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

+479新增论文
+361其中 AI 相关
+63已有论文修订
+93新增 AI 导读

arXiv 最近完成扫描:2026-11-16 01:34 · 全站最近入库:2026-11-16 13:44各信源进度 →

统计口径与数据新鲜度

统计 2026-11-15 20:26 — 2026-11-16 20:26(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:2026-11-16 01:34最近发现新增版本:2026-11-16 13:44库内论文最新更新时间:2026-11-14 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。

23,827 已收录论文19,941 AI 相关6,921 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 17,708 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-16 01:33(北京时间)
新论文 · v1arXiv:2611.16345

DRAMA: Dissecting Attention Redundancy for Accelerating Multimodal Diffusion Large Language Models

Yi-Hsin Hung, Fangfu Liu, Shao-Yuan Lo, Chu-Song Chen

原文摘要

Multimodal diffusion large language models (dLLMs) generate text through iterative masked denoising with bidirectional attention, offering a compelling alternative to autoregressive vision-language modeling. However, each denoising step requires a full forward pass over long multimodal sequences in which visual tokens dominate the sequence length, and a fixed step budget further compounds this quadratic per-step cost. Existing inference acceleration approaches for multimodal dLLMs characterize attention redundancy implicitly, motivating designs from isolated attention visualizations rather than from measurement. We close this gap with DRAMA, a training-free acceleration framework grounded in a systematic dissection of attention redundancy in multimodal dLLMs. By probing attention tensors along spatial and temporal axes, we reveal two persistent redundancy structures: (i) attention mass consistently concentrates on a sparse subset of visual tokens, and (ii) attention distributions rapidly stabilize across denoising steps after an initial formation phase. These findings motivate two complementary mechanisms: Spatial-Redundancy-Aware Visual Token Pruning (SVTP) and Sufficiency-Guided Adaptive Stopping (SGAS). Together, DRAMA achieves an average 18.67 times inference speedup across six vision-language benchmarks without training, architectural modification, or attention caching, while maintaining competitive accuracy. Our results establish a strong, measurement-grounded acceleration baseline for multimodal dLLM inference.

cs.LG · cs.AI

版本更新 · v2arXiv:2611.12356

DForkTree: Accelerating Speculative Decoding with Fork-Aware Block Diffusion Draft Trees

Zhicheng Zheng, Cheng Zhang, Chenkai Tang, Tianxiang Zhao

原文摘要

Speculative decoding speeds up language models by having a lightweight drafter propose future tokens that the target model verifies in parallel. Block diffusion drafters such as DFlash predict an entire block in one forward pass, and DDTree builds draft trees directly from their per-position distributions. Each distribution implicitly assumes that the drafter's own greedy continuation is accepted, whereas a draft tree improves over a chain only on branches that leave this greedy path. Once a path forks, the drafter's top-1 accuracy drops from 89% to 18% and its top candidates often miss the true token. Re-running the drafter from the fork token largely restores its accuracy, showing that the errors stem from conditioning on an outdated prefix rather than from limited model capacity. In this paper, we introduce DForkTree, a training-free method that re-anchors each branch on its own tokens. It adjusts the drafter's scores with pointwise mutual information from target-generated n-grams, weighted more heavily after a fork, and a layer-wise GPU search recovers the exact optimal tree under these scores. On Qwen3.5-4B and 9B, DForkTree outperforms all baselines on every benchmark, by up to 41.5% over DFlash and 15.1% over DDTree, with average speedups of up to 4.75x over autoregressive decoding.

cs.AI

新论文 · v1arXiv:2611.16341

SUMNO: Spatially Unrolled Multigrid Neural Operator for Fluid Flow Surrogates

Yingqi Hu, Nan Deng, Changxiao Shao, Xiaozhou He

原文摘要

Data-driven surrogate modeling has emerged as a powerful paradigm for accelerating computational fluid dynamics, and hierarchical U-shaped networks, which resemble the structure of classical Geometric Multigrid (GMG) methods, have become prevalent designs. Modern surrogates built on these designs achieve competitive accuracy at a fraction of the cost of numerical solvers; however, accurate prediction of complex flows remains challenging, and no existing architecture leads across them. In such networks, the multigrid analogy mainly concerns the multi-resolution topology, while their components largely follow common vision practice. To address this, this study introduces SUMNO, a multigrid neural operator for fluid flow fields on regular grids that spatially unrolls a V-cycle into a feed-forward U-shaped network. Parameter-free down- and upsampling stand in for restriction and prolongation, and additive skip connections for coarse-grid correction, with normalization-free transfer and skip pathways. Within this macro-architecture, convolutions are used at the fine and intermediate resolutions and shifted-window attention at the coarsest, whose window partitioning is generalized to handle periodicity independently per axis. Evaluated on unseen test configurations across four unsteady and steady flow benchmarks against five representative baselines, SUMNO is best in 5 of 7 settings and ranks second in the remaining two. It reduces the 10-step rollout MSE by about 30% relative to the strongest baseline, and outperforms three baselines with over 130 million parameters using 9.4 million parameters.

physics.flu-dyn · cs.AI

新论文 · v1arXiv:2611.16337

Memory-Efficient Parallel Policy Learning for Collapse-Reactive Retreat via Hybrid Terrain Modeling

Yuki Kadokawa, Ritsuki Matsunaga, Yuqi Yang, Edgar Anarossi, Takamitsu Matsubara

原文摘要

At disaster sites and on unstable slopes, robot contact can collapse uninspected terrain, requiring retreat upon support loss. Massively parallel reinforcement learning of collapse-reactive retreat requires independent state--action experiences. However, on shared rigid-body collapsible terrain, one robot's collapse alters others' contacts without direct interaction. We call this indirect interference. Duplicating terrain objects prevents this but is memory-intensive, limiting parallel sampling. Our key insight is that policy learning does not necessarily require rigid-body collapse modeling, which is feasible but computationally expensive; collapse-induced contact responses can suffice. We therefore propose a memory-efficient parallel policy-learning framework for collapse-reactive retreat via hybrid terrain modeling. Hybrid terrain modeling combines rigid-body simulation of non-collapsible terrain with virtual contact forces for collapsible regions. A State-Multiplexed Contact Model shares terrain geometry and a contact-response law while maintaining environment-specific collapse states, enabling independent experiences without terrain objects for collapsible regions. The framework trains a proprioceptive policy for both normal locomotion and collapse-reactive retreat. It supports over 100 times as many parallel environments as fine-resolution rigid-block terrain within the same GPU memory limit. This scalability and improved sample efficiency enable over 50 times faster policy convergence with comparable retreat performance. Sim-to-sim and real-robot experiments demonstrate retreat under multiple collapse conditions. A supplementary video is available at: https://youtu.be/_RrhwjzEPE0

cs.RO

新论文 · v1arXiv:2611.16336

Visibility graphs of irregularly sampled two-dimensional fields via Voronoi tessellations

Francisco J. Muñoz, Juan Carlos Nuño

原文摘要

Visibility graphs map a time series into a network by linking two data whenever the straight line between them is not obstructed by intermediate data. Their extension to two-dimensional fields has so far been restricted to regular lattices, where visibility is checked along a finite number of lattice directions. We propose a definition for scalar data sampled at arbitrary, irregularly scattered points of the plane. Each point is assigned its Voronoi cell, so that the data become a landscape of prisms, and two points are linked when the line of sight between them clears the cells crossed by the connecting segment. This yields horizontal ($\HVGV$) and natural ($\NVGV$) Voronoi visibility graphs that reduce exactly to the classical horizontal and natural visibility graphs when the points are collinear, contain the Gabriel graph of the point set, and can be built by an exact walk through the Delaunay triangulation. For uncorrelated heights, the horizontal graph depends only on the ranks of the data, so its statistics are independent of the marginal distribution. The probability that two points see each other is $2/[(M+1)(M+2)]$, where $M$ is the number of Voronoi cells crossed, which reduces the problem to the geometry of the tessellation. For Poisson points this gives a degree distribution with a scale-free tail $P(k)\simeq(π^3/4)\,k^{-2}$ and a mean degree that grows as $(π^3/8)\ln N$, in contrast with the finite value $\langle k\rangle=4$ of one-dimensional series. Two dimensions are therefore the marginal case in which isotropic horizontal visibility ceases to be local. Numerical simulations with up to $N=1.28\times10^5$ points support these predictions.

math.DS · cond-mat.stat-mech · cs.CG · physics.data-an

新论文 · v1arXiv:2611.16333

When Elements Fail: Weighted Fault-Tolerant Matroid Bases

Madhumita Kundu, Soumi Nandi, Saket Saurabh

原文摘要

Given a rank-\(r\) matroid on \(n\) elements with nonnegative costs and a partition into safe and vulnerable elements, we seek a minimum-cost set that spans after every deletion of at most \(k\) selected vulnerable elements. We resolve several open questions posed by Bentert, Fomin, Golovach, and Morelle (ESA 2025) concerning weighted costs, nonuniform failures, and explicitly specified failure scenarios. To this end, we develop two compression methods. For matroids given by an independence oracle, weighted flat-avoidance representatives yield a bounded subset of the ground set containing an optimal solution and a \(2^{\mathcal{O}((r+k)^4)}n^{\mathcal{O}(1)}\)-time algorithm. This establishes fixed-parameter tractability for weighted costs and nonuniform failures simultaneously. For explicitly represented linear matroids, polynomial certificates allow us to compress families of partial solutions while preserving successful completions under all allowed failures. This yields a deterministic algorithm using \(2^{\mathcal{O}(kr(r+k)\log(r+1))}n^{\mathcal{O}(1)}\) field operations over arbitrary fields. Over fixed finite fields, a simpler reduction uses parallel classes. For explicitly listed failure scenarios, we obtain fixed-parameter algorithms parameterized by rank and the number of scenarios. In contrast, the size-budget decision version is W[1]-hard parameterized by rank and maximum scenario size, even on uniform matroids with two-element scenarios. \keywords{Matroids \and Fault tolerance \and Fixed-parameter algorithms \and Representative families}

cs.DS · cs.DM

新论文 · v1arXiv:2611.16330

Snoopie: A Multi-GPU Communication Profiler and Visualizer

Mohammad Kefah Taha Issa, Muhammad Aditya Sasongko, Ilyas Turimbetov, Javid Baydamirli, Doğan Sağbili, Didem Unat

原文摘要

With data movement becoming one of the most expensive bottlenecks in computing, the need for profiling tools to analyze communication becomes crucial for effectively scaling multi-GPU applications. While existing profiling tools including first-party software by GPU vendors are robust and excel at capturing compute operations within a single GPU, support for monitoring GPU-GPU data transfers and calls issued by communication libraries is currently inadequate. To fill these gaps, we introduce Snoopie, an instrumentation-based multi-GPU communication profiling tool built on NVBit, capable of tracking peer-to-peer transfers and GPU-centric communication library calls. To increase programmer productivity, Snoopie can attribute data movement to the source code line and the data objects involved. It comes with multiple visualization modes at varying granularities, from a coarse view of the data movement in the system as a whole to specific instructions and addresses. Our case studies demonstrate Snoopie's effectiveness in monitoring data movement, locating performance bugs in applications, and understanding concrete data transfers abstracted beneath communication libraries. The tool is publicly available at https://github.com/ParCoreLab/snoopie.

cs.DC · cs.PF

版本更新 · v3arXiv:2603.02184

MAC: A Conversion Rate Prediction Benchmark Featuring Labels Under Multiple Attribution Mechanisms

Jinqi Wu, Sishuo Chen, Zhangming Chan, Yong Bai, Lei Zhang, Sheng Chen, Chenghuan Hou, Xiang-Rong Sheng 等 12 位作者

原文摘要

Multi-attribution learning (MAL), which enhances model performance by learning from conversion labels yielded by multiple attribution mechanisms, has emerged as a promising learning paradigm for conversion rate (CVR) prediction. However, the conversion labels in public CVR datasets are generated by a single attribution mechanism, hindering the development of MAL approaches. To address this data gap, we establish the Multi-Attribution Benchmark (MAC), the first public CVR dataset featuring labels from multiple attribution mechanisms. Besides, to promote reproducible research on MAL, we develop PyMAL, an open-source library covering a wide array of baseline methods. We conduct comprehensive experimental analyses on MAC and reveal three key insights: (1) MAL brings consistent performance gains across different attribution settings, especially for users featuring long conversion paths. (2) The performance growth scales up with objective complexity in most settings; however, when predicting first-click conversion targets, simply adding auxiliary objectives is counterproductive, underscoring the necessity of careful selection of auxiliary objectives. (3) Two architectural design principles are paramount: first, to fully learn the multi-attribution knowledge, and second, to fully leverage this knowledge to serve the main task. Motivated by these findings, we propose Mixture of Asymmetric Experts (MoAE), an effective MAL approach incorporating multi-attribution knowledge learning and main task-centric knowledge utilization. Experiments on MAC show that MoAE substantially surpasses the existing state-of-the-art MAL method. We believe that our benchmark and insights will foster future research in the MAL field. Our MAC benchmark and the PyMAL algorithm library are publicly available at https://github.com/alimama-tech/PyMAL.

cs.LG · cs.AI

版本更新 · v2arXiv:2602.21682

SunnyParking: Multi-Shot Trajectory Generation and Motion State Awareness for Human-like Parking

Jishu Miao, Han Chen, Jiankun Zhai, Qi Liu, Tsubasa Hirakawa, Takayoshi Yamashita, Hironobu Fujiyoshi

原文摘要

Autonomous parking fundamentally differs from on-road driving due to its frequent direction changes and complex maneuvering requirements. However, existing End-to-End (E2E) planning methods often simplify the parking task into a geometric path regression problem, neglecting explicit modeling of the vehicle's kinematic state. This "dimensionality deficiency" easily leads to physically infeasible trajectories and deviates from real human driving behavior, particularly at critical gear-shift points in multi-shot parking scenarios. In this paper, we propose SunnyParking, a novel dual-branch E2E architecture that achieves motion state awareness by jointly predicting spatial trajectories and discrete motion state sequences (e.g., forward/reverse). Additionally, we introduce a Fourier feature-based representation of target parking slots to overcome the resolution limitations of traditional bird's-eye view (BEV) approaches, enabling high-precision target interactions. Experimental results demonstrate that our framework generates more robust and human-like trajectories in complex multi-shot parking scenarios, while significantly improving gear-shift point localization accuracy compared to state-of-the-art methods. We open-source a new parking dataset of the CARLA simulator, specifically designed to evaluate full prediction capabilities under complex maneuvers.

cs.RO

新论文 · v1arXiv:2611.16327

Designing Information in a Forest: Algorithmic Aspects

Konstantin Zabarnyi

原文摘要

The Bayesian persuasion framework models strategic information revelation by a sender to one or more receivers. We study an intricate version of the multi-receiver persuasion setting in which the communication hierarchy is a forest - i.e., some receivers observe more informative signals than others, and the partial ordering of the receivers according to the information they get is represented by a pre-specified directed forest. This model naturally captures interactions in hierarchical organizations, and it lies in between the two most commonly studied versions of Bayesian persuasion: private persuasion and public persuasion. Inspired by the classic model of private persuasion, we focus on a setting with a binary state of nature and binary-action heterogeneous receivers s.t. the sender's utility is a monotone function of the set of receivers taking a specific desirable action. We extend the computational tractability result for symmetric sender's utility from private persuasion to the forest hierarchy framework using methodology inspired by optimal transport theory. We further provide a polynomial-time algorithm when the sender's utility is modular. On the negative side, we show that, unlike in private persuasion, the case of a supermodular sender's utility function leads to an NP-hard optimization problem which further does not admit an FPTAS unless P=NP; this result holds even when each tree component of the forest is a chain of 3 receivers. To complement the hardness result, we provide a simple approximation algorithm with the approximation ratio depending linearly on the depth of the forest hierarchy. We further provide an analogous algorithm for the submodular utility case. Finally, beyond forest communication hierarchies, we rule out an FPTAS for the sender's optimal utility even in the symmetric utility case, unless P=NP.

cs.GT · econ.TH

版本更新 · v2arXiv:2611.11698

Compute Regeneration through Re-Productization: Reorganizing Retired Edge AI Silicon into a Delivered Multilingual Service Robot

Hongyu Luo, Sun Tie, John Wang

原文摘要

Useful edge hardware can lose product support while retaining substantial computation. We treat this as systems design: compute regeneration here means re-productization, retaining the compute module while redesigning the surrounding product organization. Our case is GEdgeBot, a multilingual service robot built around a retired 32 GB Jetson AGX Xavier. A source-grounded inventory shows the box alone supplies inference, while the combined system exposes five on-device AI services, a seven-language speech chain, three sensing domains, and one deterministic seven-byte control boundary. AI and orchestration stay on the Xavier; finite motor-pulse generation on an STM32F407. Audits cover the six-mode control state machine, readiness barriers and sound orientation. Historical single-prototype measurements report 1,782 MB GPU memory with YOLO, Whisper-small, and NLLB-200 resident, a direct-control p99 deviation rising from 0.24 to 49.93 ms under concurrent AI load, and six of six trigger-to-follow demonstrations at 4.35 s mean latency, with conversational voice unavailable. A November campaign restores an isolated inference stack, re-measures AI workloads, and repeats control-frame runs. Driving two execution organisations with the same 6,000 measured 50 ms frame intervals shows the finite-pulse boundary removes host jitter from the pulse train but not the command start phase; its stopping bound is host-enforced, while the MCU 16-bit pulse field is unclamped. A six-check static audit exposes a host--MCU queue mismatch, and a 14-scenario replay executes the audited main() dispatcher across six modes with zero physical device opens and 54 contract-checked frames. Results support an existence claim and a reproducible architectural hypothesis, not fleet-level reliability. The 208 direct-control ratio remains unreplicated; MCU waveform capture and on-site fault injection remain open.

cs.RO · cs.AR

新论文 · v1arXiv:2611.16323

Does Deep Knowledge Tracing Learn Semantic Relations Between Knowledge Components?

Kia Karbasi, Mahdi Saiedi, Kevin Hong, Gregory Pottie

原文摘要

Deep Knowledge Tracing (DKT) is a knowledge tracing (KT) model that updates a shared latent state from sequences of interactions, each pairing a knowledge component (KC) with a response. An interaction with one KC can therefore affect predictions for other KCs. We ask: Are the KC relations learned by DKT semantic? We construct text-derived semantic KC representations using nine pipelines to test whether the conclusions hold across alternative ways of constructing those representations. The first test compares DKT-derived KC relation graphs with semantic relation structures induced by pairwise cosine similarities among the KC representations. Across the evaluated datasets, embedding models, and pipelines, the two structures align only weakly and inconsistently. The second test introduces semantic information into DKT; the pretrained text-derived representations do not consistently outperform the standard DKT baseline. In the reduced-data experiments, semantic initialization yields a higher test area under the ROC curve (AUC) than the baseline at eight of nine nested sample sizes for ASSISTments 2017. KC-supervised fine-tuned BERT produces small positive AUC differences in some matched comparisons, but the study does not establish their stability across runs. These effects remain limited and inconsistent. The findings suggest that behavioral, temporal, structural, or psychometric signals may also shape DKT's learned relations. DKT-derived relations are not primarily semantic: the tested text-derived semantic structures do not strongly recover the DKT-derived relation structure.

cs.LG · cs.CL

新论文 · v1arXiv:2611.16322

Climate Temporal Temperature Prediction via an Adaptive Stacked Bidirectional Residual Mamba Network

Yongxiang Lei

原文摘要

Accurate temperatures forecasting relies on traditional meteorological parameters that are essential for monitoring weather informatics and guiding forecasting efforts. This study presents Deep DualMamba, a novel deep learning framework for temporal temperature prediction, built upon an enhanced variant of the recently proposed Mamba network. Unlike conventional Transformer-based architectures, which dominate foundation models but suffer from computational inefficiencies on long sequences, Deep DualMamba leverages selective structured state space models (SSMs) to achieve both scalability and performance. We identify a core limitation of prior subquadratic-time models-namely, their inability to perform long-term and robust forecasting and introduce two key innovations. First, we propose an adaptive bidirectional Mamba as encoders to learn complex and efficient features. Second, we develop a hardware-aware, parallel recurrent algorithm to compensate for the loss of convolutional efficiency, ensuring practical viability at scale. Last, the Deep DualMamba's approximation ability and convergence is theoretical analyzed. These design choices culminate in a simplified, attention- and MLP-free neural architecture that exhibits linear complexity with sequence length and significantly improved inference throughput. To demonstrate its applicability in climate informatics, we implement a full forecasting pipeline using historical meteorological data from UK monitoring stations.

cs.LG

新论文 · v1arXiv:2611.16321

From Large Language Models to Efficient System One Models via Adaptive Computation

Minh-Anh Nguyen

原文摘要

Transferring LLMs into System One decision models typically replaces the language modeling head with a decision head while retaining the full backbone, making every decision pay for all layers. Yet it remains unclear how much depth such decisions require. We find that decision information is concentrated in intermediate layers, with early layers building essential features and the final six layers removable without loss, while independently trained smaller models largely mimic the full model rather than developing specialized decision capabilities. These findings suggest that efficiency requires not simply smaller backbones, but shared computation across different depths. Building on this, we propose \method{}, which jointly trains a full model with prefix sub-models that share its final layers and decision head through self-distillation. A lightweight gate reads the hidden state at each exit to determine whether additional computation is needed, allowing one model to operate at multiple layers. Across seven decision benchmarks, our approach matches or outperforms independently trained models at every depth, while achieving better accuracy-efficiency trade-offs through adaptive computation.

cs.CL

版本更新 · v4arXiv:2606.03159

NVIDIA OmniDreams: Real-Time Generative World Model for Closed-Loop Autonomous Vehicle Simulation

Aarti Basant, Amlan Kar, Despoina Paschalidou, Fangyin Wei, Francesco Ferroni, Guillermo Garcia Cobo, Haithem Turki, Huan Ling 等 33 位作者

原文摘要

As autonomous vehicle capabilities advance, the safe evaluation of driving policies in long-tail scenarios remains a critical bottleneck. In closed-loop simulation, the driving policy model actively interacts with the environment, where its actions dynamically update the simulator state and directly influence the next set of generated sensor observations. While recent reconstruction-based neural simulators offer photorealism, they are fundamentally constrained by their initial captured data and struggle to generalize to highly dynamic or novel scenes. To overcome these limitations, we introduce OmniDreams, a foundation generative world model mid- and post-trained from the Cosmos diffusion model to autoregressively generate action-conditioned videos in real time. By leveraging the rich visual priors of Cosmos and mid- and post-training on 21k hours of driving scenarios, OmniDreams synthesizes complex, unobserved phenomena that are hard for traditional simulators to capture, such as extreme weather and unpredictable dynamic agent behaviors. Crucially, it autoregressively conditions its photorealistic sensor generation on past frames, the current simulator state, and immediate driving actions. Deployed in a closed-loop system with the Alpamayo 1 policy model and AlpaSim orchestrator, OmniDreams acts as a highly responsive, reactive environment, providing a scalable and comprehensive solution for training and evaluating next-generation autonomous driving policies. We additionally show preliminary results indicating that a world-action model (WAM) post-trained from OmniDreams achieves strong performance on the Physical AI Autonomous Vehicles NuRec dataset, surpassing the VLA-based Alpamayo 1.5 research policy model while using only 1/5 the total parameters. These results highlight the potential for a real-time world model like OmniDreams to also serve as a backbone for policy architectures.

cs.CV · cs.AI · cs.RO

新论文 · v1arXiv:2611.16317

Wanda-Guided Sparse Perturbations for Efficient Zeroth-Order Optimization

Ziqing Wen, Xiaohui Song, Jiahuan Wang, Ping Luo, Junlin Zeng, Dongsheng Li, Tao Sun

原文摘要

Zeroth-order (ZO) optimization enables memory-efficient fine-tuning of large language models through forward-only evaluations, but high-dimensional random perturbations can limit optimization efficiency. Sparse perturbations reduce the search dimension, making the selection of informative coordinates critical. We propose Wanda-Importance Sparsity-MeZO (WIS-MeZO), which uses Wanda weight--activation saliency to construct task-conditioned sparse perturbation masks while keeping the pretrained model dense. We analyze support quality through gradient energy capture and relate it to gradient estimation and convergence. Across 16 benchmarks with LLaMA and Qwen models ranging from 3B to 32B parameters, WIS-MeZO achieves higher aggregate scores than MeZO and reaches comparable validation losses in fewer optimization steps. Comparisons with random and magnitude-based sparse perturbations further highlight the importance of informative support selection.

cs.LG · cs.AI

新论文 · v1arXiv:2611.16315

Shared Item-Level Error Limits the Aggregation of Small Language Models in Fermi Estimation

Carolina Fortuna, Blaz Bertalanic

原文摘要

Multi-agent systems of language models are often motivated by an averaging argument: if the agents' errors are unbiased and independent, the error of an aggregate of N answers falls as 1/sqrt(N). Fermi estimation, whose continuous answers are aggregated on the logarithmic scale, permits a direct test. We decompose the log10 error of each answer into the mean error of its model on the item, common to all its draws on the item, and a residual, and estimate both from 2.51M answers given before any communication by 15 open-weight models (3B-72B) on realFP, with replication under a single prompt pair on realFP, Science Olympiad Fermi questions and a guesstimation set. Given the item, teammates' errors are uncorrelated and teams behave as independent draws, but the item-level component accounts for a median 0.906 of the error variance over items. Teams of thirty therefore reduce a single agent's error by a median 1.7% (at most 5.4%), whereas draws that were also unbiased on each item would reduce it by 74-79%. All models compress their estimates toward about 10^6; this trend explains a median 17% of the item-level component, and a cross-validated linear recalibration on labelled items removes the overall bias but not that component. Three rounds of debate reduce within-team dispersion by 93% and leave the item-level component unchanged; revision without peers attains the same small gain. Mixing five models of similar accuracy reduces error by 6.6% relative to their own homogeneous teams, but no mixture improves on the most accurate model. Higher sampling temperature changes team error by about 1% at most, and contrastive activation steering not detectably; per-agent scaffold prompts (not significant after multiplicity correction) and skipping one layer in a single-model screen improve it. The results are conditional on the models, benchmarks, prompts and settings tested.

cs.MA · cs.AI

版本更新 · v2arXiv:2609.24687

Discrete Isoperimetric Inequalities via Curvature

Zejia Chen, Zongchen Chen, Xinyuan Zhang

原文摘要

Isoperimetric inequalities on the Boolean hypercube play a fundamental role in the analysis of Boolean functions. These inequalities have been established primarily for the uniform measure and for biased product measures, often through Fourier-analytic or inductive arguments. We develop a curvature-based semigroup framework to establish isoperimetric inequalities for measures with weakly dependent coordinates. In particular, we show that a Dobrushin-type condition, together with marginal boundedness of the distribution, implies the local Bobkov inequality, Talagrand's $L^1$--$L^2$ and variance--surface-area inequalities, the Kahn--Kalai--Linial inequality, and the Eldan--Gross inequality. Our results apply to zero-field Ising models with interaction matrix $J$ throughout the Dobrushin uniqueness regime $\|J\|_1<1$. The framework builds on discrete Bakry--Émery theory and gradient estimates and can also be extended to measures on Hamming slices or hypergrids.

math.PR · cs.DM

新论文 · v1arXiv:2611.16313

Optimal-Size Strong Connectivity Preservers under Two Edge Failures in Subcubic Time

Wanhee Lee, Chenglin Fan

原文摘要

A $k$-edge-failure strong connectivity preserver of a directed graph on $N$ vertices is a spanning subgraph that has the same strongly connected components as the input after each set of at most $k$ edge deletions. The lower-bound construction of Baswana, Choudhary, and Roditty [BCR18], adapted to strong connectivity in [CC20], gives the previous best bound of $Ω(2^kN)$ necessary edges. Hoppenworth, Saranurak, and Wang [HSW25] prove an $O(k4^kN\log N)$ upper bound, leaving a logarithmic gap for fixed $k$. We close this gap by giving an explicit strongly connected graph on $N=2^d$ vertices with $N-1+(N/2)\log_2N$ edges, every one of which is necessary in a two-failure preserver. Together with the upper bound of [HSW25], this proves that the optimal size in the worst case is $Θ_k(N\log N)$ for every fixed $k\geq2$, whereas it is $Θ(N)$ for $k=1$ [GIP20]. The same graph has a one-failure preserver with at most $3(N-1)$ edges. We also give a randomized algorithm that either returns a spanning subgraph or reports failure. For every fixed $c,\varepsilon>0$, with probability at least $1-N^{-c}$ it returns a correct preserver for $k=2$ with $O(N\log N)$ edges, in $\widetilde O(N^{(ω+3)/2+\varepsilon})$ time and $\widetilde O(N^2)$ space. Here $ω$ is the matrix multiplication exponent over a fixed finite field.

cs.DS

版本更新 · v2arXiv:2609.35036

Persona Following Is Not Selective Control: The Neutrality Gap in LLM User Simulation

Jiashen Ren, Wenlin Zhang, Bohan Zhang, Xiaopeng Li, Zichuan Fu, Wanyu Wang, Junyi Li, Xiangyu Zhao

原文摘要

Persona prompting is widely used in LLM user simulation, yet it relies on an untested assumption: specifying a target user attribute should shift target behavior while leaving other fixed attributes unchanged. Across eight black-box and five open-weight models, however, changing a target attribute alone also shifts unspecified non-target choices, such as risk preference shifting color choice. We test whether explicitly specifying non-target attributes restores selective control across 477 held-out synthetic worlds and 200 independent items under unspecified, directional, and neutral states. Assigning a clear non-target direction nearly eliminates cross-attribute shifts. Explicitly declaring no preference, however, still leaves 37.7%-76.3% of held-out worlds and 51%-81% of independent non-target items shifting with the target attribute while preserving target efficacy. We call this disparity the neutrality gap. Because directional tests can be passed simply by following a stated preference, the neutral state reveals selective-control failures they miss: following a stated persona does not imply keeping unrelated choices invariant. In simulated investment decisions, incidental brochure color labels shift estimated target risk treatment effects by 25.5 percentage points, even when models accurately report having no color preference. Counterfactual attention interventions show that unwanted influence enters primarily at query time, when option evaluation reads the target persona directly. Blocking query-time attention cuts non-target drift by 72.1%-87.8%, whereas directional declarations suppress drift primarily by overwriting option values rather than rerouting attention. Because target and non-target choices share query-time access, blocking it weakens target efficacy while value overwrite imposes an artificial direction, leaving task-dependent selective access as a central challenge.

cs.AI

新论文 · v1arXiv:2611.16312

Alloy: A Multi-Material Physics Engine for Robot Learning

Pranay Gupta, Deepanshu Rohilla

原文摘要

Robots handle much more than rigid objects: towels, cables, water, sand. The simulators used to train robots at scale are strongest on rigid bodies, and adding another material usually means a second engine and hand-written glue between the two. We present Alloy, an open-source physics engine for robot learning in which rigid robots, cloth, cables, liquids, granular material and soft bodies share one scene. Each object is described once; Alloy assigns it to a solver suited to its material, couples the solvers inside every control step, and runs the scene in many parallel worlds on one GPU, or on a CPU. Every world carries a health grade, to help tell a failure of the physics from a failure of the policy. Alloy is built on Newton, MuJoCo Warp and Warp, and it is written to be driven by coding agents: its documentation ships inside the package, every error names its fix, and the engine is exposed as tools over the Model Context Protocol. In a cold test, agents that had never seen Alloy installed a pre-release from the package index and trained policies for walking, cable handling and pouring on a single GPU. We also report how throughput grows with the number of worlds, depth-camera throughput, agreement with closed-form references, and a reliability test over a thousand worlds.

cs.RO

新论文 · v1arXiv:2611.16307

DEDCA: Test-Time Adaptation for Generalized AI-Generated Image Detection

Hao Tan, Qimin Zhang, Lihuang Fang, Kebing Jin, Jinghui Qin

原文摘要

AI-generated image detectors face significant challenges when deployed in real-world environments, particularly when test samples are produced by unseen generators that deviate from the source training distribution. Existing detectors often rely on a single type of evidence, such as semantic representations or low-level generation traces, making them vulnerable when the corresponding cue becomes unreliable. Moreover, adapting detectors to newly emerging generators usually requires target labels or retraining, which is costly and impractical. To address these challenges, we propose Dual-Evidence Disagreement-Constrained Adaptation (DEDCA), a test-time adaptation framework for generalized AI-generated image detection. DEDCA constructs a dual-evidence detector by combining CLIP-based semantic evidence with trace evidence extracted from shuffled entropy and residual statistics, and introduces a reliability-aware gate to perform sample-specific adaptive fusion. Our key idea is to exploit the disagreement between the two evidence streams as an unlabeled adaptation signal, enabling the detector to adjust to target-domain shifts without relying on target labels or blindly trusting a single prediction stream. Experiments show that DEDCA achieves superior performance on the GenImage benchmark by outperforming state-of-the-art AI-generated image detectors and test-time adaptation methods with 92.0\% average accuracy.

cs.CV

新论文 · v1arXiv:2611.16304

StateAlign: Internalizing State-Aware Decision-Making for Reliable Multi-Turn Tool Agents via Reinforcement Learning

Zhexi Lu, Subhajit Chaudhury, Tejaswini Pedapati, Keerthiram Murugesan, Lei Yu

原文摘要

Reinforcement learning has improved the tool-use capabilities of large language models. However, better task performance can coincide with increased tool hallucination. Reliable tool use requires decisions to follow the current task state: available information, tool capabilities, and task progress. Our analysis shows that agents sometimes execute when they should ask for information, retrieve it, or decline an infeasible request, even when reasoning identifies missing information. We introduce StateAlign, combining task-state construction with state-contrastive reinforcement learning. The data cover five task conditions and pair states with different action requirements. Outcome rewards guide task completion; token-level contrasts compare the same generated reasoning and actions under paired states to train outputs that fit the current state. Experiments across five base models from three families show improvements on BFCL v4 multi-turn, ToolSandbox, and When2Call. Ablations show that state construction improves performance and reliability under different reward designs. The complete method further reduces tool hallucination across three Qwen3 sizes. On Qwen3-8B, StateAlign improves BFCL v4 multi-turn accuracy from 45.5% to 55.2% and reduces no-tool hallucination on SimpleToolHalluBench from 62.2% under outcome-only training to 56.1%. Behavioral analyses show more consistent decisions, more complete multi-call execution, and stronger recovery after tool errors.

cs.LG

版本更新 · v2arXiv:2609.24969

Rare Event Estimation via Iterative Unalignment

Hanming Yang, Daksh Mittal, Jing Dong, Hongseok Namkoong

原文摘要

As agents are deployed with increased autonomy, even extremely rare events along their stochastic output trajectories can occur and prove catastrophic. Safe deployment therefore does not depend on whether these events can occur, but on how often they might. We study the problem of estimating the probability of rare events that arise from stochastic variation in the agent's own actions. Estimating this type of risk requires searching over the combinatorially vast space of trajectories. Naive Monte Carlo is computationally prohibitive in this regime, and constructing effective importance sampling (IS) proposals requires coordinated changes to a context-dependent chain of conditional distributions. We develop a new IS method that perturbs the original model's weights to construct the proposal. The proposal is itself a differentiably parameterized language model, enabling gradient-based search over weight space. We formulate an objective that combines a differentiable surrogate for event amplification and an adaptive regularization scheme that dynamically balances amplification against estimator stability. We evaluate our approach on $\sim$120M and $\sim$2.6B models across three event families spanning 300+ rare events as rare as $10^{-9}$, all with independently estimated reference probabilities. In our most verifiable settings, we observe that our IS estimator achieves over $800\times$ compute-weighted efficiency gains on average over naive Monte Carlo for events with probabilities lower than $10^{-7}$. Our implementation is available at https://github.com/namkoong-lab/iterative-unalignment.

cs.LG · cs.AI

版本更新 · v2arXiv:2611.13987

Communication-Learning Co-Design for Pinching-Antenna-Assisted Federated Learning in Heterogeneous Wireless Networks

Shuai Wang, Wenxuan Sun, Mingjie Shao, Wantao Jiang, Yanqing Xu, Lingxiang Li, Zhi Chen

原文摘要

Wireless federated learning (FL) suffers from two coupled bottlenecks: statistical heterogeneity across clients and communication-constrained model synchronization over unreliable wireless links. Instead of focusing on algorithmic design and/or wireless resource optimization, this paper provides a systematic study of pinching antenna (PA)-empowered FL for heterogeneous wireless networks and develops a unified communication-and-learning framework. Specifically, we integrate a segmented waveguide-enabled pinching antenna (SWAN) architecture with a heterogeneity-robust FL algorithm based on client variance reduction and dual quantization, so as to jointly mitigate communication bottlenecks and client drift. On this basis, we further develop a two-stage SWAN-assisted downlink protocol and formulate fairness-oriented uplink optimization problems for PA positioning, bandwidth allocation, and quantization design. Efficient solution frameworks are then derived for the resulting downlink and uplink optimization problems. Experimental results show the superiority of the proposed design in convergence speed and final accuracy under non-independent and identically distributed data and communication bottlenecks.

cs.IT

新论文 · v1arXiv:2611.16301

LoCoJAX: Host-Local Coordination for Small-Model RL on TPUs

Loki Chen, Gregory B. Rehm, Yang Song, Hongye Xie, Ming Lei, Chongyang Bai, Bob Kamma, Mingfu Liang 等 16 位作者

原文摘要

Reinforcement learning (RL) increasingly targets small language models, and executing these workloads efficiently across multiple accelerators requires effective coordination. On TPUs, Pathways-based RL uses a single controller to coordinate computation across hosts. This execution model is effective for large models, but its coordination computational costs are harder to amortize when small models repeatedly invoke short computations. To characterize these computational costs, we sweep dense models from 0.6B to 405B parameters on TPU v7x under a decode-heavy workload and measure how host and network computational overhead vary with model size. At 0.6B, host orchestration occupies 36% of step time and network communication 2%, whereas for large models, estimated network computational costs dominate the non-compute time. Motivated by these findings, we present Local Coordination JAX (LoCoJAX), which implements local rollout and reward processing on each TPU host with synchronous global training. In controlled comparisons against a tuned and optimized Pathways baseline on Qwen3-0.6B across three workloads and 8-64 TPU v7x chips, its colocated mode (LoCoJAX-CO) is 1.3-3.6x faster per step; on GSM8K, it completes a step faster on 8 chips than Pathways does on 64. Finally, a Ray-managed disaggregated mode (LoCoJAX-DA) overlaps rollout and training on separate hosts, and placement experiments establish when this overlap compensates for dividing devices and transferring weights: only when rollout dominates the step and receives most of the hosts. On NuminaMath-CoT at 256 chips, giving rollout 75% of hosts cuts step time by about 15% compared with LoCoJAX-CO.

cs.LG · cs.DC

版本更新 · v2arXiv:2611.02627

Recovering Missing Degrees of Freedom in Simulation-Trained Field-Level Generative Models

Masato Shirasaki, Shiro Ikeda

原文摘要

Simulation-trained generative models are emerging as a powerful framework for field-level inference in cosmology, particularly in strongly nonlinear regimes where analytical approaches become intractable. However, the finite size of training datasets often leads to variance shrinkage, indicating that part of the realization-level variability of the underlying physical process is not captured by the learned generative dynamics. In this paper, we interpret variance shrinkage as a diagnostic of missing effective degrees of freedom and introduce physics-informed latent modifications to the generative dynamics of a continuous normalizing flow. We show that these latent corrections recover a substantial fraction of the missing variability without requiring additional training simulations. Using weak-lensing field-level inference as a representative cosmological application, we further demonstrate through posterior analysis that the learned latent directions are actively exploited during inference and behave as effective nuisance parameters that absorb part of the mismatch between the training and target distributions. As a result, the model exhibits improved robustness to out-of-distribution shifts in field-level inference. Our work establishes variance shrinkage as a practical diagnostic for identifying missing degrees of freedom in simulation-trained generative models and provides a framework for recovering them through physics-informed modifications of generative dynamics.

astro-ph.CO · cs.LG

版本更新 · v2arXiv:2608.13535

Joint Communication-Control Strategy Optimization with Partially Nested Information Structures: The Linear-Quadratic Case

Haoyi You, Kaiqing Zhang

原文摘要

In this paper, we formalize a joint communication-control strategy optimization (JCCO) problem in multi-agent linear systems with quadratic costs, under the common-information-based (CIB) framework from decentralized stochastic control. For computational tractability, we focus on such JCCO problems with partially nested (PN) information structures (ISs). In particular, with a baseline communication protocol that leads to a PN IS, we establish a series of conditions under which the partial nestedness is preserved under the (additional) communication strategies to be optimized, while violating them may cause nonlinearity or non-existence of the optimal control strategies, with open-loop communication strategies. Under these conditions, we then develop a dynamic-programming-based approach to compute the optimal control strategies of JCCO with open-loop communication strategies, which yields a set of closed-form Riccati Equations. As a byproduct of independent interest, such an approach also offers a new way to solve decentralized linear-quadratic control with PN ISs and output feedback, under the CIB framework and the aforementioned conditions. Finally, we extend such an approach to JCCOs with closed-loop communication strategies, yielding a more tractable dynamic program than an infinite-dimensional CIB-belief-based one.

eess.SY · cs.MA · math.OC

新论文 · v1arXiv:2611.16290

Parallel Computing Systems for the Optimal Selection of Process Equipment

Andrey B. Borisenko

原文摘要

This monograph addresses the optimal selection of equipment for chemical-engineering systems of multiproduct batch plants, a large-scale NP-hard discrete optimization problem in which, for each processing stage, a unit of standard size must be chosen so that the total capital cost of the equipment is minimized while the system remains operable and produces the required products within a given time horizon. We present the mathematical model of system operation and analyze its computational properties. We then develop a sequential branch-and-bound algorithm and its parallel versions for shared-memory systems (OpenMP), distributed-memory clusters (MPI, master-worker), and graphics processing units (CUDA), including recursive and iterative tree traversal, shared-memory optimization, and branch-divergence reduction. To overcome the exponential growth of the run time, we propose a hybrid metaheuristic in which Ant Colony Optimization finds a feasible initial solution and Simulated Annealing improves it, and we implement it on GPUs with thousands of independent threads. Finally, we propose random, constant, and multi-constant approaches to tuning the parameters of the ant colony algorithm based on statistical analysis of many parallel runs, and we study the program on GPUs of three generations and on nodes with several GPUs. Computational experiments on multi-core processors, clusters, and GPUs show that the parallel hybrid algorithm finds solutions within 0.01% of the branch-and-bound optimum in tens of seconds instead of tens of hours.

cs.DC

新论文 · v1arXiv:2611.16289

Eliciting and Mitigating Specification Satisficing in Coding Agents

Ruixin Wang, Tianyi Zhang

原文摘要

Coding agents often fail to meet all requirements in the task description. In some of these failures, the agent considers an implementation choice that meets a requirement, but later decides to commit to another one that does not meet the requirement for a reason it finds acceptable. We call this specification satisficing. It is challenging to identify and mitigate specification satisficing because the agent accepts such choices for its own reasons, and existing self-correction methods are biased toward them. We propose SatisFix, which guides a coding agent to report these choices and their reasons, and then check each choice against the task description. We evaluate SatisFix on DeepSWE and SWE-bench Pro with five LLMs as backbones and two different agent frameworks, and compare it with three self-correction methods. SatisFix improves the pass@1 of the base agent by an average of 5.09% on DeepSWE and 7.06% on the hard subset of SWE-bench Pro. On DeepSWE, it outperforms the best baseline by 3.98%. Our ablation study shows that both steps of SatisFix contribute to the improvement.

cs.SE · cs.AI · cs.CL