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

+1112新增论文
+875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 05:42 · 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 06:37 — 2026-11-20 06:37(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 05:42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 84 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-20 05:40(北京时间)
新论文 · v1arXiv:2611.24846

Atomic MathReasoner:面向数学大模型的原子推理技能实例级组合

Atomic MathReasoner: Instance-Level Composition of Atomic Reasoning Skills for Mathematical LLMs

Jiashu Li, Tianyu Gao, Qinglong Liu, Jinping Wang, Ye Wei, Zhiqiang Gao

AI 速读 基于摘要

AMR 先训练一组能力专属的 LoRA 专家,再在验证集上无梯度搜索低损失组合原型,并蒸馏出轻量路由器,为每道题选择实例级技能组合。

cs.LG

查看原文摘要

Most existing approaches enhance mathematical reasoning in large language models (LLMs) as a monolithic capability, despite the fact that different mathematical problems require different reasoning skills. Recent work has explored a more fine-grained perspective by decomposing mathematical reasoning into atomic skills and developing specialized modules for individual capabilities. However, how to effectively utilize these atomic abilities remains underexplored. To address this challenge, we propose Atomic MathReasoner (AMR), an instance-level framework for composing atomic reasoning skills. AMR first trains a library of capability-specific LoRA experts and performs gradient-free optimization over held-out validation subsets to discover a diverse set of low-loss composition prototypes. It evaluates the preference of each prototype using the token-level negative log-likelihood of reference solutions, filters ambiguous assignments through a confidence margin, and distills the resulting pseudo-labels into a lightweight router. During inference, the router selects an appropriate composition prototype for each problem, enabling problem-specific utilization of atomic skills without costly online optimization. Experiments on the AtomicMath benchmark across general and mathematics-specialized LLMs demonstrate consistent improvements. On Qwen2.5-Math-Instruct-7B, AMR increases average field accuracy from 64.4% to 68.1% at Level 1 and from 59.3% to 63.7% at Level 2, while improving average logical reasoning performance from 39.4% to 46.8%. These results demonstrate that effectively composing specialized atomic skills provides a promising approach to strengthening mathematical reasoning in LLMs.

新论文 · v1arXiv:2611.24395

RSIBench-Infra: Can Agents Optimize Their Own Training and Inference Infrastructure at Production Scale?

Hanlin Wu, Yifei Wang, Zhicheng Cai, Huan-ang Gao, Wei-Ying Ma, Ya-Qin Zhang, Jingjing Liu, Hao Zhou

原文摘要

Improving the efficiency of training and inference infrastructure is a practical component of recursive self-improvement. Existing benchmarks for agent-driven optimization largely focus on individual software layers such as kernel, leaving unclear whether agents can deliver end-to-end performance gains under production workloads. We introduce RSIBench-Infra, a benchmark of 20 training and inference tasks spanning diverse model families and modalities, with deployments of up to 32 accelerators on NVIDIA GPUs and Ascend NPUs. Agents start from runnable baselines derived from official recipes and can modify the software stack across engine, kernel, and instruction levels. An OCI image-layer submission protocol captures these changes for reproducible deployment, while independent evaluation measures end-to-end performance under task-specific quality constraints. Across all task results, 66.2% exceed $1.10\times$ validated speedup. These results demonstrate that current agents can improve production-scale training and inference performance, but identifying effective optimization strategies and consistently translating local improvements into end-to-end gains remain challenging.

cs.LG

版本更新 · v2arXiv:2610.14094

Scenes as an Attack Surface: Uncovering Scene-Grounded Safety Failures in MLLMs

Tongxi Wu, Yuyang Cheng, Jiacheng Xian, Hu Huang, Zhihao Wu, Menglin Yang, Jian Zhang, Yang Gao

原文摘要

Existing jailbreak attacks have exposed serious safety weaknesses in multimodal large language models (MLLMs). Most prior attacks rely on specially constructed visual inputs to carry or support harmful intent. In this paper, we uncover a different vulnerability: task-relevant scene semantics alone can shift an MLLM from refusal to compliance, even when the visual input introduces no additional attack instructions and the original task intent remains unchanged. We formalize this failure as a scene-grounded visual safety vulnerability and introduce SceneSafetyBench to investigate it across eight risk categories and nine white- and black-box MLLMs. Our analysis reveals that this failure is associated with how task-relevant visual evidence influences downstream safety representations, suggesting that visual grounding alone cannot fully explain the observed safety shift. Building on this finding, we propose Scene-Grounded Attacks (SGA), which preserve the original task while constructing validated scene-grounded queries and combining their independent responses. Across nine target MLLMs, SGA achieves an average attack success rate (ASR) of 86.17%, outperforming the strongest baseline by 8.55 percentage points and achieving the highest ASR on eight models. Our results identify scene semantics as an overlooked attack surface in MLLMs and highlight a growing tension between stronger visual grounding and robust multimodal safety alignment.

cs.CV

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:2611.23206

阅读场景状态:揭示视觉语言模型中的安全失效

Reading Scene States: Uncovering Safety Failures in Vision-Language Models

Tongxi Wu, Yuyang Cheng, Jiacheng Xian, Menglin Yang, Kai Chen, Jian Zhang, Yang Gao

AI 速读 基于摘要

论文识别出视觉语言模型的“场景状态安全漏洞”,并构建基准与攻击方法 S3TA,作者报告其攻击成功率显著高于基线。

cs.CV · cs.AI

查看原文摘要

Vision-language models (VLMs) can provide harmful assistance by interpreting visual events, even without an explicit request for the underlying harmful objective. We identify this \textbf{scene-state safety vulnerability} and introduce \textbf{SceneSafetyBench}, comprising 375 image--question pairs spanning Initial, Medium, and Ultimate scene states and five visual presentations. Across eight VLMs, scene-state inputs elicit harmful assistance with attack success rates (ASRs) of 45.6--72.3\%. Comparative analysis across three VLMs shows that scene-state inputs most closely resemble QR-related images in their layer-wise safety-related signal profiles, while exhibiting lower average late-layer readouts and distinct signal dynamics during generation. To turn this vulnerability into a systematic attack, we propose \textbf{S3TA}, which translates harmful objectives into scene states, queries image--question pairs independently, and assembles the responses. On seven MM-SafetyBench categories across eight open-weight and commercial VLMs, S3TA achieves a model-average ASR of \textbf{71.2\%, exceeding the strongest baseline by 6.1 percentage points}. Text-side filters and visual moderation provide limited protection. These findings highlight the need to safeguard the assistance models infer through scene-grounded reasoning, beyond screening input content.

新论文 · v1arXiv:2611.23101

StreamingDPD:面向实时320-MHz数字预失真的1.28-GSa/s瞬时功率门控深度可分离TCN加速器

StreamingDPD: A 1.28-GSa/s Instantaneous-Power-Gated Depthwise TCN Accelerator for Real-Time 320-MHz Digital Predistortion

Yizhuo Wu, Xinling Wang, Chang Gao

AI 速读 基于摘要

提出一种基于瞬时功率门控深度可分离时序卷积网络(IPG-DWTCN)的定点流水线加速器,以1.28 GSa/s处理320-MHz信号,用于实时数字预失真。

eess.SP · cs.AR

查看原文摘要

State feedback between successive samples limits the streaming throughput of recurrent neural networks for digital predistortion (DPD). We present StreamingDPD, an accelerator that processes 320-MHz signals at 1.28 GSa/s. Its 207-parameter instantaneous-power-gated depthwise temporal convolutional network (IPG-DWTCN) uses $|x[n]|^2$ to gate feature channels and causal dilated depthwise convolutions to model power amplifier memory. The network maps to a spatially unrolled fixed-point pipeline with shift-based quadratic activations and layer-wise mixed precision, sustaining one complex sample per cycle per lane. Its algorithmic cost is 554 floating-point operations per complex sample, 30.1% fewer than the recurrent baseline with the best linearization performance. In RF playback measurements on three power amplifiers driven by 40- to 320-MHz Wi-Fi 6/7 and 5G NR waveforms, IPG-DWTCN achieves the lowest mean adjacent-channel power ratio (ACPR) and ranks first or ties for first in mean error vector magnitude (EVM) against five neural baselines of comparable size. Four-lane FPGA hardware-in-the-loop tests on a 320-MHz 4096-QAM Wi-Fi 7 waveform achieve -45.82 dBc ACPR and -42.94 dB EVM at 1.28 GSa/s. At the same sample rate, post-layout evaluation of a single-lane 12-nm FinFET ASIC yields a 0.10 mm$^2$ core, a 52-cycle (40.63 ns) latency, and an estimated typical-corner power of 142.21 mW (111.10 pJ/sample). Relative to the prior 22-nm recurrent DPD ASIC, the design provides 5.1 times the throughput at about one-seventh the energy per sample.

新论文 · v1arXiv:2611.22625

VLM-guided Grounding Expert for Generalised Referring Expression Comprehension

Mingcheng Zhu, Wentian Xu, Tao Lin, Shiying Duan, Ziyun Liang, Xuecheng Qi, Junyang Chen, Kuo Li 等 12 位作者

原文摘要

Generalised referring expression comprehension (GREC) aims to identify and localise the zero, one or multiple objects referred to by a language expression. In robotic manipulation, grounding also needs to remain accurate when requests for the same objects are paraphrased or embedded in additional context in robotics. However, grounding expert models are sensitive to instruction variation, while generative vision--language models (VLMs) struggle with complete target recovery and rejection of absent targets. These complementary limitations motivate combining the contextual representations of VLMs with the direct spatial prediction capabilities of grounding experts. We therefore introduce VLM-guided Grounding Expert (VGE), which combines a frozen VLM with a trainable grounding expert through layerwise joint attention and text-conditioned instance scoring. These interfaces leverage multimodal understanding of the VLM to guide object-query refinement and candidate selection while retaining direct bounding-box prediction. Across three datasets, VGE achieves the highest averaged F1 score among the evaluated methods on all five test splits, with differences of at most 1.2 percentage points across instruction conditions within each split. On the dataset with robotic scenes, VGE also achieves the highest F1 across instruction variants and no-target, single-target and multiple-target cases. VGE provides a $5.3$--$13.0\times$ inference speedup over the baseline with the same VLM base model. These results support combining VLM context with specialised spatial prediction to address both grounding challenges, with particularly strong performance in robotic scenes. Code will be released upon publication.

cs.RO

新论文 · v1arXiv:2611.22388

Hidden Variables Behind Real-Robot Data Quality: Lessons from Hundreds of Thousands of Hours

Maniformer Real-World Manipulation Team, Che Tsang Calvin, Xingyu Chen, Bowen Jiang, Naixiang Gao, Xiongfeng Cai, Enlin Gu, Ruiying Li 等 42 位作者

原文摘要

As real-robot data collection scales, collection-system design shapes both operator efficiency and the behaviors learned by robot policies. Workspace limitations, control-interface friction, and latency can introduce pauses, detours, and corrective motions into otherwise successful demonstrations. We investigate these effects across embodiment, teleoperation control, and data consistency, combining a kinematic study of 25 dual-arm robots, controlled teleoperation experiments, and real-robot evaluation of fine-tuned $π_{0.5}$ policies. The embodiment study shows that positional reachability overstates orientation-aware task coverage and that modest shoulder motion can substantially expand the effective workspace. Teleoperation experiments show that parallelizing body-part control reduces task completion time by 43-53%, while compliant control improves success and reduces peak contact forces on constrained and contact-rich tasks. Policy-learning experiments show that demonstrations must capture the visual information underlying action choices: active-search demonstrations improve arm-selection accuracy from chance (50%) to 100%. Calibration experiments reveal that mixed-calibration training data remain usable in our tested configurations when an inference-aligned subset is retained, whereas inference-time miscalibration causes all 30 trials to fail. Policies also retain approximately 90% of the extra path length in detour-contaminated demonstrations, showing how inefficient operator motions persist in learned behavior. Together, these findings connect collection-system design to demonstration quality and downstream policy execution. They motivate evaluating effective workspace, control efficiency, observation completeness, calibration alignment, and motion quality before scaling collection, and addressing system-induced compensations at their source.

cs.RO

版本更新 · v2arXiv:2610.23909

When the Editing Intent Is Split: A Cross-Modal Jailbreak Attack for Large Image Editing Models

Xianmin Ye, Zhiyuan Fan, Tianju Liu, Yuhang Wang, Zhenxing Niu, Haichang Gao

原文摘要

Large image editing models increasingly infer editing intent from both textual and visual inputs, creating an underexplored safety risk when harmful intent is distributed across modalities. In this work, we formalize this split-intent vulnerability and propose the Split-Intent Jailbreak Attack (SIJA), which asymmetrically distributes an unsafe editing intent by specifying the editing referent in text while concealing the safety-critical action and specification through graphical visual semantics in an auxiliary cue. To systematically study this threat, we introduce the Split-Intent Image Editing Safety Benchmark (SI2ESBench) for evaluating image editing models under split-intent inputs. Extensive experiments on representative commercial and open-source models demonstrate that SIJA effectively compromises state-of-the-art image editing systems, achieving ASRs of 68.28% on GPT Image 2 and 91.38% on Qwen-Image-3.0-Pro. Our results expose a mismatch between multimodal editing and safety assessment: the distributed intent can be recovered to execute the requested transformation, yet its harmfulness often remains undetected by existing safeguards. To mitigate this vulnerability, we propose editing-intent reconstruction, a defense principle that explicitly evaluates the complete transformation implied by the joint multimodal context and substantially improves rejection of split-intent attacks. Our findings expose a cross-modal safety gap in modern image editing systems and highlight the need to assess the jointly inferred transformation rather than individual inputs in isolation. Project page: https://github.com/yexianmin/SIJA.

cs.CV · cs.AI

版本更新 · v2arXiv:2601.21177

Flow Perturbation++: Multi-Step Unbiased Jacobian Estimation for High-Dimensional Boltzmann Sampling

Xin Peng, Ang Gao

原文摘要

The scalability of continuous normalizing flows (CNFs) for unbiased Boltzmann sampling remains limited in high-dimensional systems due to the cost of Jacobian-determinant evaluation, which requires $D$ backpropagation passes through the flow layers. Existing stochastic Jacobian estimators such as the Hutchinson trace estimator reduce computation but introduce bias, while the recently proposed Flow Perturbation method is unbiased yet suffers from high variance. We present \textbf{Flow Perturbation++}, a variance-reduced extension of Flow Perturbation that discretizes the probability-flow ODE and performs unbiased stepwise Jacobian estimation at each integration step. This multi-step construction retains the unbiasedness of Flow Perturbation while achieves substantially lower estimator variance. Integrated into a Sequential Monte Carlo framework, Flow Perturbation++ achieves significantly improved equilibrium sampling on a 1000D Gaussian Mixture Model and the all-atom Chignolin protein compared with Hutchinson-based and single-step Flow Perturbation baselines.

cs.LG

版本更新 · v2arXiv:2611.16861

Visual Cipher Bench: Benchmarking LLM Agents on Long-Horizon Visual Reasoning

Deyao Hong, Jiayi Jiang, Yuchun Chen, Yixuan Shi, Wenyi Li, Zheng Jiang, Chengxing Xie, Mingyu Su 等 15 位作者

原文摘要

While language-model agents excel at long-horizon coding and terminal tasks, this success relies on two conditions: explicit text specifications and immediate execution feedback. However, neither condition holds in real-world problems where critical information exists only in images, leaving it unclear whether long-horizon capability transfers from text to visual reasoning. In such visual settings, because rules and clues resist text transcription and lack test-case feedback, agents must repeatedly inspect images, simulate spatial transformations, and autonomously cross-check clues. Despite this challenge, existing evaluations fail to capture such capability: visual reasoning benchmarks largely reduce tasks to single-turn, short-chain question answering, whereas agent benchmarks focus on text-based environments or interactive states that permit trial-and-error. To bridge this gap, we introduce Visual Cipher Bench, a collection of 40 original visual puzzles across four categories: spatial manipulation and simulation, logical constraints and planning, rule induction, and knowledge-based decoding. Each puzzle provides only one to three black-and-white line drawings and a minimal submission instruction, requiring agents to discover the underlying solving mechanism entirely from the images. Across 29 reasoning-effort configurations of eight frontier models, even the strongest model, GPT-6 Astra (max), achieves only 26.7% accuracy, while five of the eight models score below 1%. Yet on a ten-task subset, providing a single non-spoiler mechanism hint boosts accuracy from 12.2% to 51.1%. This sharp contrast shows that the primary bottleneck of current agents lies not in executing multi-step plans, but in autonomously discovering rules from visual evidence. Visual Cipher Bench thus provides a rigorous benchmark for measuring long-horizon visual reasoning in agents.

cs.CV

版本更新 · v2arXiv:2610.16346

CellChain: Benchmarking Cross-Scale Evidence Use for Perturbation Reasoning

Xue Xia, Yue Wang, Shuizhou Chen, Lang Yu, Hao Wu, Wenxuan Huang, Chengkai Yao, Zhangyang Gao 等 9 位作者

原文摘要

Recent advances in foundation models and large-scale cellular perturbation datasets have driven growing interest in using LLMs to predict and reason about cellular responses. Biological reasoning in this setting involves relating molecular mechanisms, cellular responses, and functional outcomes across scales. Evaluating foundation models therefore requires assessing not only whether their predictions are correct, but also how those predictions depend on the biological evidence provided. We introduce CellChain, an experimentally grounded benchmark comprising 590 biological cases and 2,360 matched evidence-condition instances across gene-dependency prediction, cytokine-axis identification, and drug-sensitivity ranking. CellChain links perturbation experiments and external biological resources through a provenance-aware evidence graph spanning molecular, pathway, transcriptional, and phenotypic information. Task-specific conditions vary evidence coverage, composition, or assignment while preserving questions, candidate answers, reference targets, and scoring rules. Across 18 foundation models, the highest accuracies under prespecified target-evidence conditions are 57.5%, 75.5%, and 70.8%, respectively, and broader evidence coverage does not consistently improve accuracy. On the drug-sensitivity benchmark cohort, matched reference histories outperform swapped histories by an average of 13.0 percentage points in cluster-weighted accuracy. Decision-transition analysis separates corrections, newly introduced errors, and wrong-answer switches, revealing substantial answer changes hidden by small net accuracy differences. CellChain provides controlled comparisons for attributing changes in model predictions to prespecified expansions of biological context, including changes in evidence coverage, composition, and biological assignment.

cs.DB · q-bio.GN

新论文 · v1arXiv:2611.20524

GaugeRoPE: RoPE-Compatible Executable Latent Attention

Yuanhang Gao, Xiangrui Yang, Lingyuan Meng, Yuanfeng Chen, Dongsheng Li

原文摘要

Latent KV cache compression reduces memory footprint by mapping historical keys and values into compact representations. However, generic compressed key states require reconstruction to native dimension before applying RoPE and computing attention scores, adding sequence-scaling overhead to the decode path. We characterize when a latent KV representation can serve directly as the execution state of pretrained RoPE. For RoPE-based transformers, exact static linear execution of the rotary key path requires the latent-to-native map to preserve the pretrained RoPE action through an intertwining relation, restricting exact mixing to coordinates that share the same RoPE frequency. Guided by this characterization, we build GaugeRoPE, a training-free executable latent KV representation. It combines a RoPE-compatible carrier that preserves the rotary action with an approximate position-independent residual key--value (RKV) correction for complementary key and value information, both evaluated without per-token dense historical K/V reconstruction. On Mistral-7B, GaugeRoPE achieves $5.85\times$ KV compression while maintaining quality close to the full-cache baseline. Additional MHA results validate the construction beyond GQA. GaugeRoPE achieves up to $2.48\times$ the decode throughput of the vanilla uncompressed baseline through direct latent execution.

cs.LG · cs.CL

版本更新 · v2arXiv:2611.16647

InertiaMPDD: Multimodal Personalized Depression Detection via Emotional Inertia-Inspired Dynamics Modeling

Changzeng Fu, Songyang Wang, Chang Gao, Xinyi Zhang, Peilin Li, Mo Meng, Jiarui Wu, Mingze Li 等 12 位作者

原文摘要

Multimodal depression screening estimates risk from speech, visual behavior and language. Current methods mainly predict labels, with limited evidence of departures from a non-depressed reference. Recording-level supervision also leaves fine-grained emotional dynamics unannotated. Motivated by emotional inertia, the persistence of affect, InertiaMPDD learns healthy next-state patterns using a world-model-inspired latent predictor. Using pretrained multimodal representations, it predicts from observed history and context. Reference deviation, residual decay rate and a Jacobian spectral descriptor form reference-derived emotional dynamics descriptors, evaluated as candidate proxy biomarkers for standalone screening and personalized decision correction. Across five cohorts from four datasets, fusion reaches 98.41% F1 on D-Vlog and 84.98% on LMVD. Standalone descriptors outperform fixed additive fusion on MPDD but discriminate weakly on AVEC2014. These findings distinguish reference informativeness from fusion benefit and reveal context-dependent utility.

cs.AI

版本更新 · v2arXiv:2611.12807

Suppression Is Not Erasure: Mechanistic Analysis and Reactivation Boundaries of Subspace Projection for LoRA Backdoor Defense

Junsheng Fu, Shuangshuang Gao

原文摘要

Low-Rank Adaptation (LoRA) is widely used to customize large language models, but third-party adapters create supply-chain backdoor risks with under 1% modified parameters. Recent analytical null-space and subspace projection defenses report near-zero static Attack Success Rates (ASR), equating static silence with erasure. We audit subspace-projection defenses across four scales (Qwen2.5-0.5B/1.5B/3B/7B), three architectures (Qwen2.5, InternLM2.5-1.8B, Spark-X2.6-4B), and five trigger families (suffix, natural phrase, prefix/in-sentence, double-negation syntax, and composite triggers). Statically, module-wise rank-1 trigger-difference projection cuts ASR from 60.4% to 0.0% on standard suites and 70.3% to 0.0% on 1,000 uncontaminated prompts (p = 3.055e-299) in under 5 s. By Theorem 1 (11,880/11,880 module-sample pairs), low-rank bottleneck amplification aligns benign polite-suffix probes with u_1(B) (median cosine 0.8934-0.9705), reaching 0.0% ASR without trigger knowledge. However, static suppression is not erasure. Under 5-sample reactivation, single-round weight-editing baselines rebound to 91.7%-100.0% ASR in Epoch 1 (Fine-Pruning, ANP, NEEDLE, LoRA-Patch; 48.7% for single-round dual projection) and 100.0% by Epoch 2. Theorem 2 and Corollary 1 prove single-sided projection leaves non-zero gradients along u_1 across 168/168 modules (27.2x-31.1x above random expectation), while Haar orthogonal rotation shows Epoch-2 immunity depends on semantic alignment rather than Frobenius norm. Guided by Theorems 1-2, we propose Single-Shot Rank-k Dual-Subspace Erasure (DSE) and Benign Subspace Anchoring (DSE-BA), achieving 0.0% static ASR and 0.0% Epoch-1 reactivation across all five trigger families and Qwen2.5-0.5B/1.5B, while cutting Epoch-2 ASR to 25.0% (43.33% +/- 2.49% on 50 prompts x 3 seeds) and benign PPL to 6.54.

cs.CR

新论文 · v1arXiv:2611.19350

When Does Label-Free Deconfounding of Document Embeddings Work?

Anna Kiepura, Yingqiang Gao, Elliott Ash, Yu Fan

原文摘要

Document embeddings encode spurious attributes that can distort semantic comparisons; yet supervised erasure requires metadata that is often unavailable. We systematically investigate label-free deconfounding across corpora and embedding models. Our results show that label-free methods improve semantic utility over unprocessed embeddings, while simple geometric approaches recover much of the supervised benefit, sometimes matching metadata-supervised erasure in linear confounder recoverability and semantic utility. We identify two distinct causes of ineffective erasure: poor proxy-confounder alignment and insufficient intervention coverage. Crucially, confounder recoverability and semantic utility are distinct dimensions of deconfounding success: utility can improve despite residual confounder information, while stronger suppression does not necessarily improve utility. Moreover, reduced linear recoverability can coexist with substantial nonlinear recoverability. Together, these findings distinguish confounder discovery from intervention effectiveness and highlight the importance of evaluating nuisance suppression alongside semantic preservation.

cs.CL

新论文 · v1arXiv:2611.17915

SWIFT: Sample-Wise Instance Fine-Tuning of Time-Series Foundation Models for Forecasting

Yang Chen, Feifan Tian, Cong Wang, Cheng Feng, Wenbin Li, Yang Gao

原文摘要

While current few-shot adaptation strategies for time series forecasting foundation models are predominantly dataset-centric, the necessity of obtaining an additional dataset for the finetuning process poses a significant hurdle for adoption. To address this constraint, we propose Sample-Wise Instance Fine-Tuning (SWIFT), a novel forecasting adaptation framework that achieves fine-grained specialization by leveraging structural priors from the target instance's own history. This shift allows the fine-tuning process to be entirely transparent to the end-user, which significantly enhances the ease of adoption. To implement this framework, SWIFT synergizes Self-Retrieval for historical mining and Frequency-Aware Loss for spectral consistency, while employing Deviation-Aware Dynamic Fusion to balance adaptation with general knowledge preservation. Extensive experiments across four TSFM backbones and three long-history benchmark subsets validate SWIFT, yielding average MASE and WQL reductions of 6.41% and 4.83% across 24 combinations of four backbones and six Chronos Benchmark II datasets. To the best of our knowledge, SWIFT is the first framework to investigate sample-wise instance fine-tuning of time-series foundation models, specializing the model for an individual prediction window using only the corresponding forecasting sample's own observed history.

cs.AI

新论文 · v1arXiv:2611.17728

A Spectral Framework for Universal Proxy Regularization in Model Merging

Shangge Liu, Danyue Qian, Qi Fan, Yinghuan Shi, Wenbin Li, Lei Wang, Yang Gao, Dacheng Tao

原文摘要

Model merging constructs multi-task models without expensive joint training but often suffers from parameter interference. Some solutions mitigate this by constraining finetuning using data from the other tasks to be merged, violating the core independence and privacy assumption of model merging. Recent work offers preliminary empirical evidence that a universal, task-agnostic proxy distribution can substitute for such data, but leaves open when and why it suffices. To this end, we propose a \textit{Spectral Framework for Universal Proxy Regularization}, which characterizes the relationship between a proxy distribution and the distribution of other tasks to be merged through spectral coverage ($α$) and leakage ($β$). Based on the framework, we derive a unified parameter interference bound: smaller $(α, β)$ provably yields lower parameter interference through better spectral alignment. To exploit such a proxy, we augment the training objective with a proxy regularizer. Evaluating this regularizer during fine-tuning requires a concrete mechanism. We study three instantiations, including (J) an exact evaluation via Jacobian-vector products, (F) a frozen curvature approximation using the Fisher information matrix, and (O) an output-drift formulation that constrains the model's functional deviation on the proxy directly. We prove that all three coincide under the NTK linearization, where existing analyses of such regularizers stop, and go further to characterize standard non-linear fine-tuning, bounding how far each instantiation departs from its linearized reference and obtain a three-way hierarchy, zero for (J), proxy-independent for (F), and proxy-quality-scaling for (O). Empirical validation on the commonly used vision-language model CLIP confirms these theoretical predictions.

cs.AI

新论文 · v1arXiv:2611.15013

Protecting Sensitive Knowledge Under LLM Unlearning: The Proxy Threat and a Superposition-Aware Defense

Jiayang Li, Tianqing Zhu, Longxiang Gao, Huajie Chen, Dayong Ye, Shang Wang, Huan Huo, Wanlei Zhou

原文摘要

Machine unlearning lets the owner of large language models (LLMs) remove specific entities from a trained model without retraining from scratch. Existing machine unlearning methods focus on removing the requested entities while maintaining utility. However, we observe that the representations of distinct entities in an LLM may overlap in shared parameter subspaces, a phenomenon known as knowledge superposition. Removing one entity may then affect other entities. This exposes a new security threat: legitimate unlearning requests can be exploited to remove sensitive entities that were never requested. This threat is unlike data poisoning or backdoor attacks. We define the attack as the Proxy Unlearning Attack (PUA), which selects removable proxy entities superposed with a sensitive target. The attack can even occur unintentionally, when benign users accidentally remove entities superposed with a sensitive one. In response, we propose Superposition-Aware Unlearning Protection (SAUP), a general defense framework that can be applied to unlearning requests to ensure a correct removal. SAUP performs in two stages. Before unlearning, it uses superposition detection to estimate which sensitive entities the request puts at risk. After unlearning, it decouples the shared representations and restores the affected entities, while keeping the requested entities suppressed. Our findings reveal that legitimate unlearning requests can be maliciously exploited through knowledge superposition. When dealing with unlearning requests, LLM providers need representation-level detection and recovery mechanisms to mitigate this threat and protect sensitive knowledge.

cs.CR

版本更新 · v2arXiv:2602.04851

Pose Distance Fields for Humanoid Robots

Yi Gu, Yukang Gao, Yangchen Zhou, Xingyu Chen, Yixiao Feng, Mingle Zhao, Yunyang Mo, Zhaorui Wang 等 10 位作者

原文摘要

Pose and motion priors play a crucial role in humanoid robotics. Although such priors have been widely studied in human motion recovery (HMR) domain with a range of models, their adoption for humanoid robots remains limited, largely due to the scarcity of high-quality humanoid motion data. In this work, we introduce Pose Distance Fields for Humanoid Robots (PDF-HR), a lightweight prior that represents the robot pose distribution as a continuous and differentiable manifold. Given an arbitrary pose, PDF-HR predicts its distance to a large corpus of retargeted robot poses, yielding a smooth measure of pose plausibility that is well suited for optimization and control. PDF-HR can be integrated as a reward shaping term, a regularizer, or a standalone plausibility scorer across diverse pipelines. We evaluate PDF-HR on various humanoid tasks, including single-trajectory motion tracking, general motion tracking, style-based motion mimicry, and general motion retargeting. Experiments show that this plug-and-play prior consistently and substantially strengthens strong baselines. Code and models are publicly available at https://gaoyukang33.github.io/PDF-HR.

cs.RO · cs.CV

新论文 · v1arXiv:2611.12856

WorkflowArena:面向基于大语言模型的多智能体系统的可执行工作流编排基准

WorkflowArena: Benchmarking Executable Workflow Orchestration in LLM-Based Multi-Agent Systems

Zheng Jiang, Fang Niu, Bo Peng, Houde Qian, Wenyi Li, Deyao Hong, Mingju Gao, Yizhe Chi 等 13 位作者

AI 速读 基于摘要

提出 WorkflowArena 基准,用 145 个长周期任务评估大模型多智能体系统自主编写并执行工作流编排的能力。

cs.CL · cs.SE

查看原文摘要

Large language model agents are increasingly deployed as multi-agent systems rather than as a standalone model operating within a single context. Recent agent harnesses allow the model to instantiate worker agents and to author the executable workflow that coordinates them, shifting the design of system structure from developers to the model at run time. The quality of this organization therefore constitutes a distinct capability, yet existing benchmarks either measure end-to-end task completion or evaluate coordination under structures specified in advance. We formalize this capability as workflow orchestration and introduce WorkflowArena, a benchmark of 145 long-horizon tasks spanning 8 professional domains, from software engineering and formal mathematics to legal analysis and 3D production. Each task specifies an initial workspace and an integrated deliverable composed of interdependent components, which is assessed by a hidden verifier. Agents determine autonomously whether and how to delegate, and all delegation is recorded, enabling joint evaluation of deliverable quality and the organization of the underlying work. Experiments with nine models show that no single configuration performs best across all domains and that systems exhibit distinct delegation strategies: at high effort, Claude models in Claude Code invoke workflows less frequently but with larger worker groups, whereas GPT models in Codex delegate more frequently in smaller units. When the executing model is held fixed on a challenging subset of tasks, workflows authored by Claude Opus 5.5 substantially improve weaker executors, indicating that workflow organization is itself a measurable determinant of task performance.

新论文 · v1arXiv:2611.11905

On the Global Convergence of Stochastic Adam for Physics-Informed Neural Networks

Yihang Gao

原文摘要

The Adam optimizer is widely used to train neural networks with highly nonlinear and nonconvex objective functions. Remarkably, Adam often drives the empirical training loss nearly to zero. This phenomenon has been partially explained by neural tangent kernel (NTK) analyses of gradient descent (GD) and stochastic gradient descent (SGD). However, analyzing Adam presents additional challenges. Its first-moment momentum couples stochastic gradients across iterations, while its adaptive second-moment preconditioner depends on the same stochastic gradient used in the update. In this paper, we establish linear convergence guarantees for Adam in training sufficiently wide two-layer neural networks with either full-batch or minibatch gradients. For full-batch Adam, we construct a Lyapunov function to control the effect of historical momentum. For minibatch Adam, our analysis builds on an exact virtual-iterate representation, a conditional analysis of the dependence between stochastic gradients and adaptive preconditioners, and uniform concentration bounds for the second-moment estimates. We prove geometric decay of the empirical loss for full-batch Adam and geometric decay in restricted expectation on a high-probability event for minibatch Adam. Beyond regression, we extend our convergence analysis to physics-informed neural networks (PINNs) for solving the Poisson equation, where differential operators and separate sampling of interior and boundary points introduce additional challenges. We establish analogous global convergence guarantees for both deterministic and stochastic Adam under explicit conditions on network width and Adam's parameters, characterizing the effects of minibatch size on convergence. These results help bridge the gap between Adam's empirical success and its theoretical understanding, providing sharper convergence guarantees for PINNs than existing analyses.

cs.LG · math.NA · math.OC

版本更新 · v3arXiv:2605.14260

On the Burden of Achieving Fairness in Conformal Prediction

Ziang Gao, Pengqi Liu, Archer Yi Yang, Mouloud Belbahri, Jesse C. Cresswell, Masoud Asgharian

原文摘要

Conformal prediction is often calibrated with a single pooled threshold, but this can hide cross-group heterogeneity in score distributions and distort group-wise coverage. We study this phenomenon through the population score distributions underlying split conformal calibration. First, we derive a conservation law and lower bound showing that pooled calibration incurs irreducible group-wise coverage distortion at a scale set by cross-group quantile heterogeneity. Second, we demonstrate that the two leading fairness definitions for conformal prediction, Equalized Coverage and Equalized Set Size, are fundamentally in tension. Third, we quantify the cost of moving between policies which treat groups separately or pool them. Experiments on synthetic and real data confirm the same bidirectional trade-off after finite-sample calibration. Our results show that, for the policy families studied here, calibration choice does not remove cross-group heterogeneity; it determines whether the resulting distortion appears in the coverage or size dimension, providing a principled lens for analyzing fairness-oriented calibration choices in practice.

stat.ML · cs.LG

版本更新 · v5arXiv:2607.11976

LiteTopK: Exploiting the Curse of Dimensionality for a Fused Indexer-TopK Kernel in Long-Context Sparse Attention

Ziqi Yin, Jianyang Gao, Peiqi Yin, Jiangneng Li, Gao Cong

原文摘要

Indexer-TopK, which computes scores and selects the top-$k$ candidates, is a key primitive in long-context sparse attention for large language models, vector retrieval for recommendation systems, and vector databases. However, existing GPU-based Indexer-TopK kernels like DeepSeek Sparse Attention (DSA) typically separate scoring from selection and materialize the full score matrix in high-bandwidth memory. This strategy is inefficient for large score matrices, such as those encountered in long-context sparse attention, due to excessive global-memory traffic, costly synchronization, and prohibitive memory overhead. In this study, motivated by the curse of dimensionality, we observe that sparse attention scores exhibit a score concentration phenomenon, with scores falling within a narrow range. Based on this observation, we propose LiteTopK, a sample-filter-select framework that estimates query-data score ranges from a small subset of the data. It uses these estimates to filter scores as they are computed, materializes only the candidates that pass the filter, and selects from the surviving candidates. In particular, we develop two implementations of LiteTopK. The first uses a conservative threshold estimate, while the second pairs a predicted threshold with a fallback strategy. Building on LiteTopK, we further propose LiteDSA, which reuses overlapping key/value entries across neighboring queries to accelerate attention computation. We integrate LiteTopK into vLLM, SGLang, and Megatron-LM. Our evaluation on eight B200 GPUs demonstrates prefill speedups of up to 1.49$\times$ across GLM-5.2, DeepSeek-V4-Flash, and Hy-4-Preview at a context length of 1M tokens, without compromising model capabilities.

cs.LG

版本更新 · v3arXiv:2610.23002

Robotic Spatiotemporal Editing: A Unified Interface for Controllable Action Generation via Masked Action Reconstruction

Yanbang Li, Bu He, Yi Tian, Yu-Chiao Huang, Chenyu Zhang, Ruoqu Chen, Yuanzhe Liu, Yuxiang Gao 等 9 位作者

原文摘要

Generative policies can synthesize robot behaviors, yet controlling and editing them through diverse spatiotemporal references remains challenging. In the real world, human intent requires robot behaviors that satisfy constraints on where, when, and how actions are executed. Our key insight is that spatiotemporal control references, such as keypoints, flows, orientations, and keyposes, can be viewed as partial specifications of the same temporal Cartesian action space. We introduce the MAP (Masked Action Space) interface to unify these action references and formulate controllable manipulation as an action inpainting problem. Our Inpainting Policy (IP) reconstructs missing action dimensions of the end-effector, combining progress-guided alignment with multi-mask training supporting unseen action-constraint combinations. We further adopt a training-free RePaint variant that iteratively inpaints the action during denoising to reduce control error. On ManiSkill and LIBERO-10, our policy achieves average task success rates of 92.67% and 64.67%, respectively.Real-world experiments demonstrate spatiotemporal control with average positional and rotational errors of 44.3 mm and 7°, respectively, showing that our policy can use diverse references to guide executable robot behaviors.

cs.RO

新论文 · v1arXiv:2611.10967

BAT-WAM: Bridging Action and Touch in World-Action Models for Dexterous Hands

Dong Fang, Feixuan Huang, Yuanxin Zhong, Yuxiang Gao, Qi Song, Yunlong Wang, Xiaosong Jia, Yu-Gang Jiang

原文摘要

Dexterous manipulation with world action models (WAMs) relies on hand configuration and contact information, yet joint-state and whole-hand tactile observations may be partially or entirely missing. RGB images provide scene context and hand-object geometry but are insufficient to reliably recover these missing modalities. We further observe a correlation between hand state and tactile information. Motivated by this observation, we introduce BAT-WAM, a framework that bridges visual, joint-state, and tactile representations for WAMs. BAT-WAM learns separate codebooks and decoders for the tactile and action streams. An RGB-conditioned continuous latent bridge connects the two streams for bidirectional learning, while visual information serves as continuous context. Under the HTBench evaluation protocol, BAT-WAM reduces the root mean squared error (RMSE) of tactile reconstruction by 34.73% under fully masked conditions and 38.84% under partially masked conditions, relative to an image-conditioned HandTouch checkpoint. For future tactile prediction conditioned on target-time RGB and historical actions, it reduces mean RMSE by 85.56% relative to OpenWAM with HandTouch. Real-robot experiments further show that BAT-WAM consistently outperforms related WAM baselines in dexterous manipulation. Generalization experiments on unseen objects and scenes also demonstrate substantial improvements over baseline models. These findings support joint modeling of visual context, hand state, and touch as a representation approach for multimodal WAMs.

cs.RO

版本更新 · v2arXiv:2611.03601

DiabAgent: Toward Agentic Physiological Control

Chengxi Li, Shaoxuan Li, Deyao Hong, Qingle Liu, Yi Huang, Puheng Bian, Minzhi Xu, Lanyue Zhang 等 16 位作者

原文摘要

As pretraining scales, medical language models have advanced rapidly in clinical question answering. Their ability to achieve effective physiological control through sustained intervention still lacks systematic quantitative evaluation. Type 1 diabetes provides a concrete setting: impaired insulin secretion creates dependence on exogenous insulin, with dosing continually adjusted using information about meals, exercise, illness, and glucose measurements. We introduce DiabAgent, a research suite for agentic physiological control comprising the Agentic Physiological Control (APC) framework, the DiabetSim simulator, and the SweetBench benchmark. APC controls glucose through medical-device tools, using ongoing observations to infer individual states, accumulate personal experience, and choose treatment actions. We build DiabetSim to simulate different patients, events, and device conditions for scalable virtual testing and training. Using DiabetSim, we construct SweetBench by instantiating 32 scenario designs as 212 glucose-control tasks. With Claude Opus 5.5 and a fixed decision interval of one hour, APC achieves a COGI of 73.5, compared with 71.2 for the open-source closed-loop baseline OpenAPS. Ablations further show that individual adaptation and contextual information improve glucose control. Combining high-frequency rule-based control with lower-frequency agent decisions shows potential for balancing control performance and computational cost. Our work provides a framework, simulator, and evaluation foundation for agents to participate in human physiological control.

cs.AI · cs.CL

版本更新 · v2arXiv:2609.19134

ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments

Hejia Geng, Zesen Huang, Haoyang Li, Wenbin Li, Koutian Wu, Zihan Zhou, Yuanbo Pang, Weihao Liu 等 63 位作者

原文摘要

Scientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, implicit domain conventions, and specialized correctness criteria make this knowledge difficult to convert into reliable learning experience-a challenge we call the scientific experience bottleneck. We introduce ScienceIDE, infrastructure for turning the world's scientific code into programmable environments for scientific agents. Guided by expert-defined scientific cases and acceptance criteria, agents transform repositories into executable environments that support task generation, execution, and scientific verification. These environments provide a shared foundation for supervised fine-tuning, reinforcement learning, and evaluation. Using verified interaction trajectories, we train PhAI-IDE-72B/9B/4B. The model family shows gains in held-out scientific-code repair and across selected general-purpose benchmarks in code, reasoning, and knowledge, providing evidence of positive transfer from scientific experience to broader capabilities. ScienceIDE lays the foundation for an integrated workspace for agent learning and scientific practice, making humanity's scientific software a shared substrate for developing scientific intelligence. Code: https://github.com/aitofound/ScienceIDE

cs.CL · cs.CY

版本更新 · v3arXiv:2602.06643

HuMI: Humanoid Whole-Body Manipulation from Humanoid-Free Demonstrations

Ruiqian Nai, Boyuan Zheng, Junming Zhao, Haodong Zhu, Sicong Dai, Zunhao Chen, Yihang Hu, Yingdong Hu 等 11 位作者

原文摘要

Current approaches for humanoid whole-body manipulation, primarily relying on teleoperation or visual sim-to-real reinforcement learning, are hindered by hardware logistics and complex reward engineering. Consequently, demonstrated autonomous skills remain limited and are typically restricted to controlled environments. In this paper, we present the Humanoid Manipulation Interface (HuMI), a portable and efficient framework for learning diverse whole-body manipulation tasks across various environments. HuMI enables humanoid-free data collection by capturing rich whole-body motion using portable hardware. This data drives a hierarchical learning pipeline that translates human motions into dexterous and feasible humanoid skills. Extensive experiments across five whole-body tasks--including kneeling, squatting, tossing, walking, and bimanual manipulation--demonstrate that HuMI achieves a 3x increase in data collection efficiency compared to teleoperation and attains a 70% success rate in unseen environments.

cs.RO · cs.AI · cs.LG

新论文 · v1arXiv:2611.09017

Frontier3DBench: Benchmarking Coding Agents for 3D Creation across Objects, Scenes, and Interactive Open Worlds

Xinjie Lin, Qingle Liu, Wenyi Li, John Lam, Xiaoqiu Wang, Deyao Hong, Mingju Gao, Zheng Jiang 等 13 位作者

原文摘要

Coding agents combine visual understanding with program execution, offering a new route to automated 3D creation. By translating images and instructions into editable programs, they bring reconstruction, generation, and editing into a shared workflow. Yet existing benchmarks fragment evaluation across task scopes and protocols, leaving their capabilities across construction levels unclear. We propose Frontier3DBench, a unified benchmark of 145 tasks across 9 categories spanning object, scene, and interactive open-world creation, with joint evaluation of appearance, structure, and behavior. Experiments with 11 model-harness configurations reveal strengths that vary by level: GPT-6.1-sol leads the object and scene levels and achieves the highest overall score of 66.02/100, whereas GPT-6-astra leads interactive world construction. However, all configurations score lower on reference reconstruction than on text-driven generation in object and scene tasks. Producing plausible content thus remains easier than reproducing specified geometry, layouts, and visual details. Moreover, generated worlds average 83.08/100 on local mechanisms but only 20.82 on motion clearance and 30.26 on stateful workflows: individual mechanisms often function, whereas complete activities that require coordinated actions and persistent state remain difficult. These results identify reference fidelity and coordinated activity completion as central bottlenecks for coding agents in 3D creation.

cs.CV