跳到正文
研踪PaperTrail

论文库

检索计算机领域论文,按研究方向、作者或关键词筛选。

部分采集异常
正在读取账号与研究记录…

全站更新概况

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

arXiv 最近完成扫描:2026-11-22 08:44 · 全站最近入库:2026-11-21 12:19各信源进度 →

统计口径与数据新鲜度

统计 2026-11-21 11:24 — 2026-11-22 11:24(北京时间) · 点击数字查看对应简报;AI 相关包含在新增论文内。

最近完成检查:2026-11-22 08:44最近发现新增版本:2026-11-21 12:19库内论文最新更新时间:2026-11-20 02:59

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

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

浏览论文库

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

找到 23,345 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-22 08:42(北京时间)
新论文 · v1arXiv:2611.26145

Can World Models Predict Rice Growth? Toward Agricultural World Models

Wenkai Ji, Zeteng Li, Yinkun Guo, Jie Zhou, Liang He, Zhenong Jin

原文摘要

World models can imagine how a physical scene evolves over the next few seconds, but growers need to know how a crop will develop over weeks to months: will this rice plant have headed a month from now, and how will it look at harvest? We introduce agricultural world modeling, the task of predicting how the same plant will look at a requested horizon of days to months, from one photo of it, its cultivar, and its camera view. In preliminary analysis, even state-of-the-art image generators get this wrong: they often grow the plant too little or give it the color and form of the wrong growth stage. We argue that they do not work out the plant's developmental state, which the photo reveals only indirectly. We therefore propose AgroWM, which reasons about development before it draws growth. Its reasoning module, PhenoClock, works out the current growth stage, predicts the future one, and writes a Growth Script with numeric targets for height, plant area, and color; its generation module, the Rooted Renderer, then edits the photo to follow this script, adding new organs and colors to the photographed plant. Because pixel scores can prefer an unchanged copy of the photo, we instead check each forecast's traits and developmental timing against the plant's own later photos. On rice, 54.6% of the forecasts of AgroWM are matched to the target stage (Stage R@1), against 34.8% for GPT Image 2.5, 24.0% for Qwen Image 2.1, and 47.4% for the strongest domain-specific baseline, the same backbone fine-tuned without Growth Scripts; 9 to 20 weeks ahead, it still reaches 46.4%, against 35.7% for that baseline. On maize, a crop with a different body plan and stage scale, the same design again scores best on all four measures.

cs.CV · cs.CL

新论文 · v1arXiv:2611.26143

Revisiting Finiteness of Matrix Monoids

Rida Ait El Manssour, Roland Guttenberg, Nathan Lhote, Mahsa Shirmohammadi, James Ben Worrell

原文摘要

This paper concerns decision problems related to finite monoids of rational matrices. We show that determining finiteness of a given finitely presented monoid is in PSPACE, improving the known $\text{co}{-}\text{NEXP}^{\text{NP}}$ bound. We also show that the membership problem for finite matrix monoids is PSPACE-complete, improving the known NEXP-upper bound. Our two complexity results are corollaries of a new polynomial bit-size bound on matrix entries in finite monoids. This is obtained by reduction to the case of matrix groups, using the structure theory of noncommutative algebras and of matrix monoids. Our techniques also give us a polynomial-time algorithm for deciding whether a monoid of rational matrices is conjugate to a monoid of integer matrices.

cs.FL · cs.LO

版本更新 · v3arXiv:2508.06243

SCAR: State-Space Compression for AI-Based Radio Resource Management of Vehicular Services

Ioan-Sorin Comsa, Purav Shah, Karthik Vaidhyanathan, Deepak Gangadharan, Christof Imhof, Per Bergamin, Aryan Kaushik, Gabriel-Miro Muntean 等 9 位作者

原文摘要

The increasing demand for connected vehicular services poses significant challenges for AI-based network and service management due to the high volume and rapid variability of network state information. Traditional management and control mechanisms struggle to scale when processing fine-grained metrics such as Channel Quality Indicators (CQIs) in dynamic vehicular environments. To address this challenge, we propose SCAR (State-Space Compression for AI-Based Radio Resource Management), an edge-assisted framework that abstracts high-dimensional CQI information into compact representations for AI-based management. SCAR employs machine-learning (ML)-based compression techniques, including clustering and radial basis function (RBF) networks, to reduce the dimensionality of CQI-derived state information while preserving essential features relevant to management decisions. The resulting compressed states are used to train reinforcement learning (RL)-based management policies that dynamically adapt the throughput-fairness trade-off according to service-level fairness objectives defined by the NGMN. Simulation results show that CQI compression improves average NGMN feasible-region occupancy from 56.32% to 88.07% across ten RL algorithms, reaching 97.76% for the best-performing RL configuration, with limited throughput degradation and competitive computational overhead. Furthermore, simulated annealing with stochastic tunneling (SAST)-based clustering reduces state compression distortion by approximately 10% compared with conventional K-means clustering. These results demonstrate SCAR's ability to provide compact state representations for RL-based network and service management while improving service-level fairness in dynamic vehicular systems.

cs.LG · cs.NE · eess.SY

新论文 · v1arXiv:2611.26135

Tight Bounds for Distance-Unbiased Sorting

Mats Bierwirth, Carola Doerr, Johannes Lengler

原文摘要

We study elitist randomized local search for sorting permutations. In this model, each iteration considers a possible exchange of two positions. The entries are exchanged if and only if they are in the wrong order. We introduce \emph{distance-unbiased} selection rules, which may choose an arbitrary distribution over exchange distances but sample uniformly among all pairs at each chosen distance. For static rules, we prove that the optimal worst-case high-probability runtime is $Θ(n(\log n)^2)$. This establishes the asymptotic optimality of the harmonic sorter within this class and partially resolves an open question of Olesker-Taylor. We then allow the distance distribution to change either according to a predetermined schedule or adaptively using the current permutation and the entire history. Matching upper and lower bounds show that both classes have optimal runtime $Θ(n(\log n)^2/\log\log n)$. Thus, scheduled adjustments of the distance distribution improve the optimal static worst-case runtime by a factor of $Θ(\log\log n)$, while full-information adaptation provides no further asymptotic improvement.

cs.DS

新论文 · v1arXiv:2611.26134

RouterWeave: A Unified Observation Layer for Cross-Platform Activity Analysis via Home Routers

Chengeng Liu, Ronghua Li, Yan Long, Shinan Liu

原文摘要

Smart homes bring together devices from different operating systems and vendor ecosystems, but each platform reports activity only for the devices and features it supports, leaving gaps in household-wide activity monitoring. Interoperability standards such as Matter and software integrations connect and control supported devices, yet vendor restrictions and licensing barriers limit access to their data, while hardware constraints hinder integration. Our key insight is that encrypted network traffic provides complementary evidence of activity across platform boundaries. Accordingly, we present RouterWeave, a router-side observation framework that combines cross-platform traffic collection, task-adaptive processing, and unified activity records. These records preserve activity meaning, temporal extent, and source-device attribution while distinguishing discrete operations from sustained activity intervals. We evaluate RouterWeave across four datasets, native execution on a commodity home router, and two novel case studies: cross-platform Screen Time monitoring and appliance energy estimation. RouterWeave achieves 94.41% macro-F1 on application activities and 97.46% on IoT operations, outperforming baselines by up to 20.99 percentage points, while using under 7 MB of router memory per task and supporting 54 concurrent devices within a 5-second update window. In end-to-end case studies, RouterWeave reconstructs 600 Screen Time timelines across Android, iOS, and Windows with 92.15% accuracy (24.84-second mean active-time error over 19-minute averages) and recovers 49 of 55 appliance operations over four days, tracking ON/OFF states with 92.84% agreement and estimating electricity use within a 4.28% error. Together, these results establish the commodity home router as practical infrastructure for whole-home activity sensing. The code is open sourced at https://github.com/naisslab/routerweave.

cs.NI · cs.HC

新论文 · v1arXiv:2611.26132

Sage-Evolve-Preview:Adaptive Observation for Post-Training Self-Improvement

Yinhao Tang, Youqing Fang, Yanan Sun, Xintian Shen, Hanyu Sun, Wenran Liu, Yue Wu, Weiming Zhang 等 10 位作者

原文摘要

Post-training recursive self-improvement (RSI) iteratively improves a target capability, operationalized by a fixed anchor benchmark, through repeated intervention, training, and evaluation. Yet most loops use the anchor's aggregate score as their sole observation, so early gains give way to stalled progress and repetitive exploration. When this anchor-only feedback ceases to reveal actionable gaps, a plateau may reflect an exhausted observation rather than an exhausted capability. We introduce adaptive observation (AdaObs), an observation-side mechanism of harness evolution that keeps the anchor objective fixed while adapting the active observation. At a plateau, a controller refines diagnostic granularity, escalates difficulty, and adds related tasks; validated increments are fused back into training, while every checkpoint remains evaluated against the anchor. Across eight independently optimized anchors in long-context reading comprehension and instruction following, AdaObs raises the mean score from $66.4$ to $73.7$ under a matched data budget, outperforming FixedObs, the anchor-only baseline. The evolved harness further improves the seven-task average on the independent PostTrainBench across two agent backends. These results identify observation adaptation as a missing mechanism of harness evolution for sustaining post-training RSI.

cs.AI

版本更新 · v2arXiv:2611.23822

Mapping the Design Space of Feature Sampling for Optical Flow

Liping Zhang

原文摘要

The computation of correlation features is crucial for optical flow estimation. Existing methods (e.g., RAFT) perform fixed-radius lookups on a cost volume, yielding a set of correlation features for each position; alternatively, WAFT performs single-point feature indexing, yielding one correlation feature vector per position. None of them designs a suitable correlation feature sampling scheme for each position according to regional characteristics. Our insight is that not every position requires the same sampling scheme. Applying an identical sampling scheme to every position is either redundant or insufficient and ignores the specific characteristics of the corresponding region. This paper rethinks the design of feature sampling and investigates the following question: should the support of feature sampling---i.e., the extent and directions over which evidence is aggregated---be learned separately for each query position? To study this question, we propose SAGE (Sampling with Anisotropic Gaussian Expectation). This method replaces single-point sampling with a per-pixel anisotropic Gaussian expectation on a fixed candidate grid centered on the current optical flow estimate. A positive-definite precision matrix controls the scale, anisotropy, and orientation of the sampling support, with single-point sampling included as its limiting case. Consequently, the sampling support becomes an explicit and analyzable design variable and provides a principled way to analyze how directional feature aggregation affects correspondence learning. Extensive experiments show that SAGE, while adding only 1.5% multiply-accumulate operations (MACs), effectively improves zero-shot cross-dataset generalization. When trained only on FlyingChairs and FlyingThings, SAGE reduces the end-point error (EPE) by 24.2% and 16.4% on Sintel clean and final, and reduces KITTI F1-epe and F1-all by 7.2% and 11.6%.

cs.CV

新论文 · v1arXiv:2611.26131

SCARVA: Interpretable Agentic Reasoning in Software Supply-Chain Analysis

Tim Barsch, Bruno Nogueira, Tobias Wienand, Jan Fopma, Flavio Toffalini

原文摘要

Complex applications may inherit security vulnerabilities from their dependencies, though a vulnerable dependency does not necessarily endanger the application. Existing tools flag dependencies associated with public vulnerabilities, and more advanced ones analyze the call graph to assess whether the vulnerable code is reachable. However, these approaches are limited to specific programming languages and ecosystems. LLM agents are promising candidates for this task, yet they hallucinate, take shortcuts, and reward hack. As these behaviors are opaque, the agents' conclusions may not be trusted. We present SCARVA a framework designed to produce interpretable reports. SCARVA combines advanced agent reasoning with an evidence-collection methodology that assesses the agent's analysis and improves its trustworthiness. Specifically, the ANALYST performs a software security assessment while SCARVA records every tool call, file access, and web request. Then, one or more independent JUDGE agents inspect the ANALYST's result and produce an evidence-backed report. We evaluate SCARVA on 54 manually curated cases, balanced across exploitable, non-exploitable, and vulnerability-free projects, using both open-weight and frontier models. The best ANALYST model, Qwen 3.8, reaches 82% balanced accuracy and surpasses the evaluated proprietary models. Human annotation validates the JUDGE findings: Claude Opus 5.5 performs best, agreeing with annotators in 94% of claim checks. Finally, we discuss the most common failure modes of agents on supply-chain vulnerability assessment.

cs.CR

新论文 · v1arXiv:2611.26128

CryoGaussian: Heterogeneous Cryo-EM Reconstruction Made Explicit with Gaussian Splatting

Ziwen Wang, Jiajun Fan, Chaoran Cheng, Ruihan Guo, Yanru Qu, Juexiao Zhou, Ge Liu

原文摘要

Single-particle cryo-electron microscopy (cryo-EM) determines macromolecular structures from noisy particle images. However, rare, transient compositional states and subtle, continuous conformational variation are difficult to resolve. We present CryoGaussian, a Gaussian splatting method for heterogeneous cryo-EM reconstruction that fits an explicit Gaussian density representation directly to the particle images through a differentiable rasterizer rendering orthographic line integrals modulated by the contrast transfer function. The Gaussians are initialized randomly, with their number derived from the estimated spatial support of the molecular density and their minimum size bounded by the Nyquist frequency of the particle images. To capture the compositional and conformational variation across particle images, we learn particle-image and per-Gaussian latents that jointly modulate the base position, amplitude, and scale of each Gaussian. Each structural state is then reconstructed by fitting the Gaussians conditioned on its inferred latent. On CryoBench, CryoGaussian achieves state-of-the-art reconstruction accuracy for both compositional and conformational heterogeneity. On experimental data, it resolves a rare compositional state of the 50S ribosomal subunit (class C4, 1.3% of particles) with rRNA helix 68 docked in the absence of the central protuberance, and continuous conformational motion of the SF3b and helicase subcomplexes of the pre-catalytic spliceosome. We further optimize the rasterizer for additive cryo-EM projection, reducing training time by 12.8-28.0$\times$.

cs.CV

版本更新 · v3arXiv:2607.18935

Online Stochastic Matchings: Stability on Hypergraphs

Fabien Mathieu

原文摘要

In stochastic dynamic matching, items of several classes arrive at rates $λ$ and leave in compatible groups. When groups may exceed two items, as in quantum switches or organ exchange, compatibility is described by the incidence matrix $A$ of a hypergraph, whose entries count how many items of each class a group consumes (single-class hyperedges can model discards). We call a system stabilizable if some matching policy keeps large queues rare in the long run, and show that this holds if and only if the conservation equation $Aμ= λ$ admits a nonnegative solution whose support induces a surjective submatrix. Equivalently, $λ$ lies in the interior of the cone generated by the hyperedges. The condition is necessary for all policies, including idling and clairvoyant ones. It is sufficient through one explicit policy: the rewardless counterpart of the Extended Greedy Primal--Dual policy of Nazari and Stolyar (2019) stabilizes every stabilizable system without knowledge of the rates, and its mean queue length is explicitly bounded. Approaching a solution whose support is not surjective has a cost, under every policy: matching rates within $ε$ of it force mean queues of order at least $1/ε$. On simple graphs, some greedy policies stabilize every stabilizable system. On a small hypergraph with a single hyperedge of size three, every greedy policy is unstable on an open set of stabilizable rates. This shows that no greedy policy, nor any family of greedy policies, can approach maximal stability on hypergraphs: idling is necessary. We also settle two cases explicitly: the complete $k$-uniform hypergraph (the $k$-partite quantum switch), which is stabilizable exactly when no link generates a fraction $1/k$ or more of the total rate, and the Fano plane.

cs.NI

版本更新 · v2arXiv:2607.24890

A New Look at the Classical Estimation Problem

Paul W. Vos

原文摘要

Bahadur's \emph{Lectures on the Theory of Estimation} develop the classical theory of point estimation inside the geometry of Hilbert space, and they record with unusual honesty where the theory strains: the locally best unbiased estimate depends on the parameter, a two-point parameter space yields an estimate Bahadur calls absurd, the odds in binomial sampling have no unbiased estimate, and the virtues of maximum likelihood remain heuristics. We present a subset of the lectures in Bahadur's notation and, at each strain, make one small modification: for each sample point, an estimate $τ$ becomes a function on the parameter space rather than a point in it. The result is the continuum of null hypotheses that Fisher, in 1955, identified with the theory of estimation; we call this \emph{generalized estimation}, and $τ$ a generalized estimator. Bahadur's own definition of an estimate, square-integrable at every distribution in the family, already supplies the domain. The payoffs are tracked lecture by lecture: estimators that exist at boundary samples where point estimates do not; an elementary lemma showing that no pointwise criterion admits a uniformly optimal estimator, which explains why admissibility, minimaxity, Bayes averaging and unbiasedness arose; assessment by information, $Λ(τ)$, with the score attaining the Fisher information bound uniformly by a three-line argument; Cramér--Rao attainment and sufficiency recovered as equality cases of that bound; and the maximum likelihood heuristics converted into exact statements about the score. Nothing classical is overturned; the classical apparatus is explained by Fisher's characterization of estimation as a continuum of significance tests.

math.ST · cs.IT

版本更新 · v2arXiv:2611.11437

How to "train" a surgeon? Benchmarking generalist agents in surgical scene understanding

Gengyuan Zhang, Xiao Han, Shijie Zhou, Abdullah Koukash, Jingpei Wu, Jindong Gu, Yuqicheng Zhu, Nassir Navab 等 9 位作者

原文摘要

Evaluating general-purpose multimodal large language models (MLLMs) for surgical scene understanding remains limited due to scarce surgical data and fragmented evaluation protocols. We address these challenges through a two-stage evaluation framework for surgical scene understanding. In Stage I, we introduce SurgKnowBench, a diagnostic benchmark that harmonizes public surgical datasets across procedures and tasks and evaluates proprietary and open-source MLLMs to map the boundary of pretrained surgical competence. SurgKnowBench reveals that current models retain partial visual priors but remain weak on relational and procedural understanding and workflow assessment. Building on this failure frontier and motivated by the scaffolded nature of surgical training, we introduce SurgAgentGym, an agentic environment where general-purpose MLLM agents can use bounded knowledge search over SurgWiki and general-purpose visual grounding for surgical tasks. SurgAgentGym evaluates whether generalist agents can improve surgical understanding at inference time through tool-mediated evidence use, without additional surgical training. Our results show that agentic inference narrows the gap mainly for perception-heavy tasks, while fine-grained localization and expert-style assessment remain persistent bottlenecks. This opens a path toward future studies of agentic surgical understanding.

cs.CV · cs.AI

新论文 · v1arXiv:2611.26117

ChronoPatch: Black-Box Energy-Latency Attacks on Vision-Language Models

Jiawei Xing, Li Wang, Xiangtao Meng, Chuanchao Zang, Rui Wen, Zheng Li

原文摘要

Vision-language models (VLMs) are increasingly deployed as generative inference services, where autoregressive decoding makes response length a key determinant of latency and computational cost. This creates an availability risk: adversarial visual inputs can prolong generation and increase serving overhead. Existing energy-latency attacks typically rely on gradients, logits, or token-level feedback, leaving open whether such risks persist when an attacker observes only the final response. We present ChronoPatch, a black-box attack that induces prolonged VLM generation requiring only the final response. It discovers visual delay anchors through response-length-guided evolutionary search, then minimizes perturbation magnitude through subspace-calibrated decision-based optimization. Across four open-source VLMs, ChronoPatch increases inference latency by 47.86%--293.34% and GPU energy consumption by 41.44%--249.70% under a 5% spatial budget. It also induces more repetitive outputs, as reflected by increased n-gram repetition ratios. On three commercial VLM APIs, ChronoPatch increases response length by 37.6%--81.5% and client-observed latency by 1.82x--2.29x under query budgets of 500 per API. These results reveal a generation-based availability risk that persists under final-response-only access and motivate generation-aware safeguards for deployed generative AI services.

cs.CR · cs.AI

新论文 · v1arXiv:2611.26116

Learning Reconstruction Errors as Spatiotemporal Evidence for AI-Generated Video Detection

Jiahua Jiang, Jinglai Li, Jianru Zhang, Jin Zhu

原文摘要

The proliferation of AI-generated videos has raised growing concerns about misinformation, fraud, and potential societal risks, creating an urgent need for detection. Reconstruction-error-based methods have shown strong empirical performance for AI-generated image detection by exploiting discrepancies between an input and its reconstructed version. Their extension to video detection, however, faces two challenges: (i) a fixed reconstructor is not optimized for the detection objective and may yield weak discriminative signals; and (ii) an additional raw-video feature stream is often introduced to model temporal dependencies, increasing architectural overhead. To address these challenges, we develop a detector that learns and exploits reconstruction errors as a unified source of spatiotemporal evidence. First, we jointly optimize a reconstructor and the downstream classifier to produce reconstruction errors that are more informative for detecting AI-generated videos. Second, we show that consecutive reconstruction errors alone carry temporal evidence, and thus a 3D convolution over them captures spatiotemporal evidence without a raw-video stream. Extensive experiments on over 30 existing AI-video generators demonstrate competitive performance against baselines.

cs.CV

新论文 · v1arXiv:2611.26114

Loss-Guided Tsetlin Machines: Making the Learning Objective Explicit

Mayur Shende, Runar Helin, Ole-Christoffer Granmo, Vladimir I. Zadorozhny

原文摘要

The Tsetlin Machine (TM) has no explicit notion of a loss function. Its learning signal is fixed to an implicit, margin-based update rule (MBUR), with no way to choose an objective function suited to a specific task. This paper introduces Loss-Guided TM, which replaces MBUR with an explicit and swappable loss function. In our proposed method, clause feedback and weight updates are driven by the loss, while keeping the Tsetlin Automaton and clause logic unchanged. The approach incorporates cross entropy (CE), symmetric cross entropy (SCE), and asymmetric loss (ASL). We evaluate Loss-Guided TM across six datasets that span multiclass, multilabel, and binary classification tasks. With a suitable loss, Loss-Guided TM matches or improves accuracy and F1 over MBUR on five of the six datasets. The gains are greatest for the imbalanced multilabel datasets, where ASL improves macro F1 by up to 9 percentage points. With CE, Loss-Guided TM is also significantly better calibrated than MBUR on all six datasets, reducing calibration error by a factor of 3 to 11. A qualitative inspection shows that the clause-based global interpretability, examined at the per-class level on image and text data, is preserved when using the Loss-Guided mechanism. These results show that our explicit TM learning objective allows direct targeting of uncertainty calibration, data imbalance, and task structure, opening the TM to the broader literature on loss functions.

cs.LG

新论文 · v1arXiv:2611.26112

ZoionNET: Low-Power Animal Intrusion Detection Using an Acoustic Deep-Learning Classifier for a Cellular Remote-Sensing System

Prasanna Vijay, Adarsh Gupta, Vansh Panwar, Vatsalya Singh, Shubham Singhal, Anubhav Saha, Teena Sharma

原文摘要

Crop raiding and village intrusion by elephants and large felids impose a recurring cost on communities along forest boundaries in southern India, and existing camera-based monitoring has proven difficult to sustain on the power budgets available at such sites. We present ZoionNET, a battery-powered sensing node that detects animal intrusion acoustically rather than visually. Audio from an onboard PDM microphone is sampled at 8 kHz, converted to mel-spectrograms on-device, and classified on an Arm Ethos-U55 microNPU by two int8-quantised YOLOv8 models arranged as a two-state machine: a lightweight always-on model runs in a dormant state and gates a four-class species classifier through exponential-moving-average thresholds, so the heavier model runs only when an intrusion is likely. A digital vibration sensor supplies a complementary seismic channel. Inferences leave the node as JSON over a cellular Notecard, reaching a web dashboard and SMS alerts to forest rangers without any local gateway. The two models reach reported accuracies of 0.974 and 0.946 on held-out audio, and a complete node costs USD 133.57 in parts. We report the system as built and deployed, and identify explicitly what has not yet been measured, most notably the energy behaviour that motivates the two-state design.

cs.SD

版本更新 · v3arXiv:2611.06476

The Alignment Tax Is Graded: Task-Selective Capability Loss Under Length-Normalized DPO with SFT Retention

Ye Yuan

原文摘要

Setting. This is an empirical characterization study; we propose no new method and run no comparisons against alternative objectives. We fine-tune Qwen3.5-4B with sparse LoRA and length-normalized DPO plus an SFT retention term (the RPO objective) on structured financial-filing tasks, and track downstream capability as the implicit reward margin grows to $\sim 7$ $β$-scaled units (nf4 training; $\sim 3$ under fp16 measurement on the unquantized base). Question. Does margin growth degrade capability uniformly across tasks, or selectively? Findings. In a pre-registered study (8 seeds, 6 unseen at freeze) with artifact-aware scoring and TOST equivalence testing, the alignment tax is graded and task-selective: extraction is preserved ($Δ= -0.55$ pts, equivalent at $\pm5$ pts), json loses little ($Δ= -3.52$ pts, equivalence fails on a clustered item-level interval), and router degrades severely ($Δ= -30$ pts). Router's loss concentrates in a margin 1-3 damage window containing a transient format-break spike. The margin is accounted for almost entirely ($\geq 96\%$) by suppression of rejected completions, with chosen reward approximately flat, consistent with the retention term anchoring it; we do not attribute the behavioral damage to any single gradient channel. Gold-logp deflation is task-selective (extraction $-0.15$, json $-1.04$, router $-1.47$ nats/token under fp16), with gold selectively protected relative to incorrect controls on extraction (8/8 seeds). QLoRA reward margins differ up to $3.4\times$ between nf4 training and unquantized fp16 measurement, with an axis-dependent gap. Scope. Length-normalized rewards with SFT retention, sparse LoRA on 8/32 layers, one 4B model, easily-separable synthetic preferences, three structured tasks. A no-retention control and a measured format-alignment metric are left to future work.

cs.LG

版本更新 · v2arXiv:2609.13771

Homeostatic Continual Learning

Yue Jin

原文摘要

I formulate a Continual Learning problem for an autonomous agent that imitates the one animals face. The agent has intents to satisfy. It experiences contexts where different combinations of features are active and impact intent outcomes. I propose a method named ``Homeostatic Continual Learning'' that imitates the one animals use. The agent learns incrementally environment data and features relevant for their intents from the impact of contexts, gradually completes its model and policy and performs well in more and more contexts. The core of the method is to find outliers in the environment data when the agent experiences an outlier in its intent outcome after the context changes. This method enables the agent to learn continuously in changing contexts without catastrophic forgetting. When pursuing many intents concurrently, the agent may build a transformer-like world model with the method where it factorizes the objects in the world into features, abstract objects into comparable instances of concepts and map concepts to intents through features. The world model is composed of semantic nodes and explainable by nature. I discuss the works needed to render the method practical, the connections to many fields in Artificial Intelligence and the broader implications of the method.

cs.AI

新论文 · v1arXiv:2611.26111

Physics-Informed Deep Learning for Shear Wave Speed Estimation in MR Elastography

Stefan Martin, Jakob Schattenfroh, Patrick Schuenke, Felix Frederik Zimmermann, Ingolf Sack, Christoph Kolbitsch, Andreas Kofler

原文摘要

Objective: Magnetic Resonance Elastography (MRE) is a non-invasive imaging technique for mapping biomechanical properties of in vivo tissue, including shear wave speed (SWS), but involves intrinsically slow data acquisition and an ill-posed wave inversion. Instead of relying on handcrafted image priors, we propose a data-driven approach jointly combining image reconstruction and MRE inversion for robust SWS estimation from undersampled k-space data. Methods: Our physics-informed reconstruction framework comprises two blocks: a model-based neural network (NN)-regularized reconstruction module and a phase-gradient inversion (k-MDEV) calculating SWS from the reconstructed images. Concatenating both blocks yields an end-to-end trainable method to estimate SWS directly from measured k-space data. We evaluated the method on retrospectively highly undersampled brain MRE data and compared it to a total variation (TV) minimization-based approach. We assessed the impact of end-to-end training (qualitative images and SWS maps as targets) versus pre-training (qualitative images as targets) and applied the method also to in vivo data. Results: Our approach significantly reduces NRMSE by 30% compared to TV. End-to-end training improves SWS estimation over separate image reconstruction and SWS calculation. Conclusion: Accurate SWS quantification is possible at acceleration factors up to 19. Our method significantly outperforms TV, highlighting the need for data-driven regularization in this challenging MR problem. Further, our approach successfully generalizes to in vivo data. Significance: We present the first end-to-end trainable MRE reconstruction method for estimating SWS maps directly from k-space. NN-based reconstruction can enable rapid stiffness mapping for dynamic studies, functional imaging, and real-time clinical feedback.

eess.IV · cs.LG · physics.med-ph

新论文 · v1arXiv:2611.26109

SenseAlign: Learning Decision-Aligned Predictive Representations for Active Urban Sensing with UAVs

Weijiang Xiong, Nikolas Geroliminis

原文摘要

Active urban sensing orchestrates a mobile sensor fleet to proactively inspect a dynamic road network and gather critical information for system management. This problem simultaneously requires understanding the complex spatio-temporal evolution and making decisions with partial observations. In this work, we focus on active sensing in a large-scale urban road network with utilities grounded in dynamic traffic states, while prior work has rarely considered these factors jointly. We propose \emph{SenseAlign}, a Reinforcement Learning-based (RL) framework that learns decision-aligned predictive representations for coordinating an urban sensing drone fleet. SenseAlign first trains a privileged teacher agent that exploits future ground-truth traffic for optimal sensing choices. The student agent has an auxiliary branch that learns to predict future traffic with partial observations, in addition to the RL-trained actor and critic branches. During student learning, our framework matches the action distributions of the teacher under ground-truth and student-predicted future traffic, while keeping the teacher's observations unchanged. Such a design shapes the latent space for decision-making instead of prediction alone. At inference time, the student agent has no overhead as all this learning is done in training. We conduct comprehensive experiments with the real road network of central Barcelona containing 1570 road segments, and demonstrate that our method consistently outperforms the baselines on multiple utility formulations. The codebase will be made publicly available upon paper acceptance.

cs.AI

新论文 · v1arXiv:2611.26108

Pareto-Optimal and Strategyproof Data-Sharing Between Competitors

Morgan Colin-Mourre, Felipe Garrido-Lucero, Patrick Loiseau

原文摘要

We study data-sharing among firms competing in the same market through the quality of their predictive models. We model firms' profit through two main components, precision and market share, where the former depends on the firm's own number of datapoints and the latter depends on the endowments of all firms. We characterize Pareto-optimal contracts as those in which at least one firm receives all of the other firms' datapoints. We then propose a parametric family of data-sharing contracts and show that, under any of them, the relative market shares of all firms remain constant, eliminating the negative effect of sharing data with competitors and equalizing all firms' gains relative to the no-sharing scenario. We further show that one contract within this family satisfies both Pareto-optimality and strategyproofness against firms attempting to under-declare the size of their initial endowment. Finally, using a real-world dataset, we empirically test the gain in profit for all firms under the contract described above, and show that data-sharing is particularly beneficial when the machine learning task is harder or when firms' initial endowments are closer to homogeneous.

cs.AI · cs.GT

新论文 · v1arXiv:2611.26107

HiUni: Harmonizing Understanding and Generation in Shared Visual Space

Jiahao Hu, Jingdong Du, Yichen Zhang, Yi Song, Zhenyu Zhang, Jiaxin Niu, Ruihang Li, Xiangzhao Hao 等 13 位作者

原文摘要

Unified multimodal models aim to support visual understanding and generation within a single framework. However, prevailing approaches often separate their visual representations or backbone computation to accommodate the different requirements of these two tasks, which limits the degree of unification. In this paper, we present HiUni, which revisits this design choice and explores whether a pretrained vision language model can be directly extended into a unified model with extensive parameter sharing. Specifically, HiUni reuses the same language-aligned ViT and dense Transformer backbone for both capabilities, while confining task-specific computation to the output side. Yet such a design introduces two key challenges. Generation in pixel space is difficult to optimize, and semantic abstraction in pretrained ViT discards fine spatial details required for synthesis. To address these challenges, HiUni introduces a \emph{Pixel Diffusion Head} tailored to direct pixel-space optimization and a lightweight \emph{HiGate} pathway that provides fine-grained spatial details near the output stage. Extensive experiments show that HiUni achieves a similar average understanding score to the pretrained VLM while attaining competitive performance on image generation and editing. Controlled comparisons and analyses further demonstrate that a single dense backbone can effectively support both capabilities without task-specific backbone computation, highlighting a simpler path toward unified multimodal modeling.

cs.CV · cs.AI

新论文 · v1arXiv:2611.26106

AscMLA: Unlocking FP4 Efficiency for Long-Context MLA Decoding through Asymmetric Scaling

Zechun Zhou, Zhiyong Zheng, Xiaomin Zhang, Lin Ye, Han Li, Jingwei Sun, Guangzhong Sun, Jing Li

原文摘要

Long-context decoding repeatedly reads a growing KV cache. FP4 reduces cache storage and traffic but needs block scaling to handle its limited dynamic range. Applying FP4 to multi-head latent attention (MLA) requires reconciling different K/V scaling requirements on a shared latent. Existing approaches address this mismatch through dequantization or separate encodings, adding execution or storage overhead. We present AscMLA, which combines asymmetric K/V scaling with pipelined scale delivery to process a single FP4 payload directly. AscMLA uses 38.2% less cache memory than FP8 and achieves geometric-mean speedups of 1.35x over TRT-LLM's FP8 and 1.26x over CAKE's NVFP4 MLA kernels.

cs.DC

新论文 · v1arXiv:2611.26105

PhysVAM: Physics-oriented Future Value Action Model for Dynamic Object Manipulation

Yuxiang Yang, Jingbo Zhang, Jiaru Zhong, Junliang Chen, Yirui Zhou, Guangming Wang, Haitao Lin, Huaiyuan Xu 等 10 位作者

原文摘要

Predicting future scene or object states does not necessarily reveal how robot-object interactions will advance or hinder a task. This challenge is particularly pronounced in dynamic robotic manipulation, where objects continue moving during perception and execution. Effective action generation therefore requires not only modeling object dynamics but also anticipating how future interactions will shape task progress. To this end, we propose \textbf{PhysVAM}, a Physics-oriented Future Value Action Model that represents anticipated interactions through their effects on future task progress. This value-based formulation provides task-directed foresight without reconstructing future visual states. Conditioned on current observations and historical object and robot motion, PhysVAM jointly learns action generation and multi-horizon future-value prediction. The value branch learns horizon-specific latent representations through future task-value prediction from successful and failed trajectories. These future-value representations are integrated into the action policy through gated cross-attention, guiding action generation toward interactions that facilitate task success. Extensive simulation and real-world experiments demonstrate superior task success, with absolute success-rate improvements of 27.0 and 27.5 percentage points over the strongest baselines in simulation and real-world tasks, respectively. Project page: https://metarobo.github.io/physvam.github.io/.

cs.RO

新论文 · v1arXiv:2611.26104

How Network Topology and Data Heterogeneity Shape Decentralized Learning: A Spectral Analysis

Luigi Palmieri, Lorenzo Valerio, Chiara Boldrini, Andrea Passarella, Marco Conti, János Kertész

原文摘要

Network topology influences decentralized federated learning through information mixing and the relative contribution of individual nodes. While existing analyses primarily focus on spectral mixing properties, they often assume doubly stochastic aggregation, overlooking topology-dependent influence under irregular communication networks. We develop a spectral framework that separates two complementary mechanisms: the dominant eigenmode determines the consensus trajectory and node influence, while non-dominant eigenmodes govern the propagation and attenuation of disagreement. Under uniform neighborhood averaging, we show that the stationary influence is proportional to node degree plus one and derive a bound relating model disagreement to spectral mixing and gradient heterogeneity. We then investigate how these mechanisms interact with the data distribution across nodes. Experiments across regular and heterogeneous network topologies reveal nearly identical learning dynamics under IID data, while network size and connectivity affect how unevenly distributed information is learned under non-IID data. For non-IID settings, we characterize how the distribution of data across nodes shapes the relative contribution of local information to consensus dynamics.

cs.LG · cs.SI

新论文 · v1arXiv:2611.26103

UniVEP: Unifying Velocity and Endpoint Prediction for Robot Manipulation

Zhengtu Liang, Qiwei Liang, Tao Lin, Shihang Weng, Boyang Cai, Xiangyu Chen, Yikeng Chen, Jiaming Liang 等 11 位作者

原文摘要

Flow-matching robot policies usually use one prediction target throughout action generation, even though recovering actions from noise and polishing nearly clean actions bring different challenges. We systematically compare velocity prediction (V), endpoint prediction (X), and velocity-weighted endpoint prediction (X--V) across multiple robot policies, benchmarks, and inference budgets. On StarVLA and LiLa-WAM, X--V works best for few-step generation, while V gains more from iterative refinement. We further uncover their complementary strengths across noise levels: clean actions have concentrated covariance spectra, endpoint predictions stay more stable under high noise, and velocity predictions offer more accurate corrections near clean actions. Inspired by these findings, we introduce UniVEP, which learns both prediction targets together with consistency supervision and switches from endpoint recovery to velocity refinement as noise drops. Each step only needs one network run, and joint training also boosts single-step endpoint generation. Tests on simulated and real-world manipulation show better success rates and efficiency. On RoboTwin, UniVEP hits 61.36% success with one StarVLA run, beating the best single-target baseline at sixteen runs (59.96%). With LiLa-WAM, two runs outperform the best sixteen-step single-target baseline and cut inference latency by 77.1%.

cs.RO · cs.CV

版本更新 · v3arXiv:2410.23160

FlexTSF: A Flexible Forecasting Model for Time Series with Variable Regularities

Jingge Xiao, Yile Chen, Gao Cong, Wolfgang Nejdl, Simon Gottschalk

原文摘要

Generalizing across time series with irregular temporal structures poses a persistent challenge for universal time series forecasting models. Existing approaches often assume regular sampling or depend heavily on imputation, which limits their applicability in real-world scenarios where irregularities are prevalent due to diverse sensing devices and recording practices. To address this gap, we propose FlexTSF, a flexible and domain-adaptive forecasting model specifically designed for time series with variable temporal regularities. At its foundation lies the IVP Patcher, a continuous-time patching module leveraging Initial Value Problems (IVPs) to inherently handle uneven sampling intervals, variable sequence lengths, and missing values. Built upon a decoder-only Transformer architecture, FlexTSF integrates normalized timestamp inputs and domain-specific statistical descriptors through a specialized causal self-attention mechanism, enabling adaptability across domains. Extensive experiments on 16 datasets demonstrate FlexTSF's effectiveness, significantly outperforming existing models in classic forecasting scenarios, zero-shot generalization, and low-resource fine-tuning conditions.

cs.LG · cs.AI

新论文 · v1arXiv:2611.26101

Revealing Pathway Interactions in Network Computations with Complete Path Contribution

Shenghua Liu, Youssef Faragalla, Surya Ganguli, Stephen A. Baccus

原文摘要

Networks pass signals through many layers of interconnected units, so explaining what they compute means tracing how information travels from input to output and how pathways interact. Most existing approaches analyze only pairwise interactions between components such as units, layers or attention heads. Even when these measurements are combined post hoc into a circuit, they do not capture higher-order interactions along full causal chains (paths) from input to output, and so cannot address how exponentially many paths interact to construct network computation. This leaves a large set of mechanistic questions unanswerable, for example whether inhibitory pathways veto excitatory ones, and how the two combine to yield network decisions. To address this gap, we introduce Complete Path Contribution, a method to decompose the network output into mathematically complete contributions of individual paths to study path-level interactions directly and at scale. Partial summation then induces definitions of contributions for any circuit at any granularity level, with completeness guaranteed at every level. We further present an efficient algorithm that identifies the $k$ strongest paths without examining all of them, with runtime empirically linear in $k$ rather than exponential in depth. By applying Path Contribution to benchmark image classification networks, we quantify excitatory-inhibitory interactions and show how they collectively construct network decisions at a level of detail previously unmeasurable. Furthermore, in an interpretable model of the mouse retina, it reveals how dynamic receptive fields emerge from excitatory-inhibitory interaction motifs among pathways between different cell types in response to time-varying natural stimuli. Overall, Path Contribution reveals how a network's causal pathways interact at scale to transform input to output.

cs.LG · q-bio.NC

新论文 · v1arXiv:2611.26100

Causal Overtaking in Fair Transaction Ordering

Haozhe Feng, Di Wu, Jian Liu

原文摘要

Fair transaction ordering protocols aim to mitigate front-running using replica-local arrival orders, but prior work often assumes a single submission origin for each responsive transaction. We identify causal overtaking, in which corrupted replicas observe a target and submit identical signed copies of a responsive transaction through multiple client entries. The copies' first arrivals at different replicas can collectively provide ordering support that no individual path achieves, allowing the response to execute before its target. We formalize a class of receipt-order protocols, including Themis, characterize the timing and network conditions under which the attack is feasible, and show that delayed gossip does not rule it out. We further prove that, under our network model, histories requiring opposite transaction orders can produce identical receipt evidence. Thus, receipt-only mechanisms cannot always preserve earlier-submission priority while preventing responsive overtaking. We propose a mitigation and analyze its assumptions and limitations. Finally, we evaluate the attack and mitigation on Themis through simulations driven by AWS and WonderProxy latency measurements.

cs.CR

新论文 · v1arXiv:2611.26098

Frequency-Synchronized Boundary Coupling for Training-Free Multi-Prompt Long Video Generation

Yipan Xu, Shuyong Gao, Jiyuan Fu, Keliang Yin, Jiayu Chen, Qianyu Guo, Yang Liu, Wenqiang Zhang

原文摘要

Text-to-video generation has made remarkable progress with diffusion models and transformer-based video architectures, yet generating coherent long videos from evolving textual descriptions remains challenging. Training long-video models from scratch requires substantial computation and large-scale multi-event data, while training-free multi-prompt generation must activate new prompt semantics without breaking temporal continuity. We identify a frequency-agnostic boundary coupling problem in existing transition strategies: weak coupling causes abrupt changes, flicker, and background jumps, whereas overly strong or coarse coupling may over-preserve source semantics and produce ghosting or duplicated subjects around prompt boundaries. To address this problem, we propose FreqSync, a training-free framework that formulates prompt transitions as frequency-selective boundary coupling. FreqSync combines Frequency-Synchronized Source-Conditioned Attention (FS-SCA) with Residual-Clipped Spectral Stitching (RCSS), enabling adjacent segments to share stable low- and mid-frequency structure and motion cues while keeping high-frequency prompt-specific details target-dominant under explicit residual constraints. Extensive experiments show that FreqSync achieves state-of-the-art transition smoothness and prompt alignment while maintaining strong perceptual video quality, demonstrating its effectiveness for coherent multi-prompt long video generation.

cs.CV