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

+0新增论文
+0其中 AI 相关
+0已有论文修订
+91新增 AI 导读

arXiv 最近完成扫描:2026-11-22 14:51 · 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 16:42 — 2026-11-22 16:42(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 14:51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 24,181 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-22 14:49(北京时间)
新论文 · v1arXiv:2611.23887

Can AI Understand Indonesian Children's Book Illustrations? Compliance, Safety, and the Cultural Alignment Gap

Rian Adam Rajagede, Rochana Prih Hastuti, Fitri Hasanah Amhar

原文摘要

Children's book illustrations carry developmental, pedagogical, and cultural responsibilities that are often formally codified in guidelines. As AI systems begin entering children's publishing pipelines, both as evaluators and as generators, a natural question arises: can they reliably meet the structural, safety, and cultural standards required? We investigate this across three criteria: guideline compliance, safety, and cultural alignment, using Indonesia's Book Leveling Guideline as a government-issued evaluation framework with explicit visual criteria for four reading levels. To understand how well VLMs can identify and apply these guideline criteria, we first evaluate them on real published children's book pages and compare their assessments against a human baseline. To then examine whether AI-generated illustrations meet the same standards, we apply the VLM judge to images produced by two generation models and cross-check them against a subset of human illustrator raters. We find that VLMs are reliable on explicit criteria such as compliance and safety, but fall short on cultural alignment, systematically missing content that human raters flag as culturally inappropriate. AI-generated illustrations show the same blind spot: they satisfy structural and safety criteria but fail to capture implicit cultural details that human raters consistently identify.

cs.CV

新论文 · v1arXiv:2611.23886

Risk-Controlled Correction for Frozen Multimodal Models under Missing Modalities

Hoang Pham Cong, Long D. Nguyen, Ho Viet Duc Luong, Nguyen Tran Minh Nhat, Binh P. Nguyen, Huynh Thi Thanh Binh

原文摘要

A deployed multimodal predictor may face missing modalities after its weights can no longer be changed. This raises a question that training-time repair does not answer: what part of the resulting error is still correctable from the information that remains? We show that this part is characterized by the Bayes gap, the difference between the model's prediction and the Bayes prediction given the observed modalities. For cross-entropy and squared error, excess risk measures this gap in the geometry of the loss; ambiguity in the observed modalities cannot be removed. Thus, a badly degraded prediction is not necessarily a correctable one. Estimating the Bayes prediction normally requires labels. However, when a model observing more modalities is conditionally correct, averaging its outputs conditional on the remaining modalities recovers this target without retrieval labels; with the same fixed blend and neighbors, it is no worse in expected cross-entropy or Brier risk than averaging hard labels. Some such assumption is unavoidable: otherwise the target is not identifiable from unlabeled data alone. GUARD (Gated Utility-Aware Recalibration for Deployment) leaves the deployed predictor unchanged. It estimates a candidate correction by retrieval, learns from deployment-matched labeled data when correction is likely to help, and calibrates a threshold on a held-out split by conformal risk control. Under exchangeability, the marginal probability that GUARD applies a correction whose loss increase exceeds a prescribed tolerance is bounded by a preset level, regardless of the learned score. In controlled experiments, correction gain follows headroom rather than raw error; across nine frozen model-dataset pairs spanning six domains, GUARD retains most of blanket correction's gain within harm budgets that blanket correction often exceeds.

cs.AI · stat.ML

新论文 · v1arXiv:2611.23883

Accelerating Distributed Digital Ising Machines

Tong Wu, Ziyad Alswaidan, Shreya Mukherjee, Ting-Yu Lee, Daniel Xu, Atharva Raut, Harry Tye, Abhinav K. Puthran 等 14 位作者

原文摘要

Ising machines, dedicated accelerators built around the dynamics of Ising-like spin models, have emerged as a leading platform for combinatorial optimization and probabilistic sampling. The problems that motivate them require millions of spins, far beyond the on-chip capacity of any single device, so a practical Ising machine must distribute its computation across many devices. Prior work reports throughput at a single operating point, leaving the binding resource unidentified and platform crossovers measurable only by experiment; no calibrated cross-platform cost model of distributed Ising machines has been reported, nor have systematic techniques to accelerate their Markov chain Monte Carlo (MCMC) workloads. This work presents a calibrated cost model that decomposes digital Ising sweep time into compute, memory, synchronization, and inter-device communication terms, validated with 4.8% mean error across three GPU generations, FPGA and ASIC up to 1 million spins. Guided by the model, we address each term in turn and propose: (i) a data placement strategy, worth up to 4.2$\times$ speedup; (ii) multi-rate and bounded-staleness schedules that relax the barrier that Gibbs sampling imposes, cutting time-to-solution (TTS) by up to 3.16$\times$ on three benchmarks; and (iii) a phase-offset streaming dataflow that hides boundary exchange behind interior computation whenever a link satisfies a latency-transparency criterion, yielding up to 1.25$\times$/1.74$\times$/2.01$\times$ on an H200 pair, a ZCU106 pair, and a four-board XEM8320 $2{\times}2$ mesh. The streaming dataflow is also implemented on a fabricated four-chip 28 nm ASIC sustaining 28.1 Gflips/s at 4.81 pJ/flip.

cs.AR · cs.DC · cs.LG

新论文 · v1arXiv:2611.23881

NullTok: empirical-null token selection for training-free long video understanding

Zhang Ziyue, Juwei Jian, Baiyang Song, Yunshan Zhong, Jun Peng, Yiyi Zhou, Rongrong Ji

原文摘要

Long video understanding remains a daunting challenge for existing multimodal large language models (MLLMs), mainly due to the large number of visual tokens accumulated in long videos. Currently, frame selection and token selection are two main paradigms for addressing this issue, both aiming to select relevant visual information to form a compact input for MLLMs. A natural approach combines these two paradigms by selecting frames before pruning tokens, but noisy frame-level relevance estimates can cause irreversible evidence loss. In this paper, we study this problem from the perspective of assessing the spatiotemporal distribution of query-specific evidence and propose a novel and training-free approach termed NULLTOK. For every patch across all frames, we construct an empirical-null distribution from 64 group-matched irrelevant control queries, capturing the scores expected in the absence of target-specific evidence. Target affinities are calibrated against these same-position references into right-tail Monte Carlo p-values, which are discrete super-uniform under a conditional exchangeability assumption. Calibrated screening and a high-recall semantic guard produce candidate tokens for deterministic visual deduplication and joint spatiotemporal budget allocation. To validate NullTok, we apply it to two representative MLLMs and conduct extensive experiments on a set of video benchmarks. The experimental results demonstrate its effectiveness under a bounded visual token budget. For example, it achieves 65.81% and 68.85% on Video-MME using frozen LLaVA-Video-7B and Qwen3-VL-8B backbones, respectively, at a cap of 64 full-frame equivalents (FFE), without subtitles or audio. These results outperform all evaluated baselines on Video-MME under the same budget for both backbones. Code is available at https://github.com/potato-kitty/NullTok.

cs.CV · cs.AI

版本更新 · v2arXiv:2611.19244

Agentic On-Policy Distillation Implicitly Reshapes LLMs' ReAct Patterns

Deyang Kong, Zhaoyi Li, Sijian Ren, Ying Wei, Defu Lian, Shikun Zhang, Wei Ye

原文摘要

On-policy distillation (OPD) transfers specialized capabilities to language-model agents, but its cross-domain effects on interaction remain unclear. We investigate how OPD reshapes ReAct interaction patterns across mathematics, ALFWorld, and $τ^2$-bench, focusing on reasoning, actions, and observation acquisition. General-reasoning supervision reduces mathematical tool attempts from 70.4% to 14.4% while increasing accuracy from 16.7% to 38.0%; code-interpreter supervision improves household actions and retail queries. In contrast, ALFWorld supervision reduces $τ^2$-bench accuracy from 41.5% to 6.6%, accompanied by frequent tool calls that violate user-information dependencies. To address negative transfer using only source-domain training trajectories, we explore selective OPD guided by teacher preference changes before and after domain-specific RL. Retaining the top 95% of positions raises $τ^2$ accuracy from 11.5% to 29.9% with a source-learning trade-off.

cs.AI

新论文 · v1arXiv:2611.23880

A Rank-One Bypass of Fixed Linear Erasure

Jai Kumar Sharma, Vanshaj Khattar

原文摘要

Linear concept erasure can make a target unreadable to a specified family of linear classifiers applied to a model's current internal features. In an open-weight release, however, an attacker can modify the layers that produce those features. We study a fixed affine erasure layer whose upstream model is allowed to change. If the layer transmits even one direction visible to an allowed downstream classifier, the attacker can write a target score into that direction and carry it through the layer. We show that this condition is also necessary, and that for a nonconstant linear target score the required change is a rank-one affine update. Across six CIFAR-10 class-pair tasks, the route restores nearly all pre-erasure target performance and can leave a deployed benign linear head's scores unchanged to numerical tolerance. We then price the attack: with eight target labels and 2,000 unlabelled calibration examples, a 10% mean relative feature edit yields 79.8% test accuracy, which is 91.2% of the fitted score's gain over chance; tighter edit limits sharply reduce success. Linear guardedness therefore remains a valid guarantee about the current representation, but it is not by itself a tamper-resistance guarantee for an editable open-weight model.

cs.LG

版本更新 · v2arXiv:2609.06490

OracleZoom: Reference-Constrained Recursive Image Super-Resolution Beyond Ground Truth

Shubhashis Roy Dipta, Nobin Sarwar, Md Rizwan Parvez

原文摘要

Recursive image Super-Resolution (SR) reaches extreme magnification by feeding each prediction back into the same model, like zooming into an image again and again. Real ground truth for deep zooms is impractical to capture (a full $256\times$ target would be a 52 GB image), so predictions at deeper zooms have no supervision. Yet the last ground-truth image still contains the region that every deeper zoom magnifies. OracleZoom turns this region into a training signal. Inspired by on-policy self-distillation, the model trains on its own prediction for one recursive step ($4\times \rightarrow 16\times$), and a cross-scale loss projects each $16\times$ prediction back to the last ground truth at $4\times$. This projection can verify only coarse content, so a no-reference quality objective guides the remaining fine detail, and a latent prior stops this objective from adding hallucinated artifacts. On seven datasets, OracleZoom achieves the highest CLIPIQA at every zoom depth (mean 0.713 vs. 0.621 for Chain-of-Zoom), with the largest gains at the deepest zooms, even though training stops at $16\times$ and optimizes a different quality metric. Pooled over two ground-truth-anchored judges, this added detail comes at the same hallucination rate as Chain-of-Zoom at $16\times$ and a lower rate at $4\times$, and removing the cross-scale loss raises hallucination at $16\times$.

cs.CV · cs.AI · cs.CL · cs.LG

版本更新 · v2arXiv:2611.08099

GenRA: Ultra-Low-Bitrate Generative Video Compression via Representation Adaptation

Jiancheng Zhao, Jixiang Luo, Haibin Huang, Chi Zhang, Xuelong Li

原文摘要

Preserving fine visual details in ultra-low-bitrate video compression remains challenging due to the loss of critical information. Recent generative video compression methods exploit generative priors to improve perceptual quality. These pretrained generative representations are generally not optimized for compression. Some approaches therefore use an additional learned codec to map them into a separate latent space for compression. To this end, we propose GenRA, which adapts pretrained video representations for direct quantization and entropy coding without an additional learned codec. Specifically, we use vector quantization to adapt a pretrained video VAE into a discrete video tokenizer. A lightweight categorical autoregressive entropy model combines local and global context to model the probabilities of the resulting token indices. We jointly optimize the tokenizer and entropy model using a differentiable rate surrogate, allowing the encoder to adjust token assignments based on coding cost and reconstruction quality. Experiments on UVG, MCL-JCV, and HEVC Class B show average bitrate savings of 84.1% and 82.3% over GLC-Video in terms of LPIPS and DISTS, respectively.

cs.CV

新论文 · v1arXiv:2611.23877

Infinity-Wasserstein Distributionally Robust Principal Component Analysis: Exact Reformulation, Finite-Sample Guarantee, and Efficient Algorithm

Zhouxing Luo, Lei Wang, Xin Liu

原文摘要

Principal component analysis (PCA) is a fundamental and ubiquitous technique for dimension reduction and representation learning, yet its empirical formulation remains vulnerable to distributional uncertainty. While distributionally robust optimization (DRO) offers a natural remedy, existing order-2 Wasserstein models of PCA either preserve the nominal PCA subspace or lack an exact reformulation. This paper demonstrates how the infinity-Wasserstein distance can reconcile these two desirable properties. For the resulting DRO model of PCA, we derive an exact finite-dimensional reformulation and explicitly identify a worst-case distribution attaining the inner maximum. Crucially, the reformulated objective retains samplewise residual information beyond the empirical covariance matrix, revealing a mechanism through which distributional robustness can reshape the learned principal subspace. Under the mild assumption that the true distribution has a finite fourth moment, we further establish a uniform finite-sample performance guarantee for the proposed model. An iteratively reweighted majorization-minimization algorithm is also developed to solve the resulting nonconvex and nonsmooth optimization problem, with a rigorous analysis of its convergence properties. Numerical experiments substantiate the computational efficiency of the proposed algorithm and the practical effectiveness of the infinity-Wasserstein model in downstream applications.

math.OC · cs.LG · math.ST · stat.ME · stat.ML

版本更新 · v3arXiv:2605.11617

MIST: Reliable Streaming Decision Trees for Online Class-Incremental Learning via McDiarmid Bound

Phu-Hoa Pham, Chi-Nguyen Tran, Nguyen Lam Phu Quy, Dao Sy Duy Minh, Huynh Trung Kiet, Long Tran-Thanh

原文摘要

Streaming decision trees are natural candidates for open-world continual learning, as they perform local updates, enjoy bounded memory, and static decision boundaries. Despite these, they still fail in online class-incremental learning due to two coupled miscalibrations: (i) their split criterion grows unreliable as the class count K expands, and (ii) the absence of knowledge transfer at split time. Both failures share a common root: the range of Information Gain intrinsically scales with log2 K. Consequently, any Hoeffding-style confidence radius derived from it must inevitably grow with the class count, making a K-independent split criterion structurally impossible, taking away the potential benefits of applying streaming decision trees to continual learning. To fix this issue, we present MIST (McDiarmid Incremental Streaming Tree), which resolves both failures through three integrated components: (i) a tight, K-independent McDiarmid confidence radius for Gini splitting that acts as a structural regulariser; (ii) a Bayesian inheritance protocol that projects parent statistics to child nodes via truncated-Gaussian moments, with variance reduction guarantees strongest precisely when splitting is most conservative; and (iii) per-leaf KLL quantile sketches that support both continuous threshold evaluation and geometry-adaptive leaf prediction from a single data structure. On standard and stress-test tabular streams, MIST is competitive with global parametric methods on near-Gaussian benchmarks and uniquely robust on non-Gaussian geometry where SOTA benchmarks collapse.

cs.LG · math.ST

新论文 · v1arXiv:2611.23875

In Situ Training of Optical Neural Networks by Coding Agents

Ilker Oguz, Colin Christian, Christophe Moser, Demetri Psaltis, Gordon Wetzstein

原文摘要

Optical neural networks promise fast, energy-efficient computing, but every device departs from its design and must be trained with the hardware. In situ training relies on an expert-built calibrated digital model of each device or on gradient-free trial and error. Coding agents, large language models that write and run programs, have shown an emergent ability to understand and edit the weights of digital neural networks. Here we introduce optics as code, a method in which a coding agent trains an optical neural network from measurements alone, through a harness that returns camera images of its probes and requires a hypothesis per request. We discovered that the agent forms a physical understanding of the optics and reparametrizes the network with interpretable optical elements tuned on the hardware. It outperforms gradient-free methods in classification and image encoding, approaching backpropagation through a calibrated model at a fraction of the cost. Needing neither per-device calibration nor gradients through the hardware, and improving with more capable models, optics as code offers a practical route to large, scalable optical neural networks and a general tool for calibrating and optimizing optical experiments.

physics.optics · cs.AI · cs.LG

版本更新 · v2arXiv:2611.22925

Code World Model 2.0: Towards Infinite Controllable Video Generation

Ruoyu Wang, Hao Zhong, Yanming Yang, Jiahao Tian, Lin Liu, Biao Gong, Guoshan Lu, Yiding Tian 等 11 位作者

原文摘要

Visual proxies decouple explicit world-state simulation from photorealistic video synthesis, offering a controllable interface for both offline clip creation and real-time interactive streaming. Yet a practical renderer must address two challenges. Upstream proxies vary from coarse layouts to fine articulated geometry depending on the interaction, and streaming inference must operate in a few causal steps without accumulating drift across self-generated history. We present Invoker-Render, pairing a bidirectional creation model with a causal four-step streaming model under a shared conditioning interface over structural proxies, an appearance reference, and language. Trained on an aligned approximately 200-hour multi-granularity corpus, Invoker-Render honors fine geometric constraints when supplied while synthesizing plausible local motion under coarse proxies, enabling runtime selection of input detail. To bridge bidirectional creation and real-time streaming without compounding step compression and rollout drift, we introduce Progressive Forcing (ProForcing). ProForcing first trains a causal predictor and distills it into a four-step renderer using histories derived from training videos. Training then shifts to continuous self-generated rollouts. A pipelined serving stack supports efficient streaming inference. Experiments show versatile control across proxy families, granularities, and camera paths, consistent gains over recent streaming baselines, and stable 30-minute racing rollouts that preserve scene and subject appearance.

cs.CV

版本更新 · v3arXiv:2607.28422

QAdapt: Noise-Adaptive Neural Pre-Decoding for Surface-Code Quantum Error Correction

Ran Miao, Rui Luo, Xiaohan Shan, Xiaoming Sun

原文摘要

Fault-tolerant quantum computing relies on quantum error correction to protect logical information, requiring classical decoders that keep pace with syndrome generation. Heterogeneous, changing hardware noise and the distribution shift from simulation to experiment challenge the accuracy and robustness of fixed neural decoders. We present QAdapt, a noise-adaptive neural pre-decoding framework for surface-code quantum error correction. QAdapt learns local spatiotemporal correlations, continually adapts across noise environments, and passes the residual syndrome to a conventional global decoder. Online elastic weight consolidation limits forgetting as training moves between measurement, idle, CNOT, and Pauli-bias noise regimes. At distance nine, the HTNet pre-decoder reduces equivalent per-round logical error by 15.64% relative to Ising-Fast, with 28.75% fewer parameters and 9.07% less matching-backend time. Online EWC reduces average forgetting by 76.52% relative to sequential fine-tuning. With weights frozen, it reduces equivalent per-round logical error by 5.66% relative to continued base-environment training on a fixed valid domain of 331 unseen compound-noise conditions; whole-sequence failure probability falls by 2.54% over all 396 conditions. On native Willow records, the corresponding per-round reductions are 19.44%, 17.47%, and 17.95% at 10, 30, and 50 rounds, without target-domain fine-tuning. These results show how adaptive neural pre-decoding can improve logical accuracy, preserve performance across noise regimes, and reduce residual matching workload within an established decoding pipeline.

cs.LG · quant-ph

新论文 · v1arXiv:2611.23872

Keep the Image in Mind: A Visual Board for Multimodal Agents

Haiming Zhu, Zhiwei Chen, Ding Cao, Tianxiang Zhao

原文摘要

Multi-step visual reasoning suffers when early visual misinterpretations persist as incorrect premises in subsequent reasoning. We argue that how visual evidence is represented during reasoning matters as much as how much of it is gathered. Rather than appending observations and inferences to an ever-growing dialogue, we introduce a Visual Board-centered harness that separates visual observations, uncertain interpretations, and derived reasoning within a structured, capacity-bounded, and revisable workspace. This separation makes each premise explicit and auditable, so that conflicts between what is seen and what is derived become detectable even by a text-only critic and can trigger targeted reinspection of the original image. Without task-specific training, our harness consistently outperforms standalone models across 4B, 8B, Qwen3-VL-Flash, and 32B configurations on Euclid30K, SolidGeo, and ChartQA bar charts, with gains of up to 11.68, 18.42, and 24.82 percentage points, respectively. Under identical tools, critic, and interaction limits, the structured Board outperforms shared free-form notes by 6.67 to 13.00 points and no-Board retries by 7.35 to 18.80 points while using 7.9\% to 38.2\% fewer tokens. Supervised fine-tuning on harness trajectories further improves direct answering and critic-free tool use, with positive transfer to chart reasoning.

cs.CV

新论文 · v1arXiv:2611.23871

HOForce: Motion and Force Reconstruction of Hand-Object Interaction Using a Single-View RGB-D Video

Jiahui Wang, Wanqing Du, Yuanbin Zou, Zhefeng Xu, Feng Xu

原文摘要

Reconstructing hand-object interactions is an important task, and achieving it with a single camera makes large-scale capture practical but challenging due to heavy occlusions. We introduce HOForce to track hands and objects, and further recover interaction forces from single-view RGB-D video and known object properties. For two-hand manipulation, we jointly model fingertip forces of both hands and the contact forces between objects when two objects are presented. The reconstructed forces also improve motion tracking by giving pose constraints in the occluded regions. HOForce recovers hand and object motion more accurately on DexYCB and TACO and produces less hand-object penetration than interaction baselines on DexYCB. On our dataset of fingertip forces recorded with tactile sensors, HOForce reduces normal-force MAE by 24.5% relative to the strongest baseline. The recovered forces reproduce object motion in MuJoCo and transfer to other engines after adaptation.

cs.CV

版本更新 · v2arXiv:2407.10445

I2I Backdoor: Backdoor Attacks against Image-to-Image Tasks

Wenbo Jiang, Hongwei Li, Jiaming He, Rui Zhang, Guowen Xu, Tianwei Zhang, Rongxing Lu

原文摘要

With the rapid development of deep learning technology, deep learning-based Image-to-Image (I2I) networks have become the predominant choice for I2I tasks like image super-resolution and denoising. Despite their remarkable performance, the security of I2I networks has not been thoroughly investigated. While some studies have probed their susceptibility to adversarial attacks, none have explored the backdoor attack against I2I networks, which is a more stealthy and severe threat. In this work, for the first time, we comprehensively investigate the vulnerability of I2I networks to backdoor attacks. We propose a backdoor attack against I2I tasks, where the backdoored I2I network behaves normally on clean input images, yet outputs a specific inappropriate image when the backdoor trigger appears on the input image. To achieve such an I2I backdoor attack, we design a universal adversarial perturbation (UAP) generation algorithm for I2I networks, where the generated UAP is used as the trigger for the I2I backdoor. Besides, multi-task learning (MTL) with dynamic weighting methods is employed in the backdoor training process to gain better results. Expanding our focus beyond I2I tasks, we extend our I2I backdoor to attack downstream tasks, including image classification and object detection. Specifically, the backdoor-triggered image processed by the backdoored image denoising network can fool the downstream image classifiers and object detectors. Extensive experiments demonstrate the effectiveness of the I2I backdoor on state-of-the-art I2I network architectures as well as the robustness against different backdoor defenses.

cs.CV · cs.AI · cs.CR

新论文 · v1arXiv:2611.23867

Control-Aware Separating Hyperplanes for Control Barrier Functions in Planar Navigation

Aadarsh Ramachandiran, Rachel Kalpana Kalaimani

原文摘要

Separating hyperplanes are widely used in Control Barrier Function (CBF)-based collision avoidance, yet their orientations strongly govern control intervention-the deviation of the applied control from the robot's nominal command. Existing methods typically select hyperplane orientations from geometry alone. Jointly optimizing orientations and control yields a continuous, non-convex problem in which local nonlinear solvers suffer from initialization sensitivity, local minima, and unpredictable latency, while discretized orientation grids scale exponentially with obstacle count. For planar navigation of a polygonal robot among polygonal obstacles, we present a solver-free, control-aware separating hyperplane selection method that directly computes a global optimum of this joint problem at each step whenever it is feasible. Exploiting the geometry of Minkowski difference polygons, we reduce this continuous non-convex optimization to evaluating a polynomial-size set of candidate control interventions via closed-form algebraic checks, selecting the feasible candidate of minimum norm. Across 250 simulated navigation scenarios, our method reduces cumulative intervention cost by 3.6% at the median and 27.3% at the 90th percentile compared to a local nonlinear solver, and completion time by 0.6% and 5.6%, respectively. In scalability benchmarks, it achieves a 6.1x per-step speedup at 16 simultaneous obstacles.

eess.SY · cs.RO

版本更新 · v4arXiv:2609.18366

Bad Genius: Counterfactual-Guided Harness Evolution Beyond Task-Specific Shortcuts

Guojun Zhu, Xunheng Huang, Peng Yin, Jiahui Xie, Sanguo Zhang, Doudou Zhou

原文摘要

An optimized agent harness can improve on held-out tasks by relying on formatting conventions that those tasks share with the rest of the benchmark. We introduce Counterfactual Harness Search and Evolution (CHASE) to test whether harness improvements survive changes to the evaluation protocol. A Proposer edits the harness around a fixed foundation model. A Challenger then searches for a protocol change that preserves the tasks but reduces the updated harness's gain over the initial harness. Executable checks test whether the transformation preserves task semantics. If the gain loss is confirmed on held-out tasks, CHASE adds the transformation to an archive that constrains subsequent harness updates. We give conditions under which this finite archive bounds the gain loss on an ideal benchmark with shortcuts removed. We evaluate CHASE on Syn-Ledger, OfficeQA, FinanceBench, and SWE-Bench Pro V2. On Syn-Ledger, removing the known shortcuts yields 38.30% accuracy for CHASE and 18.27% for HarnessEvolve. On the 21 certification tasks from SWE-Bench Pro V2 HARD-51, changing the issue formatting and tool-output field order reduces HarnessEvolve's gain over the initial harness from 8.57 to 3.81 percentage points. The final CHASE harness retains 10.48 points of its 11.43-point gain under the same change.

cs.AI · cs.LG · stat.ML

版本更新 · v2arXiv:2611.07573

Integrating Learned Priors through Prior-Space Reparameterization for Physically Based Inverse Rendering

Hun Hwangbo, Minjun Park, Yongsung Kim, Wooseok Song, Jungmin Ko, Junghyun Hwang, Sungroh Yoon

原文摘要

Physically based inverse rendering is an ambiguous problem, allowing independently optimized material variables to absorb appearance variations caused by illumination. Learned material predictions provide useful cues about which surface points have similar materials. Existing approaches often incorporate these cues through auxiliary regularization losses. Instead, we propose a novel prior-guided reparameterization of the material optimization space. We fuse multi-view material predictions onto a mesh and construct a permutohedral lattice in the resulting prior feature space. Surface points with similar predictions share lattice variables, restricting the material field to a prior-defined subspace that is optimized jointly with geometry and illumination using a purely photometric objective. Experiments on three synthetic benchmarks demonstrate improved material decomposition and relighting, with results on par with the strongest baseline on Stanford-ORB. As a complete system, our method improves relighting PSNR by 1.4-1.5\,dB over the strongest evaluated mesh-based baseline on Synthetic4Relight and TensoIR-Synthetic. Analysis of the optimization shows that shared variables pool per-vertex photometric gradients into updates that are more consistent across sampled views. In a controlled comparison within the same pipeline, the restricted subspace, despite its reduced degrees of freedom, yields more accurate material decomposition than both unrestricted per-vertex optimization and a regularization loss built from the same prior.

cs.CV

版本更新 · v2arXiv:2602.10182

Signature-Kernel Based Evaluation Metrics for Robust Probabilistic and Tail-Event Forecasting

Benjamin R. Redhead, Thomas L. Lee, Peng Gu, Víctor Elvira, Amos Storkey

原文摘要

Standard sample-based metrics like Continuous Ranked Probability Score (CRPS) and Quantile Loss (QL) are frequently used in the evaluation of probabilistic time-series forecasting. However, these metrics fail to capture multivariate correlations or assess the ability to capture distribution tails. These global metrics are insensitive to errors on high-utility events in distributional tails due to over-representation of the distribution's body. To accurately measure distributional fidelity on tail-regions and regions of high utility while still having a proper scoring rule, we introduce a family of censored signature kernel metrics. Our proposed metrics concentrate evaluation of forecasts to a focus region, representing high-utility or distributional tails, by collapsing the body of the distribution to a single pivot. Our benchmarks on time-series foundation models (TSFMs) reveal that while a clear ranking can be formed for distribution capture, there is no clear winner for tasks like systemic load prediction. This indicates that models with a strong ability to capture an overall distribution do not produce forecasts with high downstream utility. To encourage the use of the proposed metrics we open-source the efficient signature kernel (ESK) library. This library facilitates batch computation, with custom triton kernels, achieving a speed-up of up to 3.58$\times$ compared to implementations via the popular SigKernel library. Code for our metrics are available on GitHub.

cs.LG · stat.ML

版本更新 · v2arXiv:2407.06676

Games played by Exponential Weights Algorithms

Maurizio d'Andrea, Fabien Gensbittel, Jérôme Renault

原文摘要

This paper studies the last-iterate convergence properties of the exponential weights algorithm with constant learning rates. We consider a repeated interaction in discrete time, where each player uses an exponential weights algorithm characterized by an initial mixed action and a fixed learning rate, so that the mixed action profile $p^t$ played at stage $t$ follows an homogeneous Markov chain. At first, we show that whenever a strict Nash equilibrium exists, the probability to play a strict Nash equilibrium at the next stage converges almost surely to 0 or 1. Secondly, we show that the limit of $p^t$, whenever it exists, belongs to the set of ``Nash Equilibria with Equalizing Payoffs''. Thirdly, we show that in strong coordination games, where the payoff of a player is positive on the diagonal and 0 elsewhere, $p^t$ converges almost surely to one of the strict Nash equilibria. We conclude with open questions.

cs.AI · math.PR

新论文 · v1arXiv:2611.23865

On Demand, Not Uniformity: Adaptive Expert Orchestration for Multimodal Continual Instruction Tuning

Mengxin Qin, Kun Wei, Weitao An, Jiahua Li, Xu Yang, Cheng Deng

原文摘要

Multimodal Continual Instruction Tuning (MCIT) enables Multimodal Large Language Models (MLLMs) to adapt to sequential instruction data while retaining previously acquired knowledge. While recent Mixture-of-Experts (MoE)-based methods mitigate forgetting by freezing historical experts and introducing new ones, they largely overlook the varying adaptation demands across tasks. Probing incoming tasks reveals pronounced heterogeneity in layer-wise gradient pressure and token-wise prediction difficulty. This observation highlights two unresolved decisions: where additional capacity is needed and how strongly new experts should contribute to individual tokens. To address both questions, we propose the Coverage-Calibrated Expert Injection and Residual Arbitration framework (CIRA-MoE), which derives expert acceptance margins via learnable threshold routing to serve as a unified coverage signal. Specifically, historical acceptance evidence guides the selective injection of new experts into insufficiently covered layers. Meanwhile, relative evidence between historical and current experts modulates token-wise residual contributions, coupled with arbitration-consistent regularization to promote expert specialization. Comprehensive experiments on UCIT and CoIN demonstrate that CIRA-MoE outperforms existing state-of-the-art methods in MCIT.

cs.CV

新论文 · v1arXiv:2611.23864

Asking but Missing: Decision Coverage in Natural-Language Data Querying

Hangle Hu, Chenyu Hou, Bin Cao, Jing Fan, Wei Luo

原文摘要

Large language models are increasingly used to translate natural language questions about structured data into executable programs. A fundamental challenge is recognizing when a question admits multiple computational interpretations and identifying the decisions needed to resolve them. We investigate whether language models recognize these unresolved decisions before generating code, or silently adopt assumptions that may lead to incorrect results. We refer to the latter behavior as Underspecified Decision Completion (UDC). To distinguish different causes of UDC, we introduce Grounded Code Generation (GCG), a diagnostic protocol that separates the identification of unresolved decisions, their resolution using evidence, and their translation into executable code. By examining each stage separately, we can determine where the decisions responsible for incorrect results were first missed or mishandled. We evaluate this framework on BIRD-Pandas with twelve models. The results show that answering clarification questions consistently improves execution accuracy, yet substantial errors remain. Our findings suggest that the main difficulty lies not in answering clarification questions, but in recognizing which questions need to be asked in the first place.

cs.AI · cs.SE

新论文 · v1arXiv:2611.23862

Grounding Radiology Reports in 3D Chest CT: Results of the ReXGrounding Challenge

Mohammed Baharoon, Mahmoud Hussain Alabbad, Sung Eun Kim, Xiaoman Zhang, Lucas Bijnens, Zirong Chen, Jiayu Wang, Xi Ouyang 等 62 位作者

原文摘要

Segmenting each reported finding in a radiology scan would allow findings to be measured and tracked over time and let AI systems generate reports that are verifiable, but so far AI models have performed poorly at this task. To advance this task, we organized the ReXGrounding Challenge at MICCAI 2026, the first challenge dedicated to it. Given a non-contrast chest CT scan and the set of findings from its report, participants trained models to produce the corresponding set of 3D segmentations. The challenge used 2,992 training and 200 validation scans from the ReXGroundingCT dataset, and evaluated methods on a held-out test set of 300 scans with 582 findings, exhaustively annotated by board-certified radiologists. In total, 48 teams from 15 countries made 760 submissions, and the best method reached a Dice of 0.375 and a hit rate of 0.77, matching agreement between two independent radiologist groups (Dice 0.32, hit rate 0.76, on a 50-scan subset). On the 100 scans of the pre-challenge benchmark, the winner reached a Dice of 0.390, against 0.285 for VoxTell, the best model before the challenge, which five of the six best teams fine-tuned. Non-focal findings remained far harder to segment than focal ones, instance F1 stayed below inter-reader agreement (0.34 vs. 0.47), and models sometimes segmented the right abnormality in the wrong location. Top methods thus match radiologist agreement in Dice, but non-focal findings, findings with multiple instances, and adherence to the anatomical location described in the text remain open problems.

cs.CV · cs.AI

版本更新 · v2arXiv:2610.16825

One Checkpoint, Many Runs: Recipe Cards for Fine-Tuned Models

Jaewon Choi, Jihye Kang, Namhyuk Ahn

原文摘要

Fine-tuned language models are shared with model cards that describe the training setup in detail, often down to the default random seed, but report the score of a single released checkpoint. Such setups are retrained routinely, by fine-tuning services for each customer, by organizations at each update, and by users on their own data, so the model that is deployed comes from a new run of the setup. Some runs fail, and a failed run can go unnoticed and reach the people the model serves. We call such a setup a model recipe and argue that its documentation should state how often a run of the recipe misses a required performance bar. In a survey of the model cards of 300 fine-tuned language models on Hugging Face, none reports this failure probability, and only three report results from more than one training run. Across 2,124 fine-tuning runs, we find that reading the mean, the median, or a bootstrap interval as an approval often approves recipes whose failure rate exceeds a stated failure tolerance, because these summaries do not measure how often a run fails. Model cards do not report how often the runs of a recipe fail, although whoever runs the training is best placed to count them, while those who rely on a retrained model bear the cost of a failed run. To report this failure probability directly, we propose recipe cards, an addendum to model cards that documents the model recipe behind a checkpoint. A recipe card states the bar and the tolerance and who chose them, reports how many runs are made and how many of them miss the bar, gives an upper bound on the failure probability, and warns when this bound exceeds the tolerance. Recipe cards record who set and who accepted this risk, and let deployers and regulators state acceptance criteria for retrained models.

cs.LG

版本更新 · v2arXiv:2611.18627

FlashREINFORCE: Critic-Free Single-Rollout Asynchronous RL for Agentic Language Models

Jian Hu, Yifan Zhang, Hao Zhang, Binfeng Xu, Shaokun Zhang, Hongqing Peng, Zhiding Yu, Pavlo Molchanov 等 10 位作者

原文摘要

Asynchronous reinforcement learning suits long-horizon agents with irregular rollout times, but group-relative updates require repeated rollouts of each prompt. However, group-based sampling limits prompt coverage at a fixed rollout budget and introduces group synchronization barriers. One rollout per prompt avoids these constraints but requires learning from stale, variable-length trajectories without a sibling baseline. We present FlashREINFORCE, to our knowledge the first critic-free single-rollout framework that scales to long asynchronous training and large models. It combines One-Batch REINFORCE, a Sequence Trust Region, and Sample-Mean Optimization, which provide prompt coverage and batch centering while controlling policy drift and length-amplified negative updates. Each component is easy to implement, and the result is their combination: a batch-mean baseline is stable under sample-mean loss, and each common alternative degrades or collapses training. FlashREINFORCE remains stable through 6,000 updates on DeepSeek-R1-Distill-Qwen-1.5B at policy lag four, and through 1,450 on the 30B MoE Qwen3-30B-A3B at lag eight. Under identical loss, data, and lag it matches a strengthened group-relative baseline within 0.5 points, and with half the rollouts exceeds published synchronous baselines on Qwen2.5-Math-1.5B (38.0 versus 36.3 five-benchmark mean). On Qwen2.5-7B-Instruct it maintains tool use where GRPO stops calling tools and achieves 98.3%/96.5% seen/unseen ALFWorld success.

cs.CL · cs.LG

新论文 · v1arXiv:2611.23860

MobileWeave: Program-First Synthesis of Dependency-Linked Tasks for Training Mobile GUI Agents

Ziyi He, Xian Wu, Qiyuan Chen, Haitao Li, Cehao Yang, Yinghao Zhu, Xichen Zhang, Xiao Chen 等 10 位作者

原文摘要

Native mobile GUI agents require large-scale training trajectories covering both individual app functions and multi-goal workflows within and across apps, which makes automatic task synthesis a promising way to scale training data. Existing synthesis pipelines, however, produce each task as free-form text, leaving implicit how its subgoals depend on one another. This matters because composition is a skill in its own right: a policy may complete each subtask of a workflow on its own, yet fail to chain these subtasks into the complete workflow. We present MobileWeave, which instead synthesizes every task as an executable program derived from the app implementations of a controllable simulator. MobileWeave abstracts each app into a typed capability graph with semantic types shared across apps, composes atomic programs into single- and cross-app composite programs through explicit value and object dependencies, and verifies each program by native execution, deriving both its instruction and its completion check from the same program. Trained on teacher trajectories collected for these tasks and on recovery supervision at the policy's own stalls, MobileWeave-8B improves its base model by 26.7 points on MobileGym and by 8.6, 21.3, and 17.6 points on AndroidWorld, AndroidLab, and MobileWorld, with comparable gains at 4B. Beyond these gains, we find that composition is learnable from task structure: explicit dependencies between subgoals act as a training signal that teaches a policy to compose skills it already masters. We release data, model, and code at https://maxxnotmarx.github.io/MobileWeave/ .

cs.CV

新论文 · v1arXiv:2611.23858

LAGR: Language-Aware Grasp Re-Ranking for Task Oriented Grasping

Manav Kulshrestha, Shreyans Jain, Apoorva Vashisth, Ankush Gupta, Damon Conover, Aniket Bera

原文摘要

Grasping is a fundamental capability for robotic manipulation, with the appropriate grasp depending on the downstream task. For example, a hammer should be held by its handle for hammering but by its head when it is handed over. Selecting such a grasp requires consideration of two criteria: physical feasibility on the object and suitability for the task. Developing such an approach requires dense grasp labels with both feasibility and task scores on real observations, but existing task-oriented grasping datasets are small, synthetic, or costly to annotate. To address this gap, we construct a dataset of 2.68 million scored 5-DoF grasp annotations over 100,911 image-task pairs from $17$ object categories. Additionally, we introduce LAGR, a task-oriented grasp detection method that explicitly addresses both these challenges. Given vision-language inputs, LAGR first generates task-independent grasp proposals along with a physical feasibility score, and then scores the generated grasps with a language-conditioned re-ranker that reasons over all proposals jointly. Experiments show that LAGR outperforms the next best method by 9.3 percentage points. We further validate LAGR on real objects by executing task-oriented grasps with a Franka Research 3 robot.

cs.RO · cs.AI · cs.LG

新论文 · v1arXiv:2611.23859

Temporal Image Representations for Mouse Dynamics Authentication

Junyi Li, Charles Devlen, Daqing Hou

原文摘要

Mouse dynamics authentication has traditionally relied on handcrafted features or time-series representations that capture both spatial and temporal characteristics of mouse movements. More recent approaches have modeled mouse data as images, enabling the use of powerful pretrained computer vision models. However, prior work primarily focuses on spatial-only representations where the cursor coordinates are plotted as a trajectory image. These representations discard the temporal characteristics which can help discriminate users. In this work, we propose image representations that incorporate speed and velocity information for mouse dynamics authentication. We investigate two mouse data representations, a XYPlot representation that intuitively shows the path of movement, and a Pairwise Distance (PWD) representation that utilizes the entire image canvas to encode relationships between mouse events. Both representations are evaluated under both location-aware and location-free settings using pretrained 2D-CNN and ViT architectures. We found that the incorporation of temporal information improved performance over spatial representations alone, and that the location-aware variants generally outperformed the location-free. We achieve state-of-the-art performance on two datasets, Balabit and TWOS, and extend our evaluation to two other datasets. We found that in general, location-aware PWD with temporal encoding combined with the 2D-CNN architecture provided the best results, with mean EERs of 5.21% and 12.70% on the Balabit Test-1 and Test-2 sets, respectively, 4.94% on TWOS, 6.16% on ChaoShen, and 0.59% on DFL.

cs.CV

版本更新 · v4arXiv:2607.21653

MOLT: A Hugging Face-Native, Minimal, and Scalable Framework for Agentic Reinforcement Learning

Jian Hu, Huiying Li, Hao Zhang, Binfeng Xu, Yifan Zhang, Shaokun Zhang, Hemil Desai, Michael Demoret 等 11 位作者

原文摘要

Agentic reinforcement learning (RL) research iterates quickly on agents, rewards, and learning algorithms, while policies have grown to hundreds of billions of parameters and beyond. Existing open frameworks force a choice: those that are easy to modify do not scale to the largest models, and those that scale rely on Megatron, which needs dedicated bridge code for every model family and several times more framework code. We present MOLT, an open-source framework built for research that is Hugging Face-native, minimal, and scalable. Its code is easy to modify: the entire RL path is approximately 9.2K lines of Python with one training backend and one rollout engine, so a researcher or an AI coding agent can read it in full and edit the code that actually runs. A Hugging Face checkpoint is loaded directly into a PyTorch FSDP2 trainer with tensor, expert, and context parallelism, so a new model family needs no bridge, and the same code scales from a 4B model to a 753B one. Existing OpenAI- and Anthropic-SDK agents are trained by redirecting their client, with automatic handling of agents that rewrite their own context; training is fully asynchronous; and trajectory data stay in a distributed object store. With MOLT, a 30B mixture-of-experts tool agent trains stably for 1,464 asynchronous updates, a visual computer-use agent improves on OSWorld, asynchronous training is 2.40 times faster than synchronous training, and pipeline throughput reaches 92% of a Megatron-based stack. Its trainer, NeMo AutoModel, trains models of one to 2.8 trillion parameters.

cs.LG · cs.CL · cs.DC