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

+1111新增论文
+874其中 AI 相关
+133已有论文修订
+97新增 AI 导读

arXiv 最近完成扫描:2026-11-20 06:44 · 全站最近入库:2026-11-19 19:09各信源进度 →

统计口径与数据新鲜度

统计 2026-11-19 07:12 — 2026-11-20 07:12(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 06:44最近发现新增版本:2026-11-19 19:09库内论文最新更新时间:2026-11-19 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。

27,982 已收录论文23,249 AI 相关8,559 有修订112 HF 社区发现

浏览论文库

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

找到 23,249 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-20 06:42(北京时间)
版本更新 · v2arXiv:2608.15341

TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models

Alireza Dehghanpour Farashah, Zhuan Shi, Negar Rostamzadeh, Golnoosh Farnadi

原文摘要

Text-to-image diffusion models can be misused to generate harmful content through adversarial or paraphrased prompts that bypass built-in safety mechanisms. Existing concept erasure methods often suffer from limited robustness against adversarial prompts, degradation of benign generation quality, or reliance on inference-time interventions that introduce persistent computational overhead. To address these limitations, we formulate concept erasure as a domain alignment problem in the text representation space. We propose a lightweight Text Encoder Alignment framework (TEA) that fine-tunes only the text encoder while keeping the generative backbone fully frozen. Given concept-anchor prompt pairs, our method trains a discriminator to distinguish token-level representations of concept-bearing prompts from those of safe anchor prompts, while updating the text encoder to make these representations indistinguishable. TEA introduces zero inference-time overhead and trains in seconds with only a few fine-tuning steps, making it practical for large-scale deployment. Despite this efficiency, TEA achieves state-of-the-art erasure robustness against black-box and white-box adversarial attacks on Stable Diffusion v1.4, while preserving generation quality on benign prompts. Furthermore, TEA is model-agnostic and achieves the lowest attack success rate on Stable Diffusion v3.5, extending concept erasure to Rectified Flow Transformer architectures with T5 conditioning, which remain largely unexplored by prior methods. Code is available at https://github.com/alirezafarashah/TEA

cs.CV

新论文 · v1arXiv:2611.23857

Neural Memory Orchestration

Michael Zizov, Boris Morose

原文摘要

Modern HPC workloads increasingly stall on the memory hierarchy rather than on compute, and page faults from mmap'd files, swap, and tiered memory are a growing fraction of wall time. Heuristic prefetchers in the kernel handle stride-like and temporally clustered patterns well but degrade sharply on the irregular, data-dependent access streams typical of these workloads. Learned prefetchers have been shown to capture such patterns in simulation but have not been taken to a deployed pipeline running against live workloads on real hardware. We present Overseer, an end-to-end learned page-prefetching system. An eBPF collector observes page faults on the target host, a transformer predictor (OverNet) runs on network-attached inference hardware and emits future page addresses, and a kernel-resident enforcer materializes those pages into the target's address space before they are demanded. OverNet encodes fault sequences through per-instance ordinal ranks over deltas, which lets a single trained model serve any target process without retraining. On two page-fault-heavy workloads, Overseer reduces runtime by roughly 35% on Hilbert-curve traversal and by roughly an order of magnitude on dense matrix multiplication, with corresponding reductions in major-fault counts.

cs.DC · cs.LG · cs.OS

新论文 · v1arXiv:2611.23856

Beyond Cross-Modal Reconstruction: Participant-Level Evaluation of sEMG-Kinematic Coupling in Post-Stroke Gait

Rafsan Jany, Md Rafid Islam, Mohammad Fahim Islam, Hafiz Abdur Rahman

原文摘要

Accurate cross-modal reconstruction is often treated as evidence of physiological coupling, although it may simply reflect temporal patterns shared across participants. We tested whether frozen bidirectional models captured subject-specific neuromuscular coupling between electromyographic activity and gait kinematics in 82 held-out able-bodied and 43 post-stroke participants. Complete source-modality removal and cross-participant source substitution, along with a cycle-position-age-sex template baseline, were applied. Learned bilateral asymmetry was also compared with transparent waveform and derivative asymmetry for predicting Performance-Oriented Mobility Assessment (POMA) scores in the case of repeated nested cross-validation. Complete source removal increased reconstruction RMSE by 6.90% for electromyography-to-kinematics and 11.38% in the reverse direction. In contrast, cross-participant substitution increased RMSE by only 1.68% and 1.67%, respectively, while the template reduced RMSE by 2.89% and 0.55%. For POMA prediction, walking speed, age, and sex achieved an MAE of 3.693, compared with 4.229 and 4.427 after adding absolute and signed learned asymmetry, respectively. In all of the 20 repeated evaluations, all asymmetry representations performed worse than the covariate baseline. No localized learned feature remained significant after covariate adjustment and false-discovery-rate correction. The model mainly relied on shared gait-phase patterns, with no clear subject-specific coupling or added clinical value. Participant-level controls are therefore needed before reconstruction accuracy is treated as evidence of individualized physiology.

cs.LG

新论文 · v1arXiv:2611.23855

RoboBuilder: Agentic Robot Co-Design with Real-world Video Feedback in the Loop

Ghadi Nehme, Dahyun Nam, Akshay Govind Srinivasan, Faez Ahmed

原文摘要

Can AI agents design robots that work in the real world? We introduce RoboBuilder, a benchmark in which coding agents co-design a robot's hardware and program from standard parts, motors and a custom 3D-printed part, and every design that passes in simulation is built and tested on a physical course. Simulation alone is not enough: under public part data, none of 15 initial passing designs crossed the real course, so an agent learns the simulator from measurements it requests and videos of real runs. Across four frontier coding agents, ten obstacles at three difficulty levels, 16,427 simulations of 12,774 distinct robots and 135 physical robots built in an estimated 240 hours, agents solve nearly all easy obstacles and invent mechanisms our search baselines do not reach, but falter at harder levels, when one robot must cross every obstacle, and in the real world, where 66% of simulated passes hold across all levels. Closing the loop helps: shown a video of its real failure, an agent redesigns the robot, sometimes correcting its simulator, and removes 72% of level-1 and level-2 failures. RoboBuilder measures progress toward agents that engineer real machines. Project Page: https://ghadinehme.com/robobuilder.github.io/

cs.AI · cs.CV · cs.RO

版本更新 · v5arXiv:2410.23066

Don't Pay Attention, PLANT It: Pretraining Attention via Learning-to-Rank

Debjyoti Saha Roy, Byron C. Wallace, Javed A. Aslam

原文摘要

State-of-the-art Extreme Multi-Label Text Classification models rely on multi-label attention to focus on key tokens in input text, but learning good attention weights is challenging. We introduce PLANT (Pretrained and Leveraged AtteNTion), a plug-and-play strategy for initializing the label-attention module that operates on top of an LLM's output representations. PLANT works by planting label-specific attention using a Learning-to-Rank model trained on mutual information gain, before end-to-end fine-tuning. This architecture-agnostic approach integrates with LLM backbones (e.g., Mistral-7B, Llama-3.1-8B, DeepSeek-V3, and Phi-3-mini). PLANT outperforms SOTA methods across tasks like ICD coding, legal topic classification, and content recommendation, and improves fine-tuned LLMs in controlled comparisons outside the medical domain. Gains are largest on rare labels, although a training-free LLM extraction baseline remains competitive on rarely seen labels. Ablation studies confirm that attention initialization is a key driver of these gains. We make our code available at https://github.com/deb-official/plant

cs.CL · cs.LG

版本更新 · v5arXiv:2410.22725

VerifyPrompt: Towards Text-to-Image Models Verification Behind Black-Box APIs

Ji Guo, Rui Zhang, Guomin Lu, Aiguo Chen, Tianwei Zhang, Hongwei Li, Wenbo Jiang

原文摘要

With the rise of diverse Text-to-Image (T2I) models such as DALL-E 3, third-party platforms increasingly integrate multiple black-box APIs to deliver cost-effective services. This also raises a new and practical security question: are these third-party platforms truly offering the models they claim? Such model substitution fraud, in which a malicious third-party platform operator acts as the adversary, enables the platform to profit illicitly while undermining user interests and damaging the commercial reputation of legitimate model providers. To address this practical concern, we first define the task of T2I model verification, which determines whether a black-box target model matches a black-box oracle reference model. After that, we propose VerifyPrompt, which performs T2I model verification through a specially designed verify prompt. Intuitively, the verify prompt is an adversarial prompt for the target model, without transferability to other models. It makes the target model generate a specific image while making other models produce entirely different images. Our key idea for generating verify prompts is to enforce strong interactions between adversarial perturbations and the target model, thereby overfitting the target and producing model-specific perturbations. To enable efficient black-box search for verify prompts, we further introduce VerifyPrompt-RL, a reinforcement learning-based strategy that leverages multi-sample exploration and group relative error estimation to stabilize and accelerate the search process. Experiments show that VerifyPrompt achieves over 90\% accuracy across various T2I platforms, confirming its effectiveness in real-world third-party services.

cs.CV · cs.CR

新论文 · v1arXiv:2611.23848

A Fast, Programmable GPU-to-CPU Invocation Layer for GPU-Initiated Communication

Jiakun Yan, Shiksha Patel, David Sidler, Gengyu Rao, Brandon Potter, Bradford Beckmann

原文摘要

Modern GPU workloads increasingly rely on device-initiated communication to let GPU kernels communicate directly without synchronizing with the host at kernel boundaries. Existing libraries commonly favor placing transport control on GPU threads, an approach known as GPUDirect Async (GDA), and treat reverse offload to host proxy threads as a slower fallback. We argue that reverse offload can outperform GDA and should be a first-class communication path. We further propose exposing reverse offload as a programming interface that lets users define workload-specific communication operations. We present KIWI, a programmable reverse-offload framework that achieves sub-microsecond one-way device-to-host latency and hundreds of millions of operations per second. We address practical challenges of scaling reverse offload, including contention on shared queues, polling overhead, and load imbalance across proxies. KIWI-P2P, a SHMEM-like library built on KIWI, outperforms state-of-the-art GDA implementations in both latency and bandwidth. More importantly, KIWI-EP redesigns low-latency mixture-of-experts dispatch and combine kernels around workload-defined operations, lowering expert-parallel communication latency by up to 2.3x over the best existing backend on AMD GPUs and time per output token in vLLM by up to 29%. The source code of KIWI is available at https://github.com/AMDResearch/kiwi.

cs.DC · cs.LG

新论文 · v1arXiv:2611.23845

ReflectScene:MLLM-Guided Closed-Loop Diffusion for 3D Indoor Scene Generation

Jiawei Shen, Jie Feng, Junpeng Zhang, Guanbin Li, Wanyi Chen, Weisheng Dong, Guangming Shi

原文摘要

Text-conditioned 3D indoor scene generation requires layouts satisfying geometric, functional, and semantic constraints. However, existing diffusion methods rely primarily on static textual conditions, limiting their ability to assess evolving scenes and translate semantic discrepancies into executable object-level corrections. MLLM-based post-hoc refinement provides semantic feedback but remains decoupled from denoising, potentially disrupting scene coherence.We propose ReflectScene, an MLLM-guided closed-loop diffusion framework. First, we introduce Hierarchical Prior-Grounded Scene Understanding that aligns structured layouts with multi-view visual evidence, to interpret object-specific context and scene global semantics via a frozen multi-modal large language model (MLLM). Second, we propose MLLM-Guided Semantic Reward Evaluation to distill MLLM's diagnoses and assessments on possible candidate corrections into a lightweight Reward Predictor, enabling correction decision making with reduced computation burden. Finally, Decision-Conditioned Trajectory-Consistent Diffusion is proposed to integrate selected corrections through transient conditioning and persistent clean-estimate projection, correcting object layout in the generated scene while preserving non-target arrangements throughout denoising.Experiments on 3D-FRONT demonstrate improved functional consistency and semantic alignment while maintaining competitive generation quality and diversity.

cs.CV · cs.AI

新论文 · v1arXiv:2611.23843

RePlan-VLA: Progress-Aware Replanning for Long-Horizon Vision-Language-Action Models

Apoorva Vashisth, Manav Kulshrestha, Damon Conover, Zachary Kingston, Aniket Bera

原文摘要

Vision-Language-Action (VLA) models have emerged as a promising paradigm for general-purpose robotic manipulation, yet their performance often degrades on long-horizon tasks that require multiple interdependent manipulations. Instead of burdening the VLA policies with long-horizon reasoning, memory, action generation, progress tracking, and error recovery, we argue for leveraging their short-horizon manipulation capabilities within an execution-aware framework. To this end, we introduce \textnormal{\textbf{RePlan-VLA}}, a VLA-agnostic framework that translates a VLA's short-horizon capabilities into reliable long-horizon execution. RePlan-VLA explicitly monitors the robot's motion for progress evaluation, and adaptively revises the future plan based on observed outcomes. We preserve the underlying VLA policy entirely, requiring no additional fine-tuning. Experiments on RoboCasa365 and CALVIN benchmarks show an improvement of 21.9% and 10.4% in success rate over the next strongest baseline. We also validate our approach for real-world deployment on a Franka Research 3 robot.

cs.RO · cs.AI · cs.LG

新论文 · v1arXiv:2611.23840

TOGBench: Benchmarking and Improving Tiny-Object Generation Fidelity in Aerial Imagery

Chanhwi Jeong, Dongmin You, Seungwook Kim

原文摘要

Advances in deep generative models now enable the synthesis of realistic imagery. However, aerial imagery is dominated by tiny objects, such as vehicles, ships, and aircraft, which occupy only a few pixels owing to the overhead, wide-area nature of aerial imaging. Existing generative models struggle substantially more with tiny objects than with larger ones. This limitation has been overlooked, as prior work on generative evaluation emphasizes prompt following or human preference rather than whether small objects are rendered faithfully. To address this gap, we introduce TOGBench, a quantitative, object-level benchmark for evaluating tiny-object generation fidelity in aerial imagery. We further propose two simple train-/inference-time techniques that improve the ability of layout-to-image generators to render small objects, raising their performance on TOGBench.

cs.CV · cs.AI

新论文 · v1arXiv:2611.23836

Online State-Representation Adaptation under Synthetic Interaction-Cue Remapping

Bin Yin

原文摘要

When cues change what they predict, continuing to update a learned mapping may help an agent choose the appropriate action. We tested this in a one-step synthetic task by comparing a logistic state predictor that continued to learn from state labels with the same predictor frozen at transfer. We evaluated ten conditions in six held-out environments, with 1,296 paired runs per condition and 1,296,000 transfer records in total. Late action accuracy was 93.39% with online updates and 64.63% with frozen coefficients, a paired difference of 28.76 [8.73, 53.19] percentage points (95% environment-bootstrap interval). A direct policy reached 95.65% using the same predictions with a different action threshold. A contextual bandit reached 94.23% and earned more cumulative utility than the adaptive predictor. The bandit used the reward of its chosen action, which in this task reveals the true state when combined with that action. Weights based on development history lowered cumulative utility relative to neutral weights (difference -3.45 [-4.48, -2.43]). The adaptive predictor outperformed its frozen counterpart in late accuracy in each of the six test environments. The other comparisons show how the choice of action threshold and the information available in feedback affect what can be attributed to online adaptation.

cs.LG

新论文 · v1arXiv:2611.23835

NarrowSpec: Draft Hidden Width Need Not Match the Target in Speculative Decoding

Alagappan Valliappan, Ankit More, Dheevatsa Mudigere, Jatin Mitra

原文摘要

Speculative decoding is widely used to speed up generation in large language models (LLMs), with a small draft model proposing tokens that one forward pass of the target model verifies. Draft models that read the target's hidden states, such as DFlash and EAGLE-3, typically inherit the target's hidden width. We show that the draft's whole residual stream can run at half or a quarter of that width while largely preserving acceptance. For DFlash, which borrows the target's vocabulary projection, the trained output adapter folds into it, giving one narrower projection with logits unchanged up to floating-point rounding. We test two drafter families, DFlash and EAGLE-3, on two targets, Qwen3-8B (NVFP4) and GPT-OSS-20B (MXFP4), in SGLang on a B200 and a DGX Spark (GB10). At batch size 1 and temperature 0, narrowing the draft to half or a quarter width speeds up end-to-end decoding by 1.06-1.12x on B200 and 1.13-1.44x on GB10. The speedup holds at larger batch sizes. From batch 32 to the largest batch each GPU fits, narrowing reaches up to 1.21x on both GPUs, and no narrow draft is slower than full width.

cs.CL · cs.LG · cs.PF

版本更新 · v2arXiv:2611.22432

Think Once, Act Fast: Cost-Effective Recursive Self-Improvement for Decision Model Harness

Siqi Lai, Haitao Li, Hao Liu

原文摘要

Fast System-1 models enable responsive actions but struggle with complex or unfamiliar tasks. Large language model (LLM)-based System-2 provides strong reasoning capabilities, yet repeated invocation incurs substantial latency and computational costs. To bridge this gap, we propose JevRSI, a System-1/System-2 collaboration framework that transforms LLM reasoning into transferable decision support through recursive harness improvement. A Jev-style System-1 model enables low-latency action selection, while an asynchronous System-2 LLM develops a modular harness. The harness integrates procedures, state memory, and world models to provide decision contexts spanning the past, present, and future. Utility-driven retrieval selects modules based on both contextual relevance and their utility on previous tasks. Environmental feedback guides targeted revisions and enables recursive self-improvement. Experiments across three datasets show that JevRSI consistently improves decision quality, matching or surpassing LLM agents in several settings. It preserves low-latency execution with favorable performance-cost trade-offs. Moreover, System-2 development interventions decrease as the harness evolves, amortizing System-2 costs through continuous improvement. Our project is available at https://github.com/usail-hkust/JevRSI.

cs.AI

新论文 · v1arXiv:2611.23832

A System 1 Decision Model for Medical Multiple-Choice Questions Built on Frozen Language Models

Pavan Kumar Mantha

原文摘要

A System 1 model answers a multiple-choice question in a single pass. A frozen language model turns the question and each option into vectors, two small, trained heads place them in a shared space, and the closest option wins. The design has been shown on agent decisions, but we are not aware of a study of how far it goes on exam questions that need medical knowledge, where the right option is often no more similar to the question than the wrong ones. We rebuilt the recipe from scratch for MedMCQA and trained only two projection heads and one temperature on top of frozen Qwen3.5-0.8B, Granite-4.2-3B and Qwen3-8B, using a contrastive loss with in-batch and hard negatives and a mask for repeated answers. With plain last-token vectors, both small encoders answer about one question in three correctly (33.1% and 33.0% on 3,842 held-out questions, where chance is 25.0%). Writing each text inside a one-word-summary prompt and lowering the learning rate of the heads raised this to 41.7% and 42.5% for the two small encoders and to 46.0% +-0.2 (three seeds) for Qwen3-8B. Shuffling the questions against the options returns accuracy to chance, so the heads do use the question. Flan-PaLM and Med-PaLM 2 report 57.6% and 72.3% on MedMCQA under a different protocol, so ours is a cheap, fast baseline and not a competitive system. We also say what our evidence cannot show: validation gains were about 1.7 to 2 times the test gains, and we could not separate the effect of reading several encoder layers from the other changes.

cs.AI · cs.CL

新论文 · v1arXiv:2611.23830

Estimation-Aware Reachability

Himanshu Paudel, Javier Borquez

原文摘要

Hamilton-Jacobi Reachability computes the states from which a system can be kept safe under worst-case disturbances, but it assumes the state is known exactly, whereas real robots act on noisy estimates whose uncertainty evolves with their motion. We extend Hamilton-Jacobi Reachability to the robot's belief, estimate and uncertainty, treating unexpected measurements as a bounded adversary and enlarging the unsafe set by the current uncertainty, so that certified states are safe with a prescribed probability. The formulation is compatible with existing solvers. In simulation, the resulting safety filters nearly eliminate the collisions of certaintyequivalent HJ and are less conservative than equally safe uncertainty-padding baselines. The implementation code is available at https://github.com/HimanshuPaudel/Estimation-Aware-Reachability

cs.RO · eess.SY

版本更新 · v3arXiv:2610.05745

Think Before You Distill: Generalization Recovery in Quantized VLAs via Selective Policy Distillation

Shen Ruan, Wenchang Gao, Jin Wang, Siao Liu, Zhaxizhuoma, Dongchun Ren, Xin Zheng

原文摘要

Post-training quantization has been shown to preserve VLA performance under standard evaluation conditions, but whether it preserves the full-precision model's robustness and generalization remains underexplored. In this study, we systematically study the robustness and generalization of post-quantized VLA policies under environmental disturbances. Empirical results show that quantized policies can become fragile to subtle environmental variations despite retaining comparable in-distribution performance. We further observe that action discrepancies are concentrated in a small subset of the rollout states, while teacher guidance has opposite effects depending on discrepancy: it improves performance at high-discrepancy states but can degrade it at low-discrepancy states. These findings reveal that effective post-quantization recovery requires selective intervention on vulnerable states rather than globally distilling the student. We therefore propose Policy-Induced Vulnerability-Oriented Tuning (PIVOT-Q), a vulnerability-aware On-Policy Distillation (OPD) framework that selectively corrects vulnerable states encountered during quantized-student rollouts using the frozen full-precision policy as a teacher. PIVOT-Q identifies vulnerable states using discounted accumulated discrepancies over a short horizon, applies phase-balanced sparse supervision, and uses a Behavioral Anchor to prevent unintended changes. Experiments under seven LIBERO-Plus environmental variations demonstrate consistent recovery across multiple VLA backbones and quantization methods. Notably, PIVOT-Q outperforms full-state distillation with smaller variance across evaluated settings while using only 20\% of its state-level distillation budget. Our code is available at https://github.com/ruanruan-andy/PIVOT-Q.

cs.LG · cs.RO

新论文 · v1arXiv:2611.23825

Learning Martial Arts Motion Skills for Animating Opponents in VR Sparring Games

Jiqing Wen, Qixuan Shi, Robert LiKamWa, Pavan Turaga

原文摘要

Animating virtual opponents is an important yet challenging task in game development. In martial arts games, opponent motions must not only be physically plausible but also exhibit appropriate technique, timing, transitions, and target-directed intent. Traditional game animations often rely on handcrafted clips, state machine transitions, and procedural adjustments, which can be labor-intensive and may limit the behavioral diversity of non-player characters. This paper presents a martial arts motion skills learning framework for character animation in virtual sparring games. Inspired by Tang Soo Do, we collected a martial arts motion capture dataset containing isolated attacking and defending techniques, technique combinations and transitions, and contextual sparring variations. We first pretrained a reusable repertoire of martial arts motion skills from the captured data using reinforcement learning (RL) with adversarial style rewards, and then adapted these skills with task-specific rewards for target-directed striking and threat-aware evasion. We evaluated the motions generated by the learned policy through a user study, in which participants assessed their physical plausibility, smoothness, and martial arts form. To demonstrate their applicability to interactive gameplay, we developed a VR sparring prototype for Meta Quest headsets. The prototype combines policy-generated motion clips with lightweight state-based control to support interactive opponent behavior. Together, these findings support the feasibility of using a learned martial arts skill repertoire as a motion foundation for animating opponents in VR sparring games.

cs.HC · cs.AI · cs.MM · cs.RO

新论文 · v1arXiv:2611.23824

Evidence-Based Credit Assignment for Search-Augmented Language Models

Wen Wang, Jindong Wang, Jiahua Bao, Tu Yongsiqi, Yihao Liu, Junwei He, Wenkui Fan

原文摘要

Reinforcement learning from terminal answer rewards enables search-augmented language models to interleave reasoning and retrieval without step-level annotations, yet outcome supervision assigns identical credit to every search in a trajectory regardless of whether it retrieved useful information. This limitation is especially severe when all sampled rollouts for a question fail, collapsing group-centered outcome advantages to zero even when trajectories gather markedly different evidence. We present Evidence-based Credit Assignment (ECA), which augments terminal outcome rewards with a training-only auxiliary signal measuring how retrieved documents improve the likelihood of a reference answer. Rather than scoring the full dialogue prefix, ECA queries a frozen snapshot of the current policy under an evidence-only prompt containing solely the question and the deduplicated documents observed so far, isolating the utility of retrieved passages from surface variations in generated thoughts or search queries. Within this framework, we compare two return formulations, ECA-Step (which accumulates step-clipped potential increments) and ECA-Horizon (which preserves the telescoping horizon gain in potential), paired with cross-fold group normalization, novelty gating, and search-local token assignment. Across seven Search QA benchmarks, ECA-Horizon improves strict macro exact match over an outcome-only GRPO baseline by 1.03 and 4.80 percentage points on Qwen2.5-3B- and 7B-Instruct, respectively, outperforming both RLOO and GiGPO. Zero-shot evaluation on MultiHop-RAG further demonstrates strong transfer to multi-hop inference queries.

cs.AI

新论文 · v1arXiv:2611.23822

Mapping the Design Space of Feature Sampling for Optical Flow

Liping Zhang

原文摘要

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

cs.CV

版本更新 · v4arXiv:2609.31620

Rethinking the Reconstruction-Generation Gap in Representation Autoencoders

Hongyang Du, Yunfei Xie, Junjie Ye, Jiawei Yang, Xiaoyan Cong, Haodong Zhang, Yongchao Huang, Haiyu Wu 等 16 位作者

原文摘要

Fusing shallower encoder layers into a Representation Autoencoder (RAE) sharpens pixel reconstruction but degrades diffusion generation. Prior studies and the research community commonly attribute this trade-off to the latent content itself, assuming detail-rich features are harder to model. We show that this gap stems from geometric mismatch at the encoder-decoder interface. Freezing a foundation encoder preserves rich visual information, but imposes an encoder-native Euclidean metric shaped without pixel decoding, whereas the decoder amplifies sampling errors sharply along sensitive directions. At fixed information, reducing pixel error admits two dual geometric interventions: regularizing the decoder to flatten its sensitivity to latent variations, and aligning diffusion coordinates with the frozen decoder's metric at identical clean reconstruction. On ImageNet-256, aligning diffusion coordinates largely closes the guided generation penalty of multi-layer fusion and reaches state-of-the-art 0.984 gFID and 2.14 FDr6 in 80 epochs. RAEs should take information from the encoder and geometry from the decoder.

cs.CV

新论文 · v1arXiv:2611.23817

Metrics-in-the-Loop: Complexity Feedback Reshapes Agent-Written Code

Nicolas Chapados, Jean-François Gagné, Nouha Dziri

原文摘要

Coding agents now write whole codebases in long autonomous runs, guided mainly by whether their tests pass. Yet passing tests alone does not reveal accumulating complexity or gaps in test coverage, which then persist as agents extend their own code. We introduce metrics-in-the-loop, an inference-time intervention that gives agents in-flight feedback: cheap, automatically computed code metrics injected into the agent's context during the trajectory. After each code-changing agent turn, we insert short feedback messages naming changed functions whose cyclomatic complexity exceeds a limit (complexity feedback) or changed lines that are skipped by tests (coverage feedback), giving the agent the opportunity to refactor code or strengthen tests before continuing. We run extensive experiments and ablations comparing several interventions across three long-horizon coding benchmarks and four strong models. We find that complexity feedback significantly reduces erosion, a measure of complexity concentration of code, without affecting held-out correctness of the code. Broadly, our experiments investigate a practical form of agent "re-railing", where timely, actionable measurements redirect agents before weaknesses become the foundation they keep building on. As agents scale to codebases with millions of lines, such interventions become critical to sustaining effective development and long-term code maintainability.

cs.AI

新论文 · v1arXiv:2611.23816

RoboDx: Repair and Diagnosis of Faults in Robot Software

Soham Dan

原文摘要

A patch to a robot's navigation software can pass every unit test and still drive the robot off its path. Checking how the repaired robot drives catches such patches. It does not show that the agent found everything that was wrong. We introduce RoboDx, a generator of repair tasks for simulated robot navigation software with known injected faults. A hidden verifier marks a repair as verified only if the robot then drives correctly in hidden scenarios, which the agent never sees. A second track hides the code and asks the agent to pick the cause from a list of hypotheses. A task with two faults can contain a quiet fault, a fault that makes no hidden scenario fail unless the other fault is also present. On our set of 100 tasks, RoboDx-Hard-100, almost every repair by Claude Opus 5.5 with standard coding tools is verified. We analyse 94 of the tasks, 44 with one fault and 50 with two. On the 50 two-fault tasks, only 17 of Opus's 49 verified repairs remove both injected edits. Most faults left behind are configuration constants whose correct value the task's documentation does not state. Opus names both faults on 10/50 of these tasks when it cannot see the code. We then give four current models the same measurements of each faulty robot, picked so that, in the simulator, only the true cause produces all of them. The models name the cause on at most 38 of 94 tasks. They are right on almost all when they also get the simulator's output for every hypothesis. Their own predictions of that output give no detectable gain. On 97 tasks from real bug-fix pull requests, Claude Haiku 5.5, the weakest model on the generated tasks, scores lowest, and the other three models are not ordered.

cs.RO · cs.SE

新论文 · v1arXiv:2611.23815

CAFE-AQI: Coverage-Aware Day-Ahead AQI Forecasting with Early Warning and Conformal Uncertainty in Bangladesh

Md Rafid Islam, Rafsan Jany, Rafid Bin Zaman, Mohammad Fahim Islam, Hafiz Abdur Rahman

原文摘要

Next-day air quality forecasting often assumes city records are equally complete and independent. However, an audit of a large Bangladesh dataset revealed a sharp coverage expansion after August 2022 and showed that 103 city labels correspond to only 61 unique AQI trajectories. To address these issues, this paper presents CAFE-AQI, a coverage-aware framework for forecasting next-day maximum AQI. It also estimates the probability of Unhealthy-or-worse conditions (AQI >= 151). Data were divided chronologically into training (2022-2023), calibration (2024), and testing (2025). To prevent duplicate records from dominating training and evaluation, identical trajectories were weighted. The framework combines temporal and pollutant features with a five-neighbor spatial context. It also includes a class-balanced warning head and risk-conditional conformal calibration aimed at providing early warnings and prediction intervals. Evaluation on the duplicate-balanced 2025 data gave an MAE of 9.76, an RMSE of 15.04, and an R-squared value of 0.910 for CAFE-AQI. Its MAE was 3.63 AQI lower than persistence (95% CI: 3.49-3.76), with lower errors in each of the 61 trajectory groups. For unhealthy-air warnings, the F1-score was 0.908 and the AUROC was 0.982. The nominal 90% intervals covered 92.09% of observations. When an entire spatial block was excluded from training, the MAE remained 10.66, showing that the lightweight model transferred reasonably well to unseen locations.

cs.LG · stat.AP

版本更新 · v2arXiv:2610.26651

The Effectiveness and Durability of Pretraining Data Filtering in Protein Language Models

Aaron Maiwald, Roderick McNeill, Jaeyoung Lee, Matthew A. Clarke, Lennart Justen, Alejandro Acelas, Ivo Andrews, Giles Harper-Donnelly 等 12 位作者

原文摘要

Pretraining filtering is one of the most common safeguards for open-weight language models. Here we study its effectiveness, durability and scalability using the ESM-2 family of protein language models. We find that filtering viral proteins from the pretraining data meaningfully degrades masked language modelling, mutation effect prediction and long-range contact prediction on viral proteins, while largely preserving performance on non-viral proteins. These effects are concentrated among proteins from human-infecting viruses, and grow as we scale models from 8M to 3B parameters. As models increase in size, we find that stringent filtering becomes more effective. However, fine-tuning on sequences from the removed viral families restores the lost performance. Although the cost to recover performance increases with model scale, even for the largest models it costs about 80 USD of H100 time. This work establishes pretraining filtering as moderately effective at selectively degrading performance on viral proteins, but not durable against fine-tuning with modest computational resources and access to relevant data.

cs.LG · cs.AI

新论文 · v1arXiv:2611.23813

SWIFT: Slow World, Instant Freshness, and Timely Action for Asynchronous Autonomous Driving

Jiazhe Wang, Tianyi Yang, Ruihang Li, Fan Lu, Xianming Zeng, Jianyun Xu, Xingliang Liu, Lei Yang

原文摘要

Driving scenes keep changing during model computation, making both observation freshness and action-delivery time central to world-action planning. To retain rich generative world priors in autonomous driving without stalling real-time control, we present SWIFT (Slow World, Instant Freshness, and Timely action), a deployment-oriented asynchronous world-action system that moves world updates off the observation-to-action critical path. On a background GPU, a slow Video DiT streams surround-view history into read-only predictive world context, trained with a predictive span that covers both the update interval and background world computation. On a foreground GPU, a lightweight freshness carrier encodes current surround-view images into a gated key/value bank at every step, letting a fast Action DiT combine live visual feedback with the latest completed world context without waiting. On open-loop NAVSIM, SWIFT reaches 94.2 PDMS at 198.4 ms per decision with a trajectory scorer; on closed-loop Bench2Drive, it achieves an 80.30 Driving Score at 109.9 ms foreground latency. SWIFT further transfers to the Waymo E2E hidden test (8.0050 Rater Feedback Score) and nuScenes (0.54 m mean endpoint error, 0.03% mean collision rate). Across all four benchmarks, SWIFT achieves the best results among compared world-action methods, establishing a favorable accuracy-latency trade-off.

cs.CV

版本更新 · v2arXiv:2609.13226

Synthetic Leprosy Image Generation Using Mask-Conditioned Latent Diffusion and Transfer Learning from Large Chronic Wound Datasets

Yusuf Abdulkadir

原文摘要

Machine learning for neglected tropical diseases is limited by data, not algorithms: public annotated image sets for leprosy (Hansen's disease) number in the hundreds. We ask whether a model trained on abundant chronic wound photography transfers to this low-data regime. We build a three-stage pipeline: a DeepLabV3-ResNet50 segmenter supplies lesion masks for two wound datasets that ship without them; a mask-conditioned latent diffusion model built from Stable Diffusion 1.5 components is trained on 3,280 wound crops; and the model is fine-tuned on leprosy image-mask pairs from CO2Wounds-V2 (764 images of 96 patients). We evaluate on 136 held-out leprosy lesions using LPIPS anchored by a real-versus-real baseline, kernel inception distance (KID), and a paired design that runs both models on the same input. Fine-tuning moved the generated distribution toward real held-out leprosy images: KID fell from 0.167 to 0.089, and LPIPS distance to real images from 0.806 to 0.765, with non-overlapping 95% confidence intervals. Giving the base model the leprosy prompt alone did not reproduce the gain (KID 0.250). For a fixed input, fine-tuning changes the output less than a new random seed does (paired LPIPS 0.544 versus 0.728): it keeps layout and alters tissue color and texture. The generated set shows no mode collapse (internal diversity 0.734 versus 0.667 for real images). A dataset audit shows that a majority of training images carry synthetic flat regions from preprocessing, which the fine-tuned model partly reproduces. Chronic wound photography is a viable donor domain for leprosy lesion synthesis; the remaining barrier is semantic control rather than image quality.

cs.CV

新论文 · v1arXiv:2611.23810

Continual Tool-Space Extension Learning for Multimodal Agents

Jinbae Seo, Jungin Park, Hyunsung Jang, Namkoo Ha, Jiyoung Lee, Kwanghoon Sohn

原文摘要

Multimodal agents largely rely on generic tools such as cropping, search, and code execution, leaving specialist vision models outside their tool space. However, bringing such a model into an already trained agent is not straightforward. The challenge is to learn each new tool efficiently while retaining the ability to use existing ones. This work proposes COTEX, continual tool-space extension learning, that adds new tools at efficient adapter-level cost. COTEX locates two sets of parameter directions; one triggers a tool call and thereby opens a multi-turn reasoning trajectory, while the other determines which tool is chosen. The adapter of a new tool is restricted to the calling directions that leave earlier tool choices unaffected by orthogonal constraint. We further introduce a protocol that converts specialist models' existing task-specific annotations into verified tool-use trajectories. Experiments on two representative agents and three specialist tools show gains on ten vision-language benchmarks, with at most a 0.3-point drop on existing tools.

cs.AI · cs.LG

新论文 · v1arXiv:2611.23808

PReMind: When Personal Memory Meets Visual Hazards

Yeongtak Oh, Donghyeok Lee, Joonhee Oh, Jaewon Jeong, Sungroh Yoon

原文摘要

Personalized vision-language assistants use visual observations and personal memories to provide advice tailored to the user. When the observed scene contains a physical hazard, personal context, such as allergies, medical conditions, or mobility constraints, can change what help is appropriate. Responding appropriately therefore requires recognizing the visible hazard and applying the relevant person's information. We introduce PReMind-Bench, a diagnostic benchmark that evaluates this combination across diverse hazards and personal-memory conditions. It separately measures memory access, grounding, abstention, and hazard response, and assesses joint success on interdependency cases. Existing models often satisfy one requirement while failing the other, and training only for safety or personalization leaves a gap in joint success. We therefore propose PReMind, a two-phase reinforcement learning framework that combines on-demand and forced-call interactions (memory always retrieved) to jointly train grounded memory use and hazard response. At 4B, PReMind raises joint success from 12.9% to 28.6% over its base on the interdependency set and improves hazard adequacy and correct abstention on the main test. Joint-success gains persist under forced-call evaluation and under a second LLM judge across three Phase II seeds; averaged over these seeds, joint success also improves on unseen hazard groups. These findings highlight the need to evaluate and train personal assistants to connect visual evidence with the relevant person's context.

cs.CV

新论文 · v1arXiv:2611.23807

Sightline: Long-Window Speculative Decoding for Fast SVG Generation

Yanchao Li, Ben Gao, Jiaqing Xie, Yusu Zhang, Beilun Wang, Yanbo Wang, Xiaohua Xu, Nanqing Dong 等 10 位作者

原文摘要

Scalable Vector Graphics (SVG) generation models write a graphic as code with one forward pass per token, so drawing is slow. Speculative decoding speeds them up losslessly, and block drafters guess many tokens per pass. On SVG, many of their rounds accept the whole draft and end only at the verification window. We find that their misses are rare and clumped, and that they stay accurate far deeper into a block than on natural language. A run-length identity and three propositions show that a longer window pays once the candidates are wide, up to the drafter's horizon. We propose Sightline, which pairs a window as long as the drafter's horizon with a draft tree wide enough to fill it. Experiments cover text-to-SVG and image-to-SVG with three models on six benchmarks from MMSVG, MMSVG-2M and SArena. In every setting, Sightline is the fastest of the compared lossless decoders. On SArena, it stays the fastest when every decoder shares one loop control. It also leads in almost every setting of six models from four other families, which write SVG code, compressed SVG tokens or Lottie JSON. Analyses trace the lead to rounds that shorter windows cut off, both in loop-free content and in loops. Speculative decoding also saves time for choosing samples by rendering, which beats choosing by likelihood. These results suggest that speculative decoding on vector graphics should verify as far as the drafter sees. Code and data are available at https://github.com/LiZaiyuan0619/Sightline.

cs.CV · cs.CL · cs.LG

新论文 · v1arXiv:2611.23804

TacPilot: Does Touch Need to Be Everywhere?

Yunzhen Zhang, Yuxiao Zhu, Jie Yang, Yufeng Miao, Bing Luo

原文摘要

Tactile feedback helps visual manipulation policies respond to local contact, but changes in tactile observations do not always require corrective motion, and sustained corrections do not always require new tactile features. We introduce TacPilot, a tactile residual framework that separates representation updates from action intervention around a frozen visual policy. TacPilot uses a perception gate to determine when tactile features need refreshing and updates only affected regions while reusing unchanged features. A separate intervention gate determines whether bounded Cartesian and gripper residuals modify the nominal visual action. This separation allows tactile perception and action correction to follow different spatial and temporal demands during manipulation. In offline expert-trajectory replay, local feature reuse reduces encoded regions by 56.4% relative to periodic full encoding, with essentially unchanged action predictions. In an exploratory simulation evaluation spanning two cylinder sizes, TacPilot succeeds in 26 of 30 trials (86.7%). We further validate TacPilot on OpenArm v2 equipped with Paxini tactile sensors across four contact-rich manipulation tasks.

cs.RO