RHEA：面向鲁棒多模态属性图聚类的可靠性协调重构与分配
该论文以 arXiv:2608.00621v1 形式发布，公告类型为 new。研究聚焦多模态属性图（MAG），即节点在关系结构之上携带文本、图像等异构属性的图数据。原摘要指出，这类图已成为无标签实体分组任务的重要基础，包括社区发现和产品相关场景；但所提供的摘要内容在此处截断，未披露具体方法细节、实验数据或结论。论文标题表明，RHEA 方法旨在通过可靠性协调的重构与分配提升多模态属性图聚类的鲁棒性。
独立 AI 资讯库
AI 前沿动态、原文归档与媒体镜像
从自动抓取数据库实时读取，适合快速浏览、检索和后续扩展成完整资讯站。
12005资讯
11873正文归档
75817媒体
Agentic Graph Token 推理
该论文以图结构中的关系数据分析为背景，指出图广泛应用于科学与工业领域，例如引文网络和商品共购图。许多此类图的节点携带丰富文本信息，因此研究者正越来越多地将大语言模型（LLM）用于图分析。原文提供的信息未包含完整摘要、方法细节、实验数据或具体结论。
Unleashing the Potential of Large Language Models: A Blueprint for Real-Time, Enterprise-Ready Deployments
arXiv:2608.00419v1 Announce Type: new Abstract: Large language models deployed in real-time, regulated settings face knowledge staleness, catastrophic forgetting, hallucination, and weak feedback loops. We present a unified, pattern-driven LLMOps architecture integrating real-time data ingestion, c
Agentic Bayesian Optimization through Surrogate-Augmented Autoresearch
arXiv:2608.00316v1 Announce Type: new Abstract: Bayesian optimization (BO) has become the standard tool for sample-efficient optimization and owes its efficiency to uncertainty-aware search driven by generic statistical priors. Richer domain priors can improve BO in principle, but encoding them thr
Abstention as an Action Can Kill Both the Reward Gradient and the KL Anchor: Collapse Law and Repair for Error-Penalized Reinforcement Learning
arXiv:2608.00301v1 Announce Type: new Abstract: Error-penalized scoring rules ($+1$ for a correct answer, $-\lambda$ for a wrong one, $0$ for abstaining) are increasingly prescribed against hallucination: a rational agent facing such a rule answers exactly when its correctness probability exceeds C
AutoCause: A Python framework that automates expert decisions in environmental time-series causal discovery
arXiv:2608.00198v1 Announce Type: new Abstract: Environmental time-series causal discovery requires expert decisions about method choice, conditional-independence tests, lag horizons, sample-size adequacy, multiple-testing control, and evidence interpretation. Applied inconsistently across datasets
Inference-Time Policy Alignment for Fair Reinforcement Learning
arXiv:2608.00175v1 Announce Type: new Abstract: Deep reinforcement learning (RL) agents achieve strong performance by optimizing scalar reward functions. However, once deployed, the policies of these RL agents are often rigid and costly to adapt to new performance criteria. For instance, an agent t
Response Magnitude as a Dominant Signal for Held-Out CRISPRi Perturbation Effect Prediction
arXiv:2608.00152v1 Announce Type: new Abstract: Predicting the magnitude of a CRISPRi perturbation's transcriptomic effect on held-out target genes is an important open problem in single-cell biology. Recent work has documented that simple baselines often match or exceed deep perturbation predictor
MetaRoute-Bench: Evaluating Meta-Decision Policies for Agentic Workflow Routing
arXiv:2608.00107v1 Announce Type: new Abstract: Agentic systems must repeatedly decide whether to answer directly, decompose a task, invoke a tool, execute code, delegate to a specialist, verify an intermediate result, or recover from failure. These meta-decisions affect not only task success but a
Learning Compositional Meta-Routing for Agentic Workflows: An Executable Benchmark
arXiv:2608.00106v1 Announce Type: new Abstract: Agentic systems must decide not only what answer to produce, but which reasoning and execution operations should precede it. A controller may answer directly, decompose a request, retrieve evidence, execute code, delegate to a specialist, or verify an
SERL-SQL: Selective Hindsight Distillation for Text-to-SQL Reinforcement Agentic Learning
arXiv:2608.00485v1 Announce Type: new Abstract: Recent Text-to-SQL systems increasingly rely on multi-turn interaction, execution feedback, and reinforcement learning. However, most existing methods use execution correctness only as a trajectory-level reward, which provides limited guidance for ide
Deep Research Pretraining via Predictive Navigation
arXiv:2608.00432v1 Announce Type: new Abstract: Deep research agents are often trained on expensive, environment-grounded tool-use trajectories that require repeated retrieval, document inspection, and report evaluation. We introduce Deep Research Pretraining (DRP), an offline framework that derive
CurveShift: Is Agent Progress Scalar? Separating Level from Shape
arXiv:2608.00355v1 Announce Type: new Abstract: Progress in large language models is often summarized using a single scalar measure, such as a time horizon, a latent ability estimate, or an aggregate benchmark score. These summaries capture the overall performance, but they do not test whether prog
A Few Neurons Reveal When LLMs Misuse Tools: Sparse Detection and Selective Steering for Reliable Tool Use
arXiv:2608.00218v1 Announce Type: new Abstract: Agentic LLMs exhibit three consequential tool-use failures: invalid arguments (validity), unnecessary calls (over-calling), and omitted calls when tools are needed (missing). We find that a small, failure-specific set of MLP neurons could distinguish
XL-DocBench: Benchmarking Evidence-Grounded Extra-Long Document Understanding
arXiv:2608.00036v1 Announce Type: new Abstract: Real-world document tasks often ask professionals to answer questions from annual reports, regulations, clinical guidelines, and technical manuals that span hundreds or thousands of pages. Some questions also require comparing related reports. Reliabl
SLMs as Multi-Agent Routers: A Progressive SFT and Reinforcement Learning Approach
arXiv:2608.00030v1 Announce Type: new Abstract: Specialised retrieval agents typically surface higher quality results than general-purpose search, but selecting the optimal agent for a given query remains an open problem. Current approaches route queries based on inferred topic or intent, however i
Role Steering of Language Models for Social Simulations
arXiv:2608.00023v1 Announce Type: new Abstract: Social simulations built from language-model agents need role-conditioned behavior that can be checked before agents are placed into a simulated population. We introduce an activation-steering screening workflow for role-conditioned agents: define a r
What Transfers from Text to Vision? Capability Scaling Laws and Transfer Dynamics for VLMs
arXiv:2608.00013v1 Announce Type: new Abstract: Choosing the right large language model (LLM) backbone is the most consequential decision when building a vision-language model (VLM), yet it remains fundamentally unprincipled: compute-based scaling laws fail to generalize across model families, and
Obshazard-bench: Benchmarking Multimodal Foundation Models for Real-Time Disaster Intelligence from Raw Earth Observation Streams
arXiv:2608.00012v1 Announce Type: new Abstract: Multimodal Large Language Models (MLLMs) are increasingly used to interpret Earth observation data, yet their capability to support real-world disaster emergency response remains insufficiently evaluated. Existing remote sensing benchmarks largely rel
AgentMemBench: A Systematic Benchmark for Evaluating Long-Term Memory Management Strategies in Conversational AI Agents
arXiv:2608.00009v1 Announce Type: new Abstract: Long-term memory remains a critical bottleneck for conversational AI agents, whose finite context windows cannot support coherent recall across thousands of turns. We present AgentMemBench, a unified, reproducible benchmark evaluating five memory mana
MemoryForge: Synthesize Lifelong Memory for Human-Like LLM Agents
arXiv:2608.00007v1 Announce Type: new Abstract: Equipping Large Language Models (LLMs) with human-like personas is crucial for agentic applications, such as role-play and user simulation. Traditional prompt-based methods rely on descriptive conditioning by injecting static textual profiles, which o
RubricReviewer: From Direct Critique to Objective and Comprehensive Rubric-Driven Peer Review
arXiv:2608.00005v1 Announce Type: new Abstract: Peer review at major venues is under unprecedented submission pressure, motivating the use of large language models (LLMs) as review assistants. Existing LLM-based reviewers, however, face two structural limitations. First, they map manuscripts direct
Cost-Effective Automated Judging of Natural-Language Mathematical Proofs
arXiv:2608.00004v1 Announce Type: new Abstract: Grading natural-language mathematical proofs is a recurring cost in evaluating math-reasoning systems, and frontier LLM judges are expensive. We ask whether cheap open-weight models can serve as reliable judges given a candidate proof, a ground-truth
从空间语义到时间上下文：利用注视轨迹进行弱监督医学图像分割
该论文提出利用眼动追踪记录的注视轨迹，为弱监督医学图像分割提供监督信号。医学图像分割通常依赖耗时且成本高昂的像素级标注，而眼动数据可较低成本地融入临床工作流程。研究关注从注视的空间语义进一步挖掘时间上下文，有望为开发者构建更低标注成本的医学影像 AI 系统提供思路。该论文为 arXiv:2607.26542v1，新发布版本。
CineWeaver：无需训练、可参考控制的多镜头长视频生成，用于电影化叙事
CineWeaver 聚焦电影化长视频生成，旨在解决文本生成视频扩散模型在多镜头、角色与场景精细控制以及长时序生成方面的联合挑战。该方法强调无需训练，并支持基于参考信息进行可控生成，以服务更连贯的电影叙事创作。对于开发者而言，这类方法有望降低定制化视频生成的训练成本，并提升多镜头内容中人物和场景一致性的可控性。该论文以 arXiv:2607.26529v1 形式发布。
EgoSafe：面向视觉安全理解的第一人称移动拍摄基准
研究团队提出 EgoSafe-Bench，用于评估大视觉语言模型（LVLM）在第一人称安全场景中的取证式推理能力。该基准包含 12,000 个评测样本，由 3,000 段视频及其问答链组成，并采用分层推理评估（HRE）协议，要求模型完成从特征锚定、盲区推断到意图推理的完整推理过程。对 Qwen3-VL、Gemini、VideoLLaMA 3 等模型的评测显示，模型的描述与感知能力可以较强，但在因果推理和逻辑闭环方面仍存在明显脆弱性。该工作为开发更具逻辑鲁棒性的视频理解系统提供了具有挑战性的数据集和系统化评估框架。
FAS-R1：用于推理型人脸活体检测的统一多任务 MLLM
FAS-R1 提出了一种面向人脸活体检测（FAS）的统一多任务多模态大语言模型（MLLM）。该研究指出，FAS 系统日益需要同时输出真人/攻击判定、攻击语义信息，以及可供人工核验的图像依据。现有判别式 FAS 模型大多仍以标签分类为中心，而该工作旨在将推理与可解释的证据呈现纳入统一框架。该方向有望帮助开发者构建更具可审查性和语义理解能力的反欺诈系统。
统一多模态模型是否在同一空间中思考？通过跨分支引导进行观察
这篇 arXiv 论文探讨统一多模态模型（UMM）中的理解与生成能力，是否共享统一且可迁移的语义空间。作者指出，尽管 UMM 试图在单一架构中整合理解和生成，但两类能力之间的内部表征关系仍不明确。论文以跨分支引导（Cross-Branch Steering）为切入点，研究这一基础问题，对多模态模型的架构设计、能力评估和可控性研究具有意义。
看见还是知道？多模态大语言模型中的视觉上下文敏感性
该论文以“看见还是知道？”为题，研究多模态大语言模型（MLLM）的视觉上下文敏感性。摘要指出，MLLM 通过将视觉输入与预训练语言模型的丰富先验相结合，已取得较强性能。与此同时，模型在以视觉为核心的任务上仍可能失效，尤其是在视觉证据与预训练先验发生冲突时。这一问题对开发者意味着，评估和改进多模态模型时需要重点关注模型是否真正依据图像证据作答，而非过度依赖语言先验。
BG-REAL：面向背景篡改检测与定位的公共真实数据锚定基准
BG-REAL 是一个聚焦图像背景篡改检测与定位的公共基准，旨在补足现有评测偏重前景对象复制、拼接和通用合成编辑的不足。该基准基于 Open Images V7 实例分割来源构建，包含 1,200 个源图组的 7,000 个处理样本，其中包括 6,000 个公共数据锚定样本和 1,000 个合成对照样本。基准覆盖六类编辑方式、配对真实对照、掩码与泄漏质量检查，并已完成 TruFor、MVSS-Net 和 HiFi-Net 三个外部基线测试。实验显示，在固定验证集阈值下，多个基线会将重新编码的真实图像误判为篡改图像，假阳性率介于 0.57 至 1.00，表明重编码伪影是跨模型普遍存在的捷径风险。