AI 论文与研究

Loneliness is a critical public health issue among older adults, linked to higher risks of depression, cognitive decline, and mortality. Scalable, objective methods for its detection remain limited, particularly in natural conversational c…

Large language models (LLMs) are increasingly deployed as economic agents, yet there is little evidence whether LLM agents are suited for participating in market mechanisms designed for humans, and whether these mechanisms deliver desired …

Volumetric video based on point cloud representations enables immersive virtual and augmented reality applications but introduces significant challenges for efficient and secure content delivery. Prior work proposed a selective coordinate …

Machine-learnt corrections can complement numerical weather prediction only if they adapt to the evolving model state while preserving dynamical consistency and numerical stability. To test this within a global forecasting model, we couple…

Drug-target interaction (DTI) prediction is an important task in AI-driven drug discovery. Although recent biochemical representation learning methods have improved DTI prediction, their passive feature aggregation tends to favor dominant …

Modern large language models (LLMs) rely on reinforcement learning to build strong capabilities in individual domains, but integrating those capabilities into a single deployable model remains challenging. By routing each sample to the tea…

新加坡科技设计大学研究人员让 140 名 60 至 89 岁老年人与其社交聊天机器人 Ami Chat 互动，研究发表于《老龄化创新》期刊。结果显示，社会关系相关三个指标（拟人化、真实性、AI 社交互动强度）在使用意愿差异中额外解释了 27%；感觉更真实、互动更具回应性时使用意愿更高，但认为机器人更像真人的参与者意愿反而更低；自评健康状况较差者使用意愿更强。

Diagnosing collective anomalies from urban trajectories is increasingly important for traffic governance, as it reveals what happened, who was involved, and where and when the event occurred. Existing detectors efficiently produce scores o…

Graph construction is a critical but underexamined design choice in deep reinforcement learning for power grid control. We present a controlled experimental comparison of different graph representations, including physical topology, electr…

Image enhancement and restoration have become standard back-end operations on short-video and social media platforms to boost UGC visual experience. Yet these processes inevitably introduce visual anomalies--especially in faces, texts, and…

Large Language Models (LLMs) are increasingly used to predict the responses of human participants in survey panels. Towards that goal, persona prompting has recently emerged as a technique to inform and align large pretrained language mode…

Uncertain knowledge graphs (UKGs) extend knowledge graphs by assigning each triple a continuous confidence score. Since most possible triples lack observed confidences, recent methods rely on semi-supervised learning to generate pseudo-lab…

We study body-only, 12-class acted-emotion classification from skeleton motion under leave-performer-out (LPO) evaluation, a hard, underdetermined setting: chance is 8.3%, and a protocol-matched reproduced STGCN++ baseline reaches only 25.…

Test-Time Adaptation (TTA) has recently emerged as a promising strategy that allows the adaptation of pre-trained models to changing data distributions at deployment time, without access to any labels. To mitigate error accumulation, resea…

Text-to-image (T2I) models have achieved remarkable success at faithfully rendering specified objects and attributes, yet their ability to produce visual metaphors, images that convey abstract ideas by combining elements from two distinct …

Recommender-system experiments often rely on a single random training seed, assuming that run-to-run stochasticity has limited impact on evaluation conclusions. This assumption is risky, as a training seed may influence several algorithm-d…

Zero-shot graph anomaly detection seeks to deploy a detector trained on source graphs to unseen, unlabeled targets, yet domain shift can make source-derived notions of normality unreliable. We introduce RINSE (Robust Iterative Normality Se…

作者转发引用 @ao_qu18465 的文章，主题为 REEF，一个面向自我改进智能体（self-improving agents）的持续学习基础设施，原文链接 https://x.com/i/article/2094850055132024832。作者本人未补充正文内容。

论文提出神经网络向量表示隐式实现了 Tensor Product Representation 符号结构，作者用 DISCOVER 方法将多种网络（含 Gemma-3-27b、GPT-2-XL、GPT-OSS-20b、Pythia-12b、Qwen3-14b、OLMo-2-13B、Llama-3.1-8b 共 7 个 LLM）的表示替换为符号化闭式方程，模型行为基本不变。

Microsoft 论文《Sliding-window beats linear attention》发现，目标是降低推理内存时，免训练的滑动窗口注意力（SWA）优于多数改造的线性注意力方法，只需保留小窗口加前 4 个 sink token。

新论文《JIT-Agent： Scaling Harness Intelligence via Just-in-Time Harness Evolution》提出 JIT-Agent，可按需为具体任务生成 agent harness，动态决定记忆、规划、动作、工具和技能的组织方式。

Meta FAIR 发布论文 ADeptS-Bench（arxiv.org/abs/2608.26204），在移动和桌面双平台上用配对的 benign/恶意 GUI 任务与歧义指令评测 7 个模型。

PromptArmor 基于一个十人团队 30 天的 310,009 条 OTel 遥测事件分析 Claude Code 和 Cowork 的安全与成本。

论文提出 AgentFold，将折叠模型开发建模为对可执行代码变体的闭环智能体树搜索。智能体从紧凑的 ESMFold 衍生模型出发，提出架构修改、实现调试、运行训练，并记录成败经验决定分支算力分配。

美联储达拉斯联储发布报告，称 GenAI 暴露度较高的得州企业招聘岗位减少 8-9%。报告将 O*NET 任务映射到 Claude 实际使用情况来度量暴露度，并与 Lightcast 数百万条在线招聘数据关联。

arXiv 论文《Selective Forgetting： A Graph-Based Memory Framework for Long-Term LLM Agents》将对话轮次抽取为带类型节点和属性边，从两跳子图作答。

一篇论文提出将 LLM agent 的执行轨迹压缩为单个紧凑的有限状态机（FSM），用于预测下一步动作和识别失败运行。

Agent Zero Memory 提出溯源感知的长期记忆架构，把情景事件时间线、实体-事件知识图谱和精选的文档式记忆三条系统并行跑在同一历史上，检索经意图门、来源路由和三路并发 agentic 搜索完成。

因果基础模型（CFMs）将基础模型范式引入因果推断，通过预训练神经网络在新数据集上利用上下文学习直接估计平均处理效应等因果量，无需更新模型。本文为该新兴领域提供实用导论，先梳理因果推断与机器学习必要背景，再讨论CFMs，并附示例代码与Jupyter notebooks。

论文提出 EarlyEval，一个通过提前结果预测降低 LLM Agent 评测成本的轻量框架。它训练一对 LightGBM 成功与失败分类器，基于行为、文本和参考解特征，在达到校准置信阈值时提前终止 Agent 运行。

新框架 Conformal Relevance 通过上下文示例筛选与集成构建评分函数，在保持覆盖率的同时提升简洁性，且几乎无需人工干预。研究在七项 NLP 任务上验证了该框架，并从理论上给出互补性条件与集成改进的饱和界，刻画了集成多样性对最坏情况句子评分的影响。

异步同策略蒸馏（OPD）中教师节点 GPU 利用率平均仅 9.8%，59% 时间低于 5%。TGOPD 利用空闲算力生成教师探针，由验证器评估各提示词下教师可靠性，仅在通过门控时启用密集 token 级 OPD 信号，否则改用验证器驱动的 GRPO，将教师节点平均利用率提升至 78.9%。

论文提出 RoboTok，一个互联网规模数据引擎，输入人类操作视频即可从网络视频中检索与操作相关的人类示范，用于训练灵巧机器人策略。方法基于估计的演员中心参考系中的 3D 手部轨迹学习潜在运动空间，使操作行为的比较不受相机视角、场景外观和演员遮挡影响，并支持互联网规模视频的高效搜索与持续索引。在检索基准和下游策略评估中，RoboTok 检索出更相关的示范并提升了下游任务成功率。

Microsoft Research 发布 VibeVoice-ASR-Streaming 技术报告，提出基于 LLM 的端到端流式说话人归属 ASR，将音频块、0.5 秒 lookahead 和已有文本交错在单一自回归上下文中，无需独立 diarization 阶段即可随语音到达输出谁说了什么。

论文提出端到端后训练管线，用 22，000 道精选题目、合成推理轨迹、SFT 和 RL 训练 Nemotron-3-Nano-CC（30B-A3B），并以 SFT 训练 Nemotron-3-Ultra-CC（550B-A55B），同时引入迭代生成、评估和改进解法的 GenCorrect 测试时计算策略。

论文研究 LLM 与非语言智能体（如棋类引擎）的协作，发布包含六个国际象棋协作任务的 LLAMIA-Bench，并提出将子智能体连续表征直接投影到 LLM token 流的潜在状态内化方法。

论文提出 Repo-To-Skill 与技能驱动的智能体 DisCo，将 GitHub 仓库中的操作性知识蒸馏为可复用技能，分为任务无关和任务导向两种形式。任务无关蒸馏产出 AREX-Skill Library，从 1，000 个常用 ML 仓库蒸馏出 5，000+ 个经核验的技能，按 20 个领域、178 个能力族组织。

论文提出 Calibrator-Output Repair for Top-1 Decision Preservation（CORD），一种后拟合适配器，通过修复完整校准概率向量，确保 argmax 恢复原始 top-1 预测，实现零 TPCR。

PaperCompiler 是一个论文到代码生成框架，将论文证据编译为显式的仓库级实现规格，区分论文支持、推断、外部委托和未解决的信息，并编码非退化要求、文件归属、跨文件依赖和文件级约束。