditions, and a Mal'tsev polymorphism condition on support and row-equivalence relations. They left open whether this criterion is decidable from the finite set of constraint functions. We resolve this question by giving a total exact algorithm for every $q$-element domain ($q\ge 1$), with algebraic weights encoded exactly. We also prove that the Block Orthogonality condition implies the other two conditions, reducing the tractability criterion to a single structural property. We obtain similar decidability and structural results for Lin's tractability criterion for $\#\mathrm{CSP}^{d}$, where each variable occurs a multiple of $d$ times. Our method first shows that every failure of Block Orthogonality can be witnessed by a generated function of arity at most $q^2+1$. At each such arity, we reduce checking infinitely many generated functions to finitely many exact algebraic tests, using tensor products and a pinned extension of Young's all-instance isomorphism theorem. For the structural implication, we use powers of generated functions to control complex cancellation and construct functions whose supports encode row equivalence. These constructions yield the Type Partition and Mal'tsev polymorphism conditions.

cs.CC

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09561

Lightweight Path-Aware Frontier Re-ranking for LiDAR-Based UAV Exploration

Duc Manh Do, Thanh Nguyen Canh, Ngoc Minh Do, Mai Thanh Thai, Xiem HoangVan

原文摘要

Autonomous 3D exploration with UAVs requires efficient target selection under limited flight time and onboard computation. In conventional Euclidean-distance-based frontier pipelines, geometric distance may poorly represent the actual collision-free travel cost in cluttered environments, leading to inefficient frontier choices. This paper presents a lightweight path-aware frontier re-ranking method for LiDAR-based UAV exploration. The method first retains a small top-N set of promising frontier candidates using a lightweight utility function and then applies time-bounded RRT* planning to estimate their feasible path costs for re-ranking. The path generated for the selected frontier is further reused for navigation to reduce redundant global planning. Experiments in three warehouse-like environments show that the proposed method consistently produces shorter exploration trajectories than conventional frontier baselines, with traveled-distance reductions of up to 14.1%, while reducing exploration time by up to 9.5% in two environments and remaining competitive in the third. The method maintains map detail comparable to the conventional OctoMap-based baseline, demonstrating an effective trade-off between exploration efficiency and mapping quality.

cs.RO

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09560

Solver-Free Decision-Focused Learning via Neural Optimization Layers for Edge Selection

Xinda Zheng, Frits de Nijs, Edward Lam

原文摘要

Decision-focused learning trains prediction models to improve downstream decisions, but many methods require costly solver calls or optimal-solution labels. We propose a solver-free decision-focused learning framework for NP-hard edge-selection problems with uncertain parameters in the objective, the constraints, or both. A neural optimization layer is first trained without solution labels, using problem objectives and constraint penalties. Graph-informed edge scoring and iterative refinement encourage concentrated and structurally consistent continuous decisions. The layer is then frozen, and a prediction model is trained through its continuous output using a problem-specific loss evaluated under observed parameters. At deployment, the predictor, optimization layer, and a lightweight decoder produce decisions without solving the original NP-hard problem. Experiments on the asymmetric traveling salesperson and endpoint-free orienteering problems show consistently better decision quality than prediction-focused learning and improvements over most evaluated decision-focused baselines. Our framework requires less training time than the evaluated decision-focused baselines, even when optimization-layer pretraining is included. Neural deployment also produces decisions faster than solver-based deployment, offering a better quality-efficiency trade-off. The layer also reduces training cost in existing decision-focused methods when used as an approximate optimization oracle.

cs.LG · math.OC

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.09558

Optimal Regret for Profit Maximization in Bilateral Trade

Matteo Castiglioni, Anna Lunghi, Alberto Marchesi

原文摘要

Learning to profit from trade requires learning both sides of a market: what a seller is willing to accept and what a buyer is willing to pay. We study repeated bilateral trade, where an intermediary posts a purchase price to a seller and a sale price to a buyer whose valuations are drawn from an unknown joint distribution. Trade occurs only if both parties accept their respective prices, in which case the intermediary earns the difference between the two prices. The intermediary observes only whether trade occurs, without observing either valuation. We show that learning both sides of the market is no harder, in terms of regret, than learning just one. Specifically, we design an algorithm achieving regret $\widetilde{\mathcal{O}}(T^{2/3})$ against the best fixed pair of prices, matching, up to logarithmic factors, the classical lower bound for one-dimensional dynamic pricing. Our result shows that the classical one-dimensional regret rate survives the transition to two dimensions, raising the question of whether this absence of a curse of dimensionality extends to arbitrarily high dimensions.

cs.GT · cs.LG

查看论文
arXiv ↗PDF ↗