论文精选 83
Type Diversity Enables Transformers to Generalise Compositionally
Compositional generalisation has been divided into lexical and structural generalisation. …
领域：cs.CL作者：Anssi Moisio、Mathias Creutz、Mikko Kurimo
📰 AI 研究简报 · 9月15日
今日头条：7名博士生仅用3个月从零训练7B大模型：代码+数据+训练日志全公开 —— 几百个Agent参与研发，AI开始造AI
共 60 条预计阅读 36 分钟Token 消耗 108,000
arXiv 最新论文
论文精选 85
SAS: Simple Attention Sparsification via End-to-End Optimization of Context Ranking· SAS：通过端到端优化实现简单注意力稀疏化
Post-training attention sparsification reduces the quadratic cumulative attention cost of …
领域：cs.CL作者：Zhiwei Li、Lei Zhu、Hao Gu
论文精选 60已解读
Rethinking Heterogeneous System Disaggregation for Subquadratic Attention· 重新思考异构系统解构以实现次二次注意力
Frontier language models are more aggressively using subquadratic attention to reduce the …
领域：cs.AI作者：Arya Tschand、Yaosheng Fu、Vikram Sharma Mailthody
论文精选 60已解读
A Hybrid LSTM-XGBoost Framework for Multi-Horizon Stock Return Prediction Across Diversified Equity Portfolios· 混合 LSTM-XGBoost 框架预测股票收益
Accurate prediction of equity returns remains a major challenge in computational finance d…
领域：cs.AI作者：Seif ElDein Mostafa、Yahia Ahmed、Farah Datwish
论文精选 82
CMA-OT: Hierarchical Expert Supervision for Dance-to-Music Generation· CMA-OT: 舞蹈到音乐生成的分层专家监督
Dance-to-music (D2M) generation aims to synthesize music that is rhythmically and stylisti…
领域：cs.AI作者：Jinting Wang、Chenxing Li、Dong Yu
论文精选 82
Continue, Adapt, or Yield: In-Turn Adaptation to Overlapping Speech in Full-Duplex Agents· 继续、适应或让步：全双工代理对话中的适应
Full-duplex evaluation often emphasizes whether an agent keeps speaking or stops. That bin…
领域：cs.CL作者：Yunqi Lu、Tyler Baumgartner、Nikhil Johri
论文精选 60已解读
Embodied-BenchForge: A Closed-Loop Agentic Workflow for Embodied Benchmark Construction· 具身化基准构建的闭环代理工作流
Agentic systems offer a promising way to automate embodied benchmark construction, but exi…
领域：cs.AI作者：Baoyang Jiang、Fengchun Zhang、Leyuan Wang
论文精选 60已解读
Autonomous Research for Open-Ended Problems: A Case Study on Telecom Ticket Retrieval· 开放式问题的自主研究：以电信工单检索为例
Recent breakthroughs in LLM-based systems and their abilities in problem solving and codin…
领域：cs.AI作者：Junghyun Min、Huseyin Uzunalioglu、Mohamed Trabelsi
论文精选 60已解读
Anchoring Clinical Events in Time: UID-Preserving Multimodal Reconstruction and Source-Grounded Adjudication· 临床事件时间锚定的多模态框架
Clinical timelines support treatment-window analysis and leakage-free modeling, but discha…
领域：cs.AI作者：Sayantan Kumar、Nicolas Grimaldi、Jack Cummins
论文精选 82
Expert-Space Exploration in MoE Reinforcement Learning· MoE强化学习中的专家空间探索
Reinforcement learning (RL) has become central to post-training of large language models. …
领域：cs.CL作者：Hongyi He、Zhenghao Lin、Xiao Liu
论文精选 60已解读
Diffusion Models and Concept Formation· 扩散模型和概念形成
Humans organize knowledge into a taxonomy of concepts with nested levels of abstraction an…
领域：cs.AI作者：Zekun Wang、Karthik Singaravadivelan、Christopher J. MacLellan
论文精选 82
Kraken: LLM-based Speech-to-Speech Translation via Low-bitrate VQ and Dual-path Source Conditioning· Kraken: 低比特率VQ和双路径源条件的LLM基础S2ST
Speech-to-speech translation (S2ST) has advanced significantly with speech LLMs, offering …
领域：cs.CL作者：Hayato Futami、Hassan Shahmohammadi、Tushar Dhyani
论文精选 60已解读
How Good Are Frontier Models at Physics? Expert Re-Grading Reveals Broken Evaluations and Near-Saturation of Leading Benchmarks· 前沿模型在物理学中的表现如何？
Low reported scores on leading physics benchmarks, including those featured in the Artific…
领域：cs.AI作者：Ali Ansari、Haoran Sun、Andy Zeyi Liu
论文精选 60已解读
EduFair-Bench: Evaluating Pedagogical Fairness of LLM Tutors Across Student Demographics· 教育公平性基准：评估大模型导师的教育公平性
Large language models (LLMs) are increasingly deployed as tutors, but it is unclear whethe…
领域：cs.AI作者：Jiaxu Zhao、Bahar Radmehr、Fares Fawzi
论文精选 60已解读
Tracing and Coordinating Cross-Layer Influence for Multimodal Model Merging
Multimodal model merging aims to consolidate task experts into a single model that retains…
领域：cs.AI作者：Pengyang Zhou、Xiaobin Tu、Zhengxi Liu
论文精选 60已解读
MedRoundsQA: A Persona and Difficulty Aware Evaluation for Multi-Turn Medical Consultations· MedRoundsQA：多轮医疗咨询的角色与难度感知评估
Medical benchmarks are dominated by single-turn, multiple-choice clinical cases that poorl…
领域：cs.AI作者：Youssef Mohamed、Ahmed Heakl、Qinrong Cui
论文精选 60已解读
Scaling Clinical Judgment to Evaluate Medical AI· 大规模评估医学AI的临床判断
Blinded physician evaluation has been considered by many to be the gold standard for asses…
领域：cs.AI作者：Thomas A. Buckley、Zahir Kanjee、Peter G. Brodeur
论文精选 82
K-Bench: A Benchmark for LLM Unlearning in Agentic Deployments· K-Bench: 代理部署下LLM遗忘能力的基准
Unlearning benchmarks such as TOFU and MUSE certify forgetting by reading the model's fina…
领域：cs.AI作者：Guangsheng Yu、Yanna Jiang、Qin Wang
论文精选 82
Interpreting the predictions of neural network classification based on a Taylor Coefficient Analysis (TCA)· 基于泰勒系数分析解释神经网络分类预测
We introduce a rigid and comprehensive taxonomy and paradigm for characterizing the influe…
领域：cs.AI作者：Markus Klute、Artur Monsch、Lars Sowa
论文精选 82
MPT: Missing Prototype Tracking via Barycentric Reconstruction in Vehicular Federated Learning· MPT: 车载联邦学习中的缺失原型追踪
Cross-vehicle federated learning enables vehicles to collaboratively improve perception mo…
领域：cs.AI作者：Hanju Jang、Gyeongmin Han、Sungmin Lee
论文精选 80
Unified Agentic Video Editing Across Levels of Complexity and Creativity· 跨复杂度和创意级别的统一代理视频编辑
Editing is a core component of video production, requiring creative planning and decisions…
领域：cs.AI作者：Surabhi S. Nath、Kim Ferres、Milan Petrović
论文精选 82
SCQ: Stabilizing Conservative Q-Learning with Sigmoid-Bounded Entropy· SCQ：使用Sigmoid限制的保守Q学习
Offline-to-online reinforcement learning reduces interaction cost for real-world robot lea…
领域：cs.AI作者：Xiefeng Wu、Shu Zhang、Zhaojie Chu
论文精选 83
Assisted Spatial Cognition Through Vision-Language Models· 通过视觉-语言模型辅助的空间认知
Multimodal AI, powered by Large Language Models (LLMs) and Vision-Language Models (VLMs), …
领域：cs.AI作者：H. Riaz、J. B. Fernandez、I. Mills
论文精选 82
What Drives Recovery in Agentic Text-to-Cypher? LAST-CQ: An LLM Agent Self-Refinement Framework· 代理Text-to-Cypher恢复机制研究
Agentic pipelines for structured-query generation are rapidly expanding, but it is unclear…
领域：cs.AI作者：Ioannis Prokopiou、Athanasios Aidinis、Panagiotis-Christos Kyrmpatsos
论文精选 82
Skill Issue: Lessons from Optimizing Repository SKILLs for Coding Agents
Coding agents increasingly read repository knowledge from SKILLs --- plain \texttt{.md} fi…
领域：cs.AI作者：Mykhailo Kozyrev、Andrei Kozyrev、Anton Podkopaev
GitHub 热门开源项目
本板块今日暂无内容
精选 AI 行业资讯
行业资讯精选 85
旗舰SoC进入融合计算时代，天玑9600 Pro首创AI原生架构！
就在这两天，联发科正式发布新一代旗舰移动平台天玑9600 Pro。作为天玑首款Pro旗舰，它率先采用台积电2nm制程，并以全新AI原生架构重构移动计算，带来双超大核CPU、重构双N…
行业资讯精选 82
Early Anthropic hire, former METR COO have found a way to rein in rogue AI agents· 初创公司控制失控AI代理
Their startup, Artificial Intelligence Underwriting Company (AIUC) has raised $40 million …
行业资讯精选 82
New insights from Google’s AI & Economy ATLAS· Google AI 经济洞察工具 ATLAS
We’ve translated ATLAS’s millions of global data points into an interactive, open-access e…
行业资讯精选 82
首个AIGC长片大赛！RunningHub单项大奖100万，科幻IP免费改编· 首个AIGC长片大赛
刘慈欣督战
行业资讯精选 85
手机替我跑了一整套流程！我就说了一句话，AI执行了100步· 一句话，YOYO全面接管我的生活
一句话，YOYO全面接管我的生活
行业资讯精选 82
Salesforce and Nvidia’s new reasoning model is everything the AI labs should fear· Salesforce 和 Nvidia 的新推理模型
Salesforce Koa is built on Nvidia's open-weight Nemotron model and is trained to do sales,…
行业资讯精选 82
无问芯穹联合清华、上交正式开源具身端侧推理引擎APXInf，Pi 0.5性能SOTA
卡位具身智能规模化落地“最后一公里”！
行业资讯精选 85
全球AI视频榜单第一梯队再添中国力量：智象发布首款物理规律导向视频模型
智象未来（HiDream.ai）正式发布首个原生全模态视频生成模型 HiDream-O1-Video-1.0
行业资讯精选 85
一张GPU跑10万原子！分子之心用AI把化学反应“拍”成了电影
打破分子模拟“不可能三角”
行业资讯精选 85
阶跃发布 StepAudio 3 ，多款语音模型登顶 Artificial Analysis 全球榜单
9月15日，阶跃发布新一代语音大模型 StepAudio 3 系列，一次性推出 StepAudio 3 Realtime、StepAudio 3 ASR、StepAudio 3 T…
行业资讯精选 82
英伟达开源 IMO 金牌配方：不仅是「人海战术」，1.5TB 显存做实 AI「 推恩令」？
<img data-src="https://static.leiphone.com/uploads/new/images/20260915/6aa911f97a99d.jpg" …
行业资讯精选 75
亚太唯一！腾讯云首次入选IDC MarketScape 全球托管边缘服务领导者类别
9 月 15 日，国际数据公司（IDC）发布《IDC MarketScape: Worldwide Managed Edge Services 2026 Vendor Assess…
行业资讯精选 82
Heart of the Matter: How a Major Children’s Hospital Uses Open Source NVIDIA AI for Cardiac Care· 儿童医院如何使用开源 NVIDIA AI 进行心脏护理
行业资讯精选 72
“阿宝”交新朋友：携手比亚迪“迪迪虾”，实现车机端一句话办事
“阿宝”又交新朋友！9月14日，“阿宝”亮相腾势N8L纯电上市发布会，并与比亚迪AI超级智能体“迪迪虾”达成跨端协同办事合作。腾势N8L车主直接对“迪迪虾”说一句话，由“迪迪虾”唤…
行业资讯精选 75
Xspark AI 丁文伯：触觉替代不了视觉，但机器人需要一套自己的“脊髓” ｜物理AI 50人
<section data-mpa-powered-by="yiban.io" style="margin-right: 0em;margin-bottom: 0px;margin…
行业资讯精选 82
数亿元！前京东副总裁“消失”4年：不造会跑的天工，他在仓库里给人形机器人“占工位”
一家“消失”了四年的机器人公司，重回公众视野时，手里已握着全球头部电商的产线订单。 默不作声的四年 它去仓库装分拣机了 9月1日，北京天下先智创机器人技术有限责任公司（下称“天工机…
行业资讯精选 82
Meta新研究：字节模型蒸馏后，天花板破了
Meta突破模型蒸馏技术，大幅提升小模型性能。
行业资讯精选 85
前OpenAI后训练VP回应陶哲轩：说AI毁了数学，可能还是太小瞧AI了
未来的AI不仅能解决问题，还能提出有价值的洞见、建立新的概念框架，让数学家在此基础上继续探索。
行业资讯精选 75
9月21日，深圳前海！聊聊工业AI与生态共创的下一步· 西门子Xcelerator繁星生态大会
西门子Xcelerator繁星生态大会，在生态合作中共赢机遇
行业资讯精选 82
超级智能体“迪迪虾”上车，比亚迪携手阿里云推进智能座舱升级
9月14日，比亚迪“超级智能体迪迪虾暨腾势N8L纯电上市发布会”举行。围绕下一代智能座舱，比亚迪正与阿里云推进AI合作：千问大模型接入比亚迪座舱体系，参与用户多轮对话与复杂意图理解…
行业资讯精选 82
字节跳动CEO梁汝波：豆包、飞书与火山引擎整合后 将加大企业市场投入
9月15日，2026飞书未来无限大会在京举办，字节跳动CEO梁汝波在会上进行了分享。他表示，在Agent进入可用阶段后，Agent、模型、协作环境三者必须紧密融合，才能真正发挥价值…
行业资讯精选 80
工作流可以自我进化了，英伟达开源 SoL-Pi，每小时省13.5刀！
<img data-src="https://static.leiphone.com/uploads/new/images/20260915/6aa8b4ebe58da.jpg" …
行业资讯精选 85
横扫四榜，DM0.5 凭什么面面俱到？
<img data-src="https://static.leiphone.com/uploads/new/images/20260915/6aa8ab33e2ac9.jpg" …
行业资讯精选 80
强化学习大本营新作：如何破解「学新忘旧」困局
<img class="rich_pages wxw-img" data-aistatus="1" data-croporisrc="https://mmbiz.qpic.cn/s…
行业资讯精选 95
7名博士生仅用3个月从零训练7B大模型：代码+数据+训练日志全公开
几百个Agent参与研发，AI开始造AI
行业资讯精选 81
不止一颗CPU？智能体经济时代，Arm对算力平台有了新理解
导语：200瓦功耗之差的新战场 记者：郭思 编辑：包永刚 <img border="0" style="" src="https://static.leiphone.com/upl…
行业资讯精选 82
Nvidia CEO Jensen Huang tells Trump ‘we’re not going to let [an AI slowdown] happen’· 英伟达 CEO 黄仁勋告诉特朗普：我们不会让 AI 发展放缓
Though Elon Musk and Sam Altman have supported Dario Amodei's calls to slow the pace of AI…
行业资讯精选 75
Fashion app Daydream uses Apple Intelligence to help you shop the outfits in your camera roll· Daydream 利用苹果 AI 优化购物
Thanks to the launch of iOS 27, Daydream's app now includes features that can turn saved o…
行业资讯精选 80
Microsoft’s new AI ‘code of conduct’ tells models not to hack systems or trick humans· 微软新 AI 行为准则
The code of conduct lays out general principles that Microsoft AI models should uphold — s…
行业资讯精选 75
DevFest is back· DevFest 2026 回归
DevFest 2026 is back and here’s how you can connect with one of the more than 800 global e…
行业资讯精选 82
Perplexity Portable Computer Is Now Available on Windows, Powered by NVIDIA RTX· Perplexity 可携带电脑现支持 Windows，由 NVIDIA RTX 驱动
As local models become more capable, AI agents can handle more work directly on a PC while…
行业资讯精选 85
蚂蚁发布大模型内生式安全护栏SingProbe，让AI边生成边识别风险
大模型正在进入日常问答、办公辅助、客户服务等真实业务场景。用户期待AI快速给出有用的回答，也需要这些回答安全、可靠。如何及时发现不安全内容和编造信息，同时减少安全审核对响应速度和用…
行业资讯精选 75
国产智驾芯片新纪录，被产业伙伴联合书写的里程碑
<section data-mpa-powered-by="yiban.io" data-pm-slice="0 0 []" style="-webkit-tap-highligh…
YouTube 热门 AI 视频
AI 视频精选 65
Claude Is Now Leaving Invisible Fingerprints In Its Text· Claude 现在在文本中留下不可见指纹
YouTube · Two Minute Papers
📺 Two Minute Papers▶ 3.4k 次观看
AI 视频精选 80
Detecting wildfires with ChatGPT· 用 ChatGPT 检测野火
YouTube · OpenAI
📺 OpenAI▶ 2.8k 次观看