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

+979新增论文
+781其中 AI 相关
+119已有论文修订
+95新增 AI 导读

arXiv 最近完成扫描:2026-11-20 08:48 · 全站最近入库:2026-11-20 19:58各信源进度 →

统计口径与数据新鲜度

统计 2026-11-20 09:10 — 2026-11-21 09:10(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:2026-11-20 08:48最近发现新增版本:2026-11-20 19:58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。

28,961 已收录论文24,030 AI 相关8,912 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 588 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-20 08:46(北京时间)
新论文 · v1arXiv:2611.26342

Training LLMs to Follow Reasoning Instructions with Minimal Edits

Alexandra Yakovleva, Luat Gia Khoi Nguyen, Juho Kannala, Harri Valpola, Alexander Ilin

原文摘要

Large language models increasingly rely on extended reasoning traces, yet the training signals used to shape this behavior remain coarse. Supervised fine-tuning on full trajectories forces the model to imitate an entire demonstration, overconstraining its output distribution, while trajectory-level reinforcement learning provides only weak credit assignment, rewarding or penalizing whole completions without identifying where they succeed or fail. We propose edit-based training, a general framework for learning from a model's own trajectories via localized textual corrections. Given a sampled completion, an editor identifies the first point of failure under a specified criterion and proposes a minimal replacement span. The model is then supervised only on these edited spans. By design, this concentrates supervision where the model's reasoning first breaks down and aims to preserve its existing reasoning patterns elsewhere. The method applies both offline and online, with the latter continually adapting supervision to the model's evolving failure modes. We instantiate this approach on reasoning instruction following with Qwen3-32B on the ReasonIF benchmark. Compared to full-trajectory imitation and trajectory-level reinforcement learning baselines, edit-based training substantially improves compliance with the reasoning instruction while largely preserving the model's skills on untargeted tasks. These results suggest that localized correction is an effective complement to these approaches for shaping how LLMs reason.

cs.AI

版本更新 · v3arXiv:2607.26921

From Keypoints to Predictive Distributions: Post-Hoc Uncertainty for YOLO-Pose Models

Alexej Klushyn, Juan Rivero Sesma, Florian Seligmann, Richard Kurle, Kinh Tieu, Jayant Sen Gupta

原文摘要

YOLO-Pose models provide efficient multi-object keypoint localization, but do not quantify the associated spatial uncertainty. We introduce a lightweight post-hoc probabilistic extension that augments a trained YOLO-Pose model with calibrated bivariate predictive distributions over keypoint locations, centered at the model's original predictions. Concretely, we train additional probabilistic heads with an importance-weighted negative log-likelihood to predict an input-dependent $2\times2$ dispersion matrix for each keypoint, followed by Gaussian calibration for broad downstream compatibility or Student-$t$ calibration for distributional fidelity. Complementing this, we propose an evaluation protocol that combines a suite of distributional calibration diagnostics with average keypoint precision (AKP), a keypoint-level extension of the COCO AP protocol for assessing reliability rankings. Experiments on COCO with YOLO-NAS-Pose and YOLO26-Pose show that the learned uncertainty estimates enable keypoint-level reliability ranking, Student-$t$ calibration best captures the empirical residual distribution, and uncertainty-based pruning removes unreliable keypoints. In vision-based aircraft landing, calibrated predictive distributions over runway keypoints support uncertainty-aware position estimation and downstream sensor fusion.

cs.CV

版本更新 · v2arXiv:2605.30727

MosaicLeaks: Privacy Risks in Querying-in-the-Open for Deep Research Agents

Alexander Gurung, Spandana Gella, Alexandre Drouin, Issam H. Laradji, Perouz Taslakian, Rafael Pardinas

原文摘要

Deep research agents increasingly combine private local documents with external tools like web retrieval, creating a privacy risk: an agent's external queries may leak sensitive information from its local context. This risk is amplified by the mosaic effect, where individual queries may appear harmless but become revealing in aggregate. We introduce MosaicLeaks, a benchmark of 1,001 multi-hop deep research tasks that chain private enterprise documents and a public web corpus, forcing agents to make external queries that depend on local information. We evaluate leakage with an adversary LLM that observes only the agent's external queries and attempts to infer private information at three levels: the agent's research intent, answers to specific private questions and verifiable claims about the enterprise documents. We find that models across families and sizes frequently leak at all three levels, that zero-shot privacy prompting has inconsistent effects across models and that reinforcement learning for task performance alone worsens leakage. To address this, we propose Privacy-Aware Deep Research (PA-DR), an RL framework that combines situational rewards for task success with a learned privacy classifier to provide dense credit assignment over both per-query and mosaic-level leakage. Training Qwen3-4B-Instruct with PA-DR improves accuracy from 48.7% to 58.1% and reduces Privacy Leakage from 34.0% to 6.5%.

cs.CL

新论文 · v1arXiv:2611.26172

Budget Internalization: Fixed-Budget RLVR Trades Test-Time Scaling for In-Budget Accuracy

Ilia Zavidnyi, Alexander Kovrigin, Arina Puchkova, Kirill Gelvan, Sergey Titov

原文摘要

Reasoning models often become more accurate when they can generate longer answers at test time. This behavior is known as test-time scaling. However, RL with verifiable rewards (RLVR) usually trains them with a fixed rollout cap. We find that this cap changes model behavior, not only training cost. After training, models use few additional tokens beyond the training budget. They outperform the base models near that budget, but their accuracy stops improving and falls below the base model at larger budgets. The rollout cap thus becomes a learned constraint, that limits model's test-time scaling. We call this effect budget internalization. We observe it across Qwen3.5-4B, Gemma4-E2B, and Nemotron-3-Nano-4B and two reasoning training tasks: mathematical reasoning and diverse, procedurally generated problems. This effect also transfers from math training to science and coding tasks. Training-objective level ablations on Qwen3.5-4B show that most changes recover little test-time scaling. Stronger KL regularization and lower training-data truncation preserve more scaling, but stronger KL reduces in-budget accuracy. Forcing an answer on truncated training rollouts largely removes budget internalization, but also removes most of the in-budget gain. SFT on artificially truncated traces reproduces budget internalization, whereas SFT on naturally short traces does not, which shows that the effect is not specific to RL. This result links budget internalization to length-censored post-training rather than to an RL-specific objective term. Thus, a rollout cap does not only control training cost. It can trade low-budget gains for lower performance when additional inference compute is available.

cs.LG · cs.AI

版本更新 · v2arXiv:2611.18729

SCALE: Towards Whole-Slide Counterfactual Explanations in Cancer Histopathology

Weiyi Qin, Yingci Liu-Swetz, Shiwei Tan, Yi Wang, Mohammad Alexanderani, Dimitrios Korentzelos, Peizi Li, Soo Yeon Sohn 等 14 位作者

原文摘要

Accurate cancer diagnosis depends on histopathological evaluation, in which tumor morphology drives disease classification and therapeutic decision-making. While deep learning models have demonstrated strong performance in predicting diagnostic and prognostic properties from histopathology images, their clinical adoption remains limited by a lack of interpretability. Unlike conventional medical images where diagnostic findings are spatially localized, histopathological images are characterized by complex textural patterns distributed across the entire whole slide image (WSI), making it difficult to explain model predictions through region-based localization alone. Counterfactual explanations offer a promising alternative by asking "what changes in the image could have altered the model prediction." To achieve this, we propose SCALE, a framework for generating such counterfactual explanations for histopathological DL models. Across nine counterfactual directions spanning multiple cancer types, SCALE generates counterfactual explanations that consistently shift model predictions toward the intended diagnostic target while maintaining biological feasibility. In a blinded review by pathologists, SCALE demonstrated superior morphological plausibility compared to baseline methods (win rate of 91.7%) and robust counterfactual effectiveness (target consistency of 96.7%). These results suggest that SCALE generates morphologically plausible and effective counterfactual explanations that can be applied to improve the interpretability of deep learning models for histopathology.

cs.CV

新论文 · v1arXiv:2611.26003

AI agents can redesign engineering parts with more than a thousand CAD parameters and nonlinear simulations

Alexandra Starostina, Alexander Hermann, Mahish Guru, Kartik Bali, Roland Aydin, Christian C. Cyron

原文摘要

Simulation-driven design of real engineering components has to meet three conditions at once. It runs on few solves, since a nonlinear finite-element simulation of a 3D part takes hours; its design space has many values, the hundreds to thousands of values of the part's CAD model; and a better design often needs a redesign, coordinated changes of the part's proportions rather than local polishing of its surface. Automated optimisers meet at most two of these conditions. Classical optimisers, Bayesian or derivative-free, need a person to cut the design down to a few parameters, and on all the values they need far more solves; industrial shape optimisers move every surface node of the mesh in one solve per cycle, but only in small steps that polish rather than redesign. We ask whether a general-purpose large language model (LLM) can meet all three by joining what it knows with what the simulation shows. GPT-6 Astra edits a magnesium stent through a CAD geometry parametrisation of about 1,650 values, with a budget of 15 simulations; every proposal is built in CAD, meshed and simulated with large-deformation plasticity and contact, and the LLM reads the full results of every solve. Within that budget it lowered the objective, a high percentile of the strains in the stent's bends, by 32%; on the same budget, the industrial shape optimiser Tosca improved the original design by a tenth and then stopped improving. An earlier GPT-6 Astra design, 25% lower in the same objective, was checked in an independent, higher-fidelity model of the full deployment with crimping and damage: it was 14% less strained than the original after the full chain, and its most damaged spot began to fail later and carried 63% less damage at full expansion. LLMs can thus join their learned knowledge with simulation data to adjust more than a thousand design values and reach valid designs with real improvements.

cs.AI

版本更新 · v2arXiv:2611.00686

Generative Forecasting from Coarse Observations: Does Cascading Help?

Alexandra Spitzer, Giovanni Abati, Lorenzo Baldassari, Till Muser, Cheng Shi, Ivan Dokmanić

原文摘要

We study probabilistic forecasting of dynamical systems such as fluids and waves from coarse observations. Since the unseen fine scales influence the dynamics, the same coarse initial state can evolve into a whole distribution of futures. Generative forecasters may produce visually plausible samples, but it is hard to tell whether they sample the right conditional distribution, since trajectory data contain a single future for each initial state. We address this problem by generating reference data in a hierarchical way: for each coarse initial state, we generate many different fine-scale components and numerically evolve the full states. Using these conditional ensembles, we show that standard diffusion models and stochastic interpolants can make substantial distributional errors when trained over the full forecast horizon with finite data and model capacity, despite being exact in principle. We then show that cascading shorter forecasts greatly reduces these errors for the turbulent flows and waves we study, but we also find and characterize cases in which it hurts. Since the coarse process is not Markov, cascading discards information about the unresolved fine scales even with exact transitions. To understand this, we propose a solvable analytic model which shows that the optimal cascade depth decreases as the coupling between the unresolved and observed scales gets stronger, and increases as the dependence of the forecast on the observation becomes more oscillatory. Consistent with our experiments, cascading is detrimental when the direct forecast is "easy" to learn, for example Gaussian as in the passive scalar case. We release our datasets as a first benchmark for probabilistic forecasting with ground-truth conditional ensembles.

cs.LG

版本更新 · v3arXiv:2611.22639

Hierarchical Control of Antiphase Synchronization in Biomorphic Spiking Neural Architectures for Locomotor Control

Artem Guba, Nikolai Kovalev, Alexey Zharinov, Vyacheslav Demin, Viktor B. Kazantsev

原文摘要

Central pattern generators (CPGs) provide a biologically inspired foundation for adaptive locomotor control, yet systematically characterizing their dynamic regimes in spiking neural networks remains challenging. This study introduces a framework for identifying and stabilizing antiphase synchronization in a biomorphic spiking CPG model embedded in a closed neuromuscular loop. A composite synchronization metric is developed, integrating phase coherence, amplitude balance, and temporal stability. Its performance is evaluated across systematically varied stimulation waveforms and synaptic configurations. Statistical analysis reveals that stimulation type dominates regime determination, with tonic drive yielding the most robust antiphase synchronization. Periodic inputs induce nonlinear saturation and context-dependent network routing. Although synaptic weights alone show limited discriminative power, they significantly modulate robustness within a given stimulation context, with reciprocal Ia interneuron connections emerging as key stabilizers. The proposed composite metric achieves robust antiphase detection, outperforming traditional phase-only measures while preserving biophysical interpretability. These findings establish a statistically validated, hierarchical modeling approach in which external drive defines the operating regime and targeted synaptic configurations fine-tune dynamic stability, providing actionable guidelines for closed-loop neuromechanical controllers and adaptive neuroprosthetic interfaces.

q-bio.NC · cs.AI

新论文 · v1arXiv:2611.25822

Limits of local curvature for constructing search subspaces in segmentation models

Maxim Bolshim, Alexander Kugaevskikh

原文摘要

We examine how well subspaces constructed from local curvature represent the change in the gradient update field after a finite step in semantic segmentation models. The local response is the action of the curvature operator on the actual input displacement. In a study of 50 images and two models, we measure the ratio of the loss increase attained in a fixed subspace to that attained in the full space. This ratio decreases as the search radius increases. On ten images from the same sample, we compare the one-dimensional subspace spanned by the initial gradient with two rank-16 subspaces for both models. The two equal-rank subspaces are assessed at the same initial states, using the same actual steps and measured vectors. Adding generalised Gauss-Newton directions to the initial gradient reduces the mean relative error in representing the field change from 0.922 to 0.874 for the first model and from 0.819 to 0.763 for the second. At the same rank, including the response direction for the initial gradient nearly eliminates the error in representing the local response. The mean error in representing the field change does not decrease. A decomposition with fixed pixel weights shows that the finite-step remainder is comparable in norm to the field change and makes the largest diagonal contribution to the squared projection error of the field after the step. Evaluation of a search subspace should measure how accurately it represents the field change and the local response.

cs.CV · cs.LG

新论文 · v1arXiv:2611.25811

Inject Where It Matters: Objective-Dependent Feature Control in Rectified-Flow Image Editing

Nina Goncharova, Alexandra Ivanova, Aibek Alanov

原文摘要

Many prompt-driven FLUX editing methods rely on injecting source features into target-conditioned generation, creating an inherent trade-off between source preservation and the desired edit. We address this problem by controlling where source features are injected and which transformer blocks are used. First, we introduce adaptive masks that spatially gate feature injection using differences between source- and target-conditioned block outputs at matched inversion states. Second, we use REINFORCE to select Top-K injection blocks under different preservation-editability objectives. The selected blocks are fixed at inference and require no online search. On PIE-Bench with FLUX.1-dev, spatial masking controls the alignment-preservation trade-off. On 60 validation examples, a learned nine-block set gives lower whole-image LPIPS than the 18-block RF-Edit set within the same masked pipeline, with slightly lower alignment. Among 50 random nine-block sets, it has the lowest LPIPS but not the highest reward. Separate objective sweeps reveal different block preferences for preservation and alignment.

cs.AI

版本更新 · v2arXiv:2605.08963

Survey-aware Machine Learning: A Guideline for Valid Population Health Inference based on Scoping Review

YongKyung Oh, Henry W. Zheng, Jeffrey Feng, Alex A. T. Bui

原文摘要

Machine Learning (ML) models trained on complex health surveys such as the National Health and Nutrition Examination Survey (NHANES) often ignore primary sampling units, stratification variables, and sampling weights. This practice violates the independence assumptions of standard evaluation methods. As a result, estimates become biased, uncertainty is underestimated, and fairness assessments fail to reflect population-level disparities. We propose Survey-aware Machine Learning (SaML), a nine-step guideline that incorporates survey design metadata across the ML lifecycle. Through a scoping review of 16 methodological papers, we summarize existing work on weighted model training, design-based cross-validation, and survey-adjusted performance evaluation. We also identify gaps in hyperparameter tuning and deployment. We provide task-specific guidance that clarifies which steps are required for different analytical objectives. SaML provides a checklist for valid population inference from survey data.

stat.ML · cs.LG

版本更新 · v2arXiv:2611.02011

MedPsy: State-of-the-Art Small Medical Language Models for Efficient Edge Deployment

Davide Vitabile, Alexandro Buffa, Akshay Nambiar, Amril Nazir

原文摘要

Medical language models promise expert clinical support, but the strongest open-source LLMs remain too large for private edge-device deployment, while compact medical models still trail substantially on knowledge-intensive and real-world healthcare tasks. Reasoning models that fit edge memory budgets can produce long reasoning traces, increasing sequential decoding, end-to-end latency, energy consumption, and KV-cache memory pressure. We present MedPsy, a family of text-only 1.7B and 4B medical language models designed for edge deployment, targeting this accuracy-generation-efficiency trade-off. Our recipe combines a synthetic medical data pipeline over biology, medicine, and health seeds with chain-of-thought targets from a 235B medical-focused LLM teacher; a four-stage post-training curriculum comprising two SFT stages followed by two RL stages with hard-sample mining; and an edge deployment evaluation spanning multiple GGUF quantization variants and real-device inference on laptop and smartphone hardware. On seven closed-ended medical benchmarks, MedPsy-4B surpasses MedGemma-1.5-4B by 19.34 points and matches MedGemma-27B while being 6.75x smaller; MedPsy-1.7B outperforms MedGemma-1.5-4B by 11.42 points despite being less than half its size. On HealthBench-Hard, MedPsy-4B surpasses MedGemma-27B by 15.33 points, while MedPsy-1.7B surpasses it by 11.66 points despite being ~16x smaller. Beyond accuracy, MedPsy reduces average response length by 1.7x (1.7B) and 3.2x (4B) versus its Qwen3 backbones, while Q4_K_M quantization preserves the aggregate score within 1 point of BF16 with ~69% lower disk footprint. In real-device inference, MedPsy-4B's shorter generations translate into 75% lower median end-to-end latency and estimated energy per answer than its Qwen3 backbone at matched quantization, while peak RAM remains below 5 GB across our tested laptop and smartphone configurations.

cs.CL

新论文 · v1arXiv:2611.25433

RoCoKV: Accelerating LLM Inference Serving through Uncertainty and Redundancy Aware KV Cache Compression

Payman Behnam, Elaheh Sobhani, Saad Amin, Yahya Hassan, Daniyal Khan, Biswadeep Chakraborty, Tanish Patwa, Souvik Kundu 等 9 位作者

原文摘要

KV cache compression is critical for large reasoning models, where most of the cost comes from long generated reasoning traces rather than the input prompt. These traces often contain redundant and low-utility tokens due to looping, self-verification, and corrections. However, such tokens may still receive high attention scores, making it difficult to remove them without harming the reasoning process. We propose RoCoKV, a KV cache compression method designed for large reasoning models. RoCoKV introduces two key components. First, uncertainty-guided adaptive pruning adjusts the compression strength based on the model's confidence, allocating more memory during uncertain phases and applying stronger pruning when the model is confident. Second, redundancy-aware token selection reduces the impact of tokens that are highly similar to recent reasoning content, encouraging the model to retain more informative and diverse tokens. Beyond the algorithm, we co-design RoCoKV with the serving stack: we develop fused three-phase CUDA decode kernels and integrate them into vLLM as a drop-in plugin that preserves its paged KV layout and CUDA graph capture, turning algorithmic savings into real end-to-end speedups. Extensive experiments across three reasoning benchmarks and two reasoning models demonstrate that RoCoKV consistently improves reasoning accuracy under constrained KV cache budgets, achieving higher accuracy compared to state-of-the-art KV cache compression methods. For DeepSeek-R1-Distill-Llama-8B in vLLM, RoCoKV achieves a 1.27X speedup and 38% higher throughput on AIME25, the most challenging evaluated reasoning benchmark, and a 1.43X speedup and 45% higher throughput on the long synthetic benchmark, compared with Full-KV.

cs.CL

版本更新 · v4arXiv:2606.14990

Rational Sparse Autoencoder

Naiyu Yin, Maosen Tang, Alex Townsend, Yue Yu

原文摘要

Sparse autoencoders (SAEs) are standard tools for mechanistic interpretability, but current SAE families are constrained by fixed encoder nonlinearities such as ReLU, JumpReLU, and TopK. This hard-codes a particular sparsity mechanism into the model and can distort the reconstruction-versus-sparsity trade-off. We introduce the Rational Sparse Autoencoder (RSAE), which replaces the fixed encoder activation with a trainable rational function. The design is motivated by a reconstruction advantage that we prove for RSAE at matched expected sparsity, with guaranteed non-inferiority to ReLU and, under stated assumptions, provable improvements over JumpReLU and TopK. The model is then implemented through a two-stage pipeline: an initialization procedure that copies the pre-trained baseline SAE weights and calibrates the scale parameters along with the rational coefficients; followed by a fine-tuning step under a sparsity-regularized reconstruction objective that holds $\ell_0$ at the teacher's value. Consistent with this theory, on residual-stream activations of three open-weight language models and across all three baseline activation families, the RSAE improves reconstruction fidelity after the fine-tuning step at the same level of sparsity. These gains are consistent across host language models, across baseline activation families, and across the full range of baseline sparsity we tested, while the upgrade itself adds only 20 scalar parameters per autoencoder and one training run takes between about 4 and 25 minutes on a single consumer GPU.

cs.LG · cs.AI

新论文 · v1arXiv:2611.25288

Intelligent Embedding Pruning for Large-Scale Recommendation Models

Anish Khazane, Alex Saad-Falcon, Guanchu Wu, Marcio Porto, Jasper Shan, Fei Yu, Eugene Shulga, Intaik Park 等 12 位作者

原文摘要

Modern recommendation systems face substantial memory pressure, largely from the sparse embedding layer that maps high-cardinality categorical features into dense vectors. In-training embedding pruning (ITEP) removes less important embedding rows during training, yielding end-to-end memory savings but sometimes degrading accuracy because per-table pruning ratios are set by size-based heuristics. We use feature importance (FI) to guide these ratios (FI-ITEP), assigning higher pruning to tables tied to less important features. Across three large-scale recommendation models, FI-ITEP achieves 20--40% total memory reduction while matching or improving accuracy relative to percentile-based ITEP. Because the feature-importance signal requires minimal resources to compute, these memory savings translate into concrete systems benefits: the same models train on fewer GPU trainers (e.g., 112 to 80) with higher per-GPU throughput, saving 1,000+ GPU-hours per model launch. We reproduce these trends on the open-source DLRM (Criteo 1TB) and quantify the end-to-end training-time and GPU-cost tradeoff---including the burden of computing FI---offering practical guidance for embedding pruning in real-world systems.

cs.LG · cs.AI

版本更新 · v2arXiv:2608.00799

CADENA:逐步式CAD逆向工程

CADENA: Stepwise CAD Reverse Engineering

Soslan Kabisov, Gennadiy Savrasov, Maksim Elistratov, Antonio Rodriguez, Daniil Ignatiev, Nikita Gavrilov, Rustam Uzdenov, Alexey I. Boyko 等 12 位作者

AI 速读 基于摘要

CADENA 以逐步生成并每步比对当前几何与目标的方式,将3D网格重建为参数化CAD程序,并引入按机械零件类别评测的 CADENA-Bench。

cs.CV

查看原文摘要

Computer-Aided Design (CAD) underpins modern engineering, yet converting existing shapes into editable models still demands substantial expert effort. Most AI systems emit the entire CAD program in a single pass, never inspecting the intermediate geometry. In contrast, human engineers build a part feature by feature, checking after each operation what remains to be modeled. We introduce CADENA (Spanish for "chain"), a model that reconstructs a 3D mesh as a parametric CAD program, growing its sequence of operations one at a time and comparing the target with the currently predicted geometry at every step. We also address the lack of benchmarks for evaluating reverse-engineering methods on mechanical parts, introducing CADENA-Bench, a benchmark that measures performance across categories of mechanical parts. CADENA outperforms prior methods on CADENA-Bench and on the DeepCAD, Fusion 360, and MCB datasets. Code is available at https://github.com/zhemdi/cadena, model weights at https://huggingface.co/kulibinai/cadena, and CADENA-Bench at https://huggingface.co/datasets/kulibinai/cadena-bench.

版本更新 · v2arXiv:2610.14226

CAPTOR:用于胸腹盆腔肿瘤CT的视觉-语言基础模型

CAPTOR: a vision-language foundation model for Chest-Abdomen-Pelvis Oncologic CT

Séléna Jean-Mactoux, Sophie Loizillon, Ghina Jardali, Amandine Crombé, Nathalie Lassau, Alexandre Bône

AI 速读 基于摘要

作者报告CAPTOR是一个基于解剖结构化预处理的视觉-语言基础模型,在肿瘤胸腹盆腔CT的分类、检索和生存预测任务上优于五个CT基础模型。

cs.CV · cs.AI

查看原文摘要

Objectives: Radiologists interpreting oncologic chest-abdomen-pelvis (CAP) CT must also detect non-oncologic conditions, yet existing CT foundation models are trained on single anatomical regions or on non-oncologic data. We developed and evaluated CAPTOR, a vision-language foundation model for oncologic CAP CT that uses an anatomically structured preprocessing strategy to learn from whole examinations. Materials & Methods: CAPTOR is a contrastive vision-language model pretrained on 25,512 predominantly oncologic CAP CT-report pairs (single or multiple series) from a retrospective multicenter dataset (North America, Europe, Asia). It was compared with five CT foundation models, a larger-scale variant (CAPTOR-100K) and an ablation without anatomical preprocessing (CAPTOR-NoPre) on classification and retrieval using held-out internal and public test sets, and on survival prediction using TumorImagingBench. Confidence intervals and pairwise comparisons used patient-level bootstrap and permutation tests. Results: On 10,794 test studies, CAPTOR achieved higher AUC than all five models on CAP (0.767 vs. 0.695 for the next-best model; p<0.001) and oncologic subsets (0.784 vs. 0.734; p<0.001), with superior retrieval on both, and ranked first on three of four oncologic survival benchmarks. Despite substantially fewer training examinations, it performed comparably to CAPTOR-100K. Removing the anatomically structured preprocessing reduced classification performance, including on the institutionally held-out test subset (macro-AUC 0.737 vs. 0.691 for CAPTOR-NoPre, p<0.01). Conclusion: Anatomically structured preprocessing enabled whole-examination CAP CT interpretation, including multiple-series acquisitions, and improved downstream performance. CAPTOR outperformed state-of-the-art CT foundation models on CAP and oncologic subsets while matching a substantially larger-scale variant.

新论文 · v1arXiv:2611.24711

Openhagen Worlds: Calibrated Environments for Measuring and Training Scientific Discovery Agents

Alexander Boesgaard Lorup

原文摘要

Agents that do science must choose informative experiments and reason from their results, but most discovery benchmarks cannot separate that skill from recall, unstated scoring rules or shortcut policies. We introduce Openhagen Worlds, a suite of 27 procedurally generated environments (some with finite truth spaces), from classic puzzles of experimental reasoning to simulator-grounded physics and chemistry, in which an agent runs experiments under a budget and commits an answer that a deterministic verifier checks against a hidden truth. Every observation states the complete scoring rule, and each of the 178 difficulty levels is calibrated: it is admitted only if, on 64 development instances, a scripted oracle succeeds on at least 0.75, the strongest plausible non-deductive policy on at most 0.25 and each gated shortcut on at most 0.35. The 92 core levels span all 27 families and are sized to train small models; the 86 frontier rungs extend the 15 discovery worlds with harder questions and with campaigns of three, five and eight or more committed stages on one hidden system. On the same 84 withheld frontier instances, one per family and rung, DeepSeek V4.1 Flash solves 32 at its default reasoning effort with 16,384 tokens per call and 57 at its maximum effort with up to 393,216 tokens per call (oracle: 79); the headroom that remains is largest in eight-stage campaigns. On-policy expert iteration gives a 9B model modest gains: its core-tier success on trained families rises from 0.090 to 0.148, partly through better action format, with no detectable loss on GPQA Diamond. A hosted service plays the core tier's withheld instances.

cs.LG · cs.AI · cs.CL

新论文 · v1arXiv:2611.24641

Choosing the Right Representations: Task-Aware Fusion for Multimodal Materials Language Models

Huu Binh Ta, Alex Morehead, N. Benjamin Erichson, Ferdinando Fioretto, Sandeep Madireddy

原文摘要

Multimodal language models for materials science rely on effective mechanisms for incorporating three-dimensional structural representations. MatterChat addresses this challenge using a large bridge module between a pretrained structural encoder and the LLM, together with a separate contrastive alignment stage. We instead ask whether this bridge can adaptively select and combine representations from multiple structural encoders according to the target property. We find that supervised and generative material encoders provide complementary information for property prediction. We found that, for Zatom-1, this information is not concentrated in the final layer. Instead, the most informative representations occur at different depths depending on the target property. This motivates Task-Conditioned Attention Fusion (TCAF), a lightweight attention-based bridge that uses task-conditioned latent queries to combine CHGNet features with intermediate Zatom-1 representations. The bridge is trained directly on downstream prediction tasks, allowing representation selection and material-language alignment to be learned jointly without separate contrastive pretraining. Across classification and regression tasks, TCAF outperforms MatterChat, with up to 47.8% lower RMSE, while using substantially fewer trainable parameters and halving the overall training time. Further experiments incorporating UMA-S demonstrate TCAF's flexibility across encoder combinations, with the three-encoder configuration achieving the highest macro-average classification accuracy of 89.61%.

cs.LG · cs.AI

版本更新 · v3arXiv:2611.01335

Profile-aware assessment of block-diagonal Hessian approximations in neural networks

Maxim Bolshim, Alexander Kugaevskikh

原文摘要

Evaluating block-diagonal curvature models in neural networks involves two questions: how much curvature they omit, and what a specific correction changes at a finite step. We measure the squared off-block fraction of the parameter Hessian against references for group sizes and the observed within-group curvature profile. In CIFAR-100 ResNet-18 and a skip-free counterpart, similar initial discarded fractions accompany different profiles. At saved trained ResNet-18 weights, changing from stage to layer groups moves estimated off-block mass from below to above the profile reference. An approximate assessment of probe uncertainty conditional on fixed weights and inputs resolves this crossing; some skip-free comparisons remain unresolved. Separately, in nine saved ResNet-50/Stanford Cars states from three training seeds, we compare candidates formed from off-group Hessian action, the remaining model residual and their combination. For each state, we construct two full-Hessian-informed maps on separate geometry draws. Within each draw, all three inputs share one map. At unit scale off-group input alone raises evaluated minibatch loss relative to the base preconditioner candidate in 17 of 18 state/draw comparisons, yet the combined candidate has lower loss than the remainder-only candidate in nine of those cases; preferences vary with scale and correction-norm matching. The profiles help assess block models and alternative groupings, while same-state candidate comparisons test a particular correction without equating structural mass with training gain.

cs.LG

版本更新 · v2arXiv:2611.21365

Exploring the link between predictive uncertainty and the sim-to-real gap in deep learning models utilizing synthetic data

Alexander Moriz, Lukas Krebs, Amon Göppert, Robert H. Schmitt

原文摘要

Today, state-of-the-art methods for quality control (QC) are based on Machine Vision, combining deep learning (DL) models with camera-based inspection. Yet, the integration of such solutions in industrial applications is often impeded by data sparsity and trust in model reliability. While synthetic data offers a promising approach to generate the necessary data with perfect annotations, it poses risks related to performance degradation due to the possible occurrence of the sim-to-real gap. This study conducts experiments using the ensemble method to estimate the predictive uncertainty of DL models trained on synthetic data. The results show that predictive uncertainty increases under the sim-to-real domain shift. However, the observed performance drop is dominated by a systematic sim-to-real bias rather than by ensemble disagreement.These findings contribute to the effective utilization of synthetic data in industrial settings while highlighting the importance of risk assessment for the deployment of DL solutions for QC tasks.

cs.CV

版本更新 · v2arXiv:2607.08514

Do Egocentric Video-Language Models Capture Both Hand- and Object-Centric Cues?

Masatoshi Tateno, Alexandros Stergiou, Risa Shinoda, Yoichi Sato, Dima Damen

原文摘要

Recognizing hand-object interactions (HOIs) requires capturing both hand dynamics and object transformations, which provide complementary information for distinguishing visually similar manipulations. However, Video-Language Models (VLMs) trained solely on egocentric videos tend to over-rely on hand appearance, struggling to distinguish interactions with similar visual appearances but different dynamics (e.g., taking vs. kneading dough). To better utilize complementary hand- and object-centric cues, we propose a learning paradigm with (i) hand-object masked training to encode both hand- and object-centric cues and learn their relationships through cross-cue reconstruction, and (ii) an HOI-dynamics-aware decoder that grounds each cue through auxiliary location and semantic predictions. To assess how effectively models leverage hand- and object-centric cues for action recognition, we introduce Cue-Isolated HOI (CI-HOI), which independently evaluates recognition performance based on each cue. To enable this evaluation, we curate the DEHOI testbed using video inpainting to selectively remove either hands or manipulated objects, allowing us to evaluate action recognition with each cue isolated. DEHOI reveals that existing VLMs trained only on egocentric video rely more on hand-centric cues than VLMs trained on more non-egocentric video data. We show that our training paradigm mitigates this hand-centric bias, outperforming VLMs trained on the same egocentric data across DEHOI, standard action recognition, and object state recognition benchmarks.

cs.CV

新论文 · v1arXiv:2611.24375

SearchDiffusion: Information-Guided Diffusion Planning for Object Search

Sausar Karaf, Dmitrii Plotnikov, Alexey Kotcov, Dzmitry Tsetserukou

原文摘要

Object search in unknown environments requires robots to gather useful observations while avoiding obstacles and accounting for limited sensor visibility. We present SearchDiffusion, a diffusion-based planner that generates position and heading trajectories using a learned dense utility map. The map represents geometric accessibility, useful visibility, and semantic progress. It conditions trajectory generation and provides an objective for sampling guidance. Sampled map hypotheses provide additional geometric context, while geometric and heading guidance, validity checks, and analytical selection support execution. Across 90 simulated tasks on 24 map layouts, SearchDiffusion achieved 95.6\% target detection at greater computational cost. Ablations show complementary benefits from utility and guidance, with limited additional detection benefit from runtime map hypotheses. In ten physical quadruped trials using OneMap relevance maps, eight were successful under operator assessment, without fine-tuning on real-robot data. These results support combining learned spatial utility with guided trajectory generation for object search under partial observability.

cs.RO

新论文 · v1arXiv:2611.24321

Benchmarking LLM abilities on postdoc-level natural science problems

Alina Avanesyan, Alexey Boldyrev, Liqun Deng, Denis Derkach, Mary Godunova, Mikhail Hushchyn, Dmitry Ilvovsky, Irina Knyazeva 等 13 位作者

原文摘要

Scientific problems have no answer key, so evaluating an agent means deciding what counts as a better solution before any agent is run. Existing benchmarks either match a symbolic answer against a key or, where they require executable work, score it by unit tests, by a rubric, or by position on a competition leaderboard. None of these is the criterion by which a field decides that one published method beats another. STEMBench adopts that criterion directly. Each of its twelve research problems in high-energy physics, astrophysics and neuroscience is scored by its own field's figure of merit, and each comes with two measured anchors: a submission that uses no domain knowledge, and an expert reference reproducing a published method, both obtained by running code shipped with the benchmark. The interval between them states how much of a problem is crossed by scientific insight rather than by engineering. Across eight frontier models, task-level wins and domain-level leadership dissociate: a model that wins five of the twelve tasks leads no domain mean. In the high-effort cross-model comparison, at least one model's mean score exceeds the expert anchor on seven of twelve tasks. The remaining five span uncertainty estimation, latent and causal inference, and cross-site clinical classification. Numerical scores and official-submission validity capture different aspects of performance and are reported separately. We release the tasks, graders, anchors and harness.

cs.AI · physics.data-an

新论文 · v1arXiv:2611.24283

X-split Argus: Learning Reusable History Tokens for Efficient CTR Prediction

Alexander Ploshkin, Nikita Stepanov, Alexey Pismenny, Viacheslav Ovchinnikov, Nikolay Savushkin

原文摘要

Accurate click-through rate (CTR) prediction depends on user representations that capture both persistent preferences and recent intent, yet processing extensive behavioral histories for every ranking request is costly under production compute and latency constraints. We present X-split ARGUS, a framework for ad ranking whose history compressor and online ranker are trained end to end. The compressor encodes past interactions into a compact set of tokens that are computed offline and reused across candidates, while the online ranker scores ads through cross-attention to these tokens and representations of recent events. Training combines an autoregressive loss for click feedback with distillation of candidate-specific click logits from a larger teacher that scores each candidate using the uncompressed history. The number of compressed history tokens and the length of recent history can be adjusted to fit serving compute and latency budgets. Offline evaluation shows a 0.81% relative reduction in normalized entropy (NE) compared with the production baseline. Deployment in a large-scale real-time bidding system increases online inference cost by 15% relative to the production baseline. An online A/B test against this baseline yields relative gains of 4.65% in a business metric reflecting the monetary value of conversions generated for advertisers and 2.96% in clicks.

cs.IR

版本更新 · v2arXiv:2610.35385

EmbodiedLab: Beyond Manipulation, Benchmarking Embodied Agents That Act to Know

Hongcheng Gao, Zelin Zheng, Alex Zhou, Jay Zhu, Shijia Ge, Yazhe Wang, Xiping Li, Jianshu Zeng 等 21 位作者

原文摘要

Physical action not only changes the state of an environment but is also the way to know its hidden properties: what cannot be seen must be inferred from the feedback to the change. As embodied agents advance from executing procedures to pursuing goals, success depends on properties neither given nor visible, so they must act to know. Existing benchmarks separate changing from knowing: manipulation benchmarks give the target state, so nothing has to be found out, whereas exploration and discovery benchmarks let the agent find out by looking or querying, so nothing physical has to be changed. Quantitative science experiments offer a natural testbed: the properties sought are new in every instance, appear only in instrument readings under conditions the agent sets up, and can be checked exactly. We introduce EmbodiedLab, a simulated laboratory of 35 such experiments. Given only a goal, the agent must decide what to measure, operate the apparatus with a robot arm, read instruments through cameras and judge when its readings suffice. Its 30 determination tasks specify what to find out, whereas its 5 optimization tasks ask only for the best result the apparatus allows. With native control, the strongest of seven frontier agents succeeds on 41.4% of the tasks. Ablations reveal two bottlenecks: contact-rich manipulation, where ground-truth depth and motion planning change the success rate by at most 10 points, whereas executing every operation for the agent on five of the tasks raises it from at most 20% to at least 70%, so what an agent can know is bounded by how reliably it acts; and closed-loop experimental rigor, where with manipulation supplied, agents fail to gather sufficient readings or read instruments accurately, and most optimization runs stop at one design. Beyond evaluation, EmbodiedLab offers a resamplable environment with verifiable rewards for training embodied agents.

cs.RO

版本更新 · v2arXiv:2510.05364

Challenging Quadratic Attention$\unicode{x2013}$A Holistic View on the Rise of Alternative Language Model Architectures

Alexander M. Fichtl, Jeremias Bohn, Josefin Kelber, Edoardo Mosca, Georg Groh

原文摘要

Transformers have dominated sequence processing tasks for the past seven years$\unicode{x2013}$most notably language modeling. However, the inherent quadratic complexity of their attention mechanism remains a significant bottleneck as context length increases. We review and distill the recent efforts to overcome this bottleneck, including advances in (sub-quadratic) attention variants, recurrent neural networks, state space models, and hybrid architectures. We critically analyze approaches regarding compute and memory complexity, benchmark results, and fundamental limitations to assess whether the dominance of pure-attention transformers may soon be challenged, which we consider possible, particularly in domain-specific and edge-device applications.

cs.CL

新论文 · v1arXiv:2611.23996

ReSynth: Constructing Synthetic Supervision for Multi-Image Understanding

Yilin Wang, Xin Yu, Xinyi Guo, Xiangxi Zheng, Dongxing Mao, Linjie Li, Zhengyuan Yang, Alex Jinpeng Wang

原文摘要

People often understand the world by comparing different views of it. Vision-language models (VLMs), however, receive these views as separate images and still struggle to relate them. One problem is data. Training on these tasks requires groups of images with a known relationship: the same object from different viewpoints, or the same scene before and after a change. Such examples are hard to collect at scale from existing photo collections. We introduce ReSynth, which turns single-image perception and controllable image generation into cross-image relational supervision. Localized editing produces before-and-after pairs, viewpoint generation produces multiple views of the same instance, and verified scene contents support comparisons of counts and attributes. ReSynth combines Qwen3-VL-8B's captioning, grounding, and counting capabilities with Qwen-Image-2.1's generation and editing capabilities. It verifies single-image facts such as object counts and attributes, then combines these facts to construct multi-image QA. Using GRPO on approximately 440K ReSynth questions, with no real images during training, Qwen2.5-VL-7B and Qwen3-VL-8B improve by 17.73 and 6.10 average points on BLINK, MUIRBench, and VLM2-Bench, while single-image performance is broadly preserved on Qwen2.5-VL-7B and improves across all four evaluated benchmarks on Qwen3-VL-8B. Notably, this supervision enables the 8B perception model itself to outperform Qwen3-VL-235B on BLINK and MUIRBench.

cs.CV

版本更新 · v3arXiv:2602.16169

Unlocking Continuous Diffusion Language Models via Discrete Stochastic Localization

Yunshu Wu, Jiayi Cheng, Longxuan Yu, Partha Thakuria, Rob Brekelmans, Evangelos E. Papalexakis, Greg Ver Steeg

原文摘要

Continuous diffusion is a natural framework for non-autoregressive generation but has generally lagged behind masked discrete diffusion models (MDMs) on discrete sequence generation. We argue that continuous diffusion can effectively model discrete sequences through the joint design of noise parameterization and token representation geometry. We introduce \emph{Discrete Stochastic Localization} (DSL), a continuous-state framework that combines a stochastic localization channel with unit-sphere token embeddings, resulting in a Bayes-optimal denoiser that depends only on the noisy state, without additional timestep conditioning. A single trained denoising network then supports multiple sampling methods, from continuous denoising to masked diffusion, random-order autoregressive generation and hybrid continuous-then-discrete generation. Parallel generation in this continuous-state framework, therefore, need not imply synchronized denoising: some tokens can establish context earlier while others remain open to refinement. Fine-tuning a pretrained MDLM checkpoint with DSL substantially improves generation quality, as measured by MAUVE, on OpenWebText across all step budgets from $T{=}128$ to $T{=}1024$. The same checkpoint supports random-order autoregressive sampling, and a hybrid continuous-then-discrete sampler using as few as $T=48$ total steps, without distillation or further training.

cs.LG · cs.CL

新论文 · v1arXiv:2611.23954

SimplePro: A Simple Baseline for Proactive Video Understanding

Yilin Wang, Xiangxi Zheng, Dongxing Mao, Linjie Li, Zhengyuan Yang, Alex Jinpeng Wang

原文摘要

Proactive video understanding requires vision-language models (VLMs) to decide not only what to answer as a video unfolds, but also when newly arrived evidence warrants a response and when to remain silent. Although modern VLMs excel at offline long-video understanding, they struggle in proactive streaming settings, leading prior work to introduce task-specific reinforcement learning or auxiliary triggering modules. We find that explicit reporting rules and structured streaming context substantially improve a frozen VLM's proactive response performance. We exploit this observation to build SIMPLEPRO, a training-free baseline for proactive video understanding. At each decision step, the model combines recent high-resolution frames, bounded historical observations, and prior answer records to determine whether new information warrants a response and, when needed, generate a concise answer. The system then filters consecutive duplicate answers while retaining the raw outputs as context for subsequent decisions. Without any parameter updates or auxiliary triggers, SIMPLEPRO with a 4B backbone consistently outperforms specialized proactive systems across OVO-Bench, StreamingBench-PO, and ProactiveVideoQA. SIMPLEPRO shows overall performance gains as backbone size increases from 2B to 32B. We hope this simple yet effective method will serve as a strong baseline for proactive video understanding, and we will release our code and inference pipeline to support future research.

cs.CV