1. Lightning-AI/pytorch-lightning
分类：开源项目来源：github_search分数：100作者：Lightning-AI时间：2026-02-16T19:06:03Z
Lightning-AI 的 PyTorch Lightning 项目提供对原生 PyTorch 的工程化封装，自动处理分布式训练、混合精度与扩展部署，重要性在于显著降低训练基础设施复杂度并支持从单机到多节点的无缝扩展。
- 项目定位为深度学习训练框架，强调“你写模型逻辑，Lightning 处理工程细节”。
- 核心包含 PyTorch Lightning 与 Lightning Fabric 两个包，分别覆盖高层训练与专家级控制。
- 支持多GPU、多节点、TPU、16-bit 精度、早停、检查点、实验管理等 40+ 高级能力。
- 提供丰富示例，覆盖分类、分割、检测、文本摘要、音频生成、LLM 微调、扩散模型等任务。
- 与 Lightning Cloud 和 LitServe 生态联动，可用于云端训练与 Python 推理服务部署。
#GitHub #repo #开源项目 #PyTorch Lightning
2. microsoft/TypeChat
分类：开源项目来源：github_search分数：28作者：microsoft时间：2026-02-16T19:11:53Z
微软开源 TypeChat，通过“schema engineering”用类型定义意图并自动完成提示构造、结果校验与修复，降低构建自然语言接口的复杂度并提升结构化与可靠性。
- TypeChat 主打以类型/Schema替代复杂提示工程来构建自然语言应用。
- 可将用户输入映射到开发者定义的意图类型，支持联合类型与分层 meta-schema 设计。
- 框架会基于类型自动构造提示词，并校验 LLM 输出是否符合 schema。
- 当输出不符合约束时，TypeChat 可通过额外交互进行修复，提升结果可用性。
- 项目提供 TypeScript/JavaScript 使用方式，并支持从 Python、TypeScript、C#/.NET 源码体验与示例。
#GitHub #repo #开源项目 #TypeChat #Microsoft #LLM #TypeScript
3. taubyte/tau
分类：开源项目来源：github_search分数：16作者：taubyte时间：2026-02-16T23:15:25Z
Taubyte 发布开源项目 tau，提供 Git-native、可自托管的 PaaS，用 Git 定义基础设施并集成函数、存储、消息与 AI 能力，重要性在于可替代多家云平台并简化本地到生产的一体化部署。
- tau 是开源、Git-native 的 PaaS，强调通过 Git 管理基础设施而非调用 API。
- 支持完全自托管，定位能力接近 Vercel、Firebase、Cloudflare，并内置 AI 特性。
- 平台包含函数计算、Web 托管、KV 数据库、对象存储、PubSub、负载均衡、HTTP/DNS、自动 HTTPS 和私有网络。
- 提供 DreamCLI、桌面端与 Web Console，覆盖本地云创建、项目开发、构建运行到部署全流程。
- 文档、演示视频、示例项目和手动/自动化（Spore Drive）部署路径较完整，便于快速上手与社区贡献。
#GitHub #repo #开源项目 #Taubyte #tau
4. “Dear Upstairs Neighbors” (BTS)
分类：视频/演讲来源：youtube_rss分数：0作者：Google DeepMind时间：2026-01-26T18:00:04+00:00
《Dear Upstairs Neighbors》作为新动画短片在圣丹斯完成预映，Pixar前员工Connie He介绍其与45人资深团队及研究工程人员协作创作过程，体现跨学科动画制作的行业价值。
- 新动画短片《Dear Upstairs Neighbors》已在圣丹斯进行预映
- 由Pixar校友Connie He分享该片的创作与制作经历
- 项目集结约45人的动画资深团队参与
- 团队成员包含奥斯卡获奖者以及研究人员和工程师
- 内容强调动画创作中的跨领域协作模式
#YouTube #视频/演讲 #Connie He #Pixar
5. Gemini 3 Flash: Creative UI in a spark
分类：视频/演讲来源：youtube_rss分数：0作者：Google DeepMind时间：2025-12-17T16:00:34+00:00
Google DeepMind在视频中展示Gemini 3 Flash可即时生成代码与多风格UI方案并支持创意探索，这对加速产品原型设计和前端开发迭代具有重要价值。
- Gemini 3 Flash强调面向快速原型开发的能力，如即时代码生成。
- 模型具备审美风格多样性，可围绕同一需求产出不同UI设计。
- 支持在设计流程中原生驱动创意探索，例如自动生成风格探索提示词。
- 可一键生成多个差异化UI变体，用于加速方案比较与开发决策。
- 内容由Google DeepMind YouTube渠道发布，并提供官方模型页面进一步了解。
#YouTube #视频/演讲 #Gemini 3 Flash
6. Gemini 3 Flash: Transform messy data
分类：视频/演讲来源：youtube_rss分数：0作者：Google DeepMind时间：2025-12-17T16:00:29+00:00
Google DeepMind 发布视频介绍 Gemini 3 Flash 可用于将两份杂乱数据进行提取、转换与合并，借助多模态和复杂推理能力提升企业非结构化数据入库效率，因而对自动化数据工程具有实际价值。
- 主题聚焦企业常见的多步骤数据处理流程，如 ETL（提取、转换、加载）。
- 示例场景是合并两份“杂乱”的数据源并完成结构化处理。
- 强调 Gemini 3 Flash 的多模态能力与复杂推理能力可减少人工解析工作。
- 内容以 YouTube 频道视频形式发布，并附官方模型介绍链接。
#YouTube #视频/演讲 #Gemini 3 Flash
7. From Diagnosis to Inoculation: Building Cognitive Resistance to AI Disempowerment
分类：研究/论文来源：arxiv_search分数：100作者：Aleksey Komissarov时间：2026-02-16T23:47:13Z
该论文提出基于“接种理论”的AI素养教学框架与八项学习目标，并通过在线课程案例验证其可帮助学习者识别与抵御AI导致的认知失权，这对从问题诊断走向可实施教育干预具有重要意义。
- 基于Sharma等（2026）提出的现实扭曲、价值判断扭曲与行动扭曲，论文聚焦如何进行教学层面的具体干预。
- 作者提出八项跨领域学习目标（LOs），源于教学实践，后与既有“认知失权”分类体系呈现收敛一致。
- 案例采用“人机共教”模式，让AI作为共同授课声音，进行有引导的失败模式暴露训练。
- 论文将McGuire接种理论引入AI素养教育，强调仅靠陈述性知识不足，需预先演练应对谄媚迎合与权威投射等模式。
- 作者认为“教学框架”与“实证分类”两条独立路径的趋同，增强了问题诊断与教育方案的可信度。
#arXiv #paper #研究/论文
8. Fast and Effective On-policy Distillation from Reasoning Prefixes
分类：研究/论文来源：arxiv_search分数：98作者：Dongxu Zhang时间：2026-02-16T23:28:54Z
该论文提出仅对学生输出前缀进行在策略蒸馏并提前终止采样的方法，在数学与跨域基准上达到与完整OPD相当效果，同时将训练算力开销降低2到47倍，提升了长推理模型训练效率与可扩展性。
- 传统在策略蒸馏需在线采样学生完整轨迹并逐token监督，训练成本在长回答场景下很高。
- 作者发现有效训练信号常集中在输出前缀，且短教师前缀就能显著引导学生得到正确答案。
- 方法上将蒸馏目标限制为学生生成前缀，并在蒸馏时提前终止采样，减少无效计算。
- 在AI-for-Math及域外基准实验中，前缀在策略蒸馏与完整OPD性能匹配。
- 训练FLOP相对完整OPD降低约2x到47x，显示出明显的效率优势。
#arXiv #paper #研究/论文 #On-policy Distillation #AI-for-Math
9. Knowing Isn't Understanding: Re-grounding Generative Proactivity with Epistemic and Behavioral Insight
分类：研究/论文来源：arxiv_search分数：95作者：Kirandeep Kaur时间：2026-02-16T23:28:17Z
该论文提出生成式智能体应从“回答已知问题”转向在用户认知不完备时进行受约束的主动干预，这对避免误导与提升人机协作质量至关重要。
- 指出现有生成式AI将“理解”简化为解答显式提问，忽视用户对未知风险和缺失信息的盲区。
- 提出“认知不完备（epistemic incompleteness）”概念，强调在未知未知存在时主动性是认知层面的必要条件。
- 批评当前主动性方法多基于历史行为外推并假设目标已明确，难以真正支持用户决策与探索。
- 强调主动干预并非越多越好，缺乏约束会导致注意力偏移、信息过载甚至潜在伤害。
- 主张将认识论基础与行为约束结合，借鉴无知哲学与主动行为研究来设计负责任的生成式主动智能体。
#arXiv #paper #研究/论文
10. How to Train Your Long-Context Visual Document Model
分类：研究/论文来源：arxiv_search分数：92作者：Austin Veselka时间：2026-02-16T23:26:51Z
该论文系统研究了最长344K上下文的长文档视觉语言模型训练流程，并在MMLongBenchDoc上达SOTA，重要性在于提供可复现配方并验证视觉长上下文训练可迁移提升长文本能力。
- 开展首个大规模、系统性的长上下文视觉语言模型训练研究，覆盖持续预训练、监督微调与偏好优化。
- 在24B与32B参数规模上实现MMLongBenchDoc当前最优结果，并提供大量长上下文评测与消融分析。
- 发现训练上下文长度与评测长度匹配优于盲目训练更长上下文。
- 在训练与评测中加入页码索引可显著提升长文档问答表现。
- 合成数据管线支持模型自提升，并验证了“视觉长上下文训练→长文本能力”反向迁移。
- 发布人工修正的MMLBD-C基准版本，以减少原基准中的错误与低质量样本。
#arXiv #paper #研究/论文 #MMLongBenchDoc
11. Scaling Laws for Masked-Reconstruction Transformers on Single-Cell Transcriptomics
分类：研究/论文来源：arxiv_search分数：90作者：Ihor Kendiukhov时间：2026-02-16T23:20:58Z
该论文系统验证了单细胞转录组掩码重建Transformer的缩放定律：在数据充足时损失随模型规模呈幂律下降，而数据稀缺时几乎不缩放，说明数据-参数比是构建高效单细胞基础模型的关键。
- 针对scRNA-seq首次系统研究掩码重建Transformer的神经缩放规律。
- 设置两种实验：数据充足（512基因、20万细胞）与数据受限（1024基因、1万细胞）。
- 评估7种模型规模（533到3.4×10^8参数），并对验证集MSE拟合参数化缩放律。
- 数据充足条件下出现明显幂律缩放，且存在不可约损失下限 c≈1.44。
- 数据受限条件下缩放效应微弱，表明瓶颈主要在数据而非模型容量。
- 将渐近下限初步换算为信息论量约2.30 bit/被mask基因位，指向后续基础模型设计与测量方向。
#arXiv #paper #研究/论文 #scRNA-seq #Transformer
12. Decision Making under Imperfect Recall: Algorithms and Benchmarks
分类：研究/论文来源：arxiv_search分数：88作者：Emanuel Tewolde时间：2026-02-16T23:19:01Z
In game theory, imperfect-recall decision problems model situations in which an agent forgets information it held before. They encompass games such as the ``ab…
- In game theory, imperfect-recall decision problems model situations in which an agent forgets information it held before
- They encompass games such as the ``absentminded driver'' and team games with limited communication
- In this paper, we introduce the first benchmark suite for imperfect-recall decision problems
- Our benchmarks capture a variety of problem types, including ones concerning privacy in AI systems that elicit sensitive information, and A…
- Across 61 problem instances generated using this suite, we evaluate the performance of different algorithms for finding first-order optimal…
- In particular, we introduce the family of regret matching (RM) algorithms for nonlinear constrained optimization
#arXiv #paper #研究/论文
13. Artificial Intelligence Specialization in the European Union: Underexplored Role of the Periphery at NUTS-3 Level
分类：研究/论文来源：arxiv_search分数：85作者：Victor Herrero-Solana时间：2026-02-16T23:01:14Z
This study examines the geographical distribution of Artificial Intelligence (AI) research production across European regions at the NUTS-3 level for the perio…
- This study examines the geographical distribution of Artificial Intelligence (AI) research production across European regions at the NUTS-3…
- Using bibliometric data from Clarivate InCites and the Citation Topics classification system, we analyze two hierarchical levels of themati…
- 61)
- We calculate the Relative Specialization Index (RSI) and Relative Citation Impact (RCI) for 781 NUTS-3 regions
- While major metropolitan hubs such as Paris (IIle-de-France), Warszawa, and Madrid lead in absolute production volume, our findings reveal …
- Notably, we find virtually no correlation between regional specialization and citation impact, identifying four distinct regional profiles:…
#arXiv #paper #研究/论文
14. Predicting Invoice Dilution in Supply Chain Finance with Leakage Free Two Stage XGBoost, KAN (Kolmogorov Arnold Networks), and Ensemble Models
分类：研究/论文来源：arxiv_search分数：82作者：Pavel Koptev时间：2026-02-16T23:00:39Z
Invoice or payment dilution is the gap between the approved invoice amount and the actual collection is a significant source of non credit risk and margin loss…
- Invoice or payment dilution is the gap between the approved invoice amount and the actual collection is a significant source of non credit …
- Traditionally, this risk is managed through the buyer's irrevocable payment undertaking (IPU), which commits to full payment without deduct…
- However, IPUs can hinder supply chain finance adoption, particularly among sub-invested grade buyers
- A newer, data-driven methods use real-time dynamic credit limits, projecting dilution for each buyer-supplier pair in real-time
- This paper introduces an AI, machine learning framework and evaluates how that can supplement a deterministic algorithm to predict invoice …
#arXiv #paper #研究/论文
15. MyoInteract: A Framework for Fast Prototyping of Biomechanical HCI Tasks using Reinforcement Learning
分类：研究/论文来源：arxiv_search分数：80作者：Ankit Bhattarai时间：2026-02-16T22:51:57Z
Reinforcement learning (RL)-based biomechanical simulations have the potential to revolutionise HCI research and interaction design, but currently lack usabili…
- Reinforcement learning (RL)-based biomechanical simulations have the potential to revolutionise HCI research and interaction design, but cu…
- Using the Human Action Cycle as a design lens, we identify key limitations of biomechanical RL frameworks and develop MyoInteract, a novel …
- MyoInteract allows designers to setup tasks, user models, and training parameters from an easy-to-use GUI within minutes
- It trains and evaluates muscle-actuated simulated users within minutes, reducing training times by up to 98%
- A workshop study with 12 interaction designers revealed that MyoInteract allowed novices in biomechanical RL to successfully setup, train, …
- By transforming biomechanical RL from a days-long expert task into an accessible hour-long workflow, this work significantly lowers barrier…
#arXiv #paper #研究/论文
16. GenAI for Systems: Recurring Challenges and Design Principles from Software to Silicon
分类：研究/论文来源：arxiv_search分数：78作者：Arya Tschand时间：2026-02-16T22:45:33Z
Generative AI is reshaping how computing systems are designed, optimized, and built, yet research remains fragmented across software, architecture, and chip de…
- Generative AI is reshaping how computing systems are designed, optimized, and built, yet research remains fragmented across software, archi…
- This paper takes a cross-stack perspective, examining how generative models are being applied from code generation and distributed runtimes…
- Rather than reviewing each layer in isolation, we analyze how the same structural difficulties and effective responses recur across the sta…
- Our central finding is one of convergence
- Despite the diversity of domains and tools, the field keeps encountering five recurring challenges (the feedback loop crisis, the tacit kno…
- We organize these into a challenge--principle map that serves as a diagnostic and design aid, showing which principles have proven effectiv…
#arXiv #paper #研究/论文
17. Size Transferability of Graph Transformers with Convolutional Positional Encodings
分类：研究/论文来源：arxiv_search分数：75作者：Javier Porras-Valenzuela时间：2026-02-16T22:38:56Z
Transformers have achieved remarkable success across domains, motivating the rise of Graph Transformers (GTs) as attention-based architectures for graph-struct…
- Transformers have achieved remarkable success across domains, motivating the rise of Graph Transformers (GTs) as attention-based architectu…
- A key design choice in GTs is the use of Graph Neural Network (GNN)-based positional encodings to incorporate structural information
- In this work, we study GTs through the lens of manifold limit models for graph sequences and establish a theoretical connection between GTs…
- Building on transferability results for GNNs under manifold convergence, we show that GTs inherit transferability guarantees from their pos…
- In particular, GTs trained on small graphs provably generalize to larger graphs under mild assumptions
- We complement our theory with extensive experiments on standard graph benchmarks, demonstrating that GTs exhibit scalable behavior on par w…
#arXiv #paper #研究/论文
18. Closing the Distribution Gap in Adversarial Training for LLMs
分类：研究/论文来源：arxiv_search分数：72作者：Chengzhi Hu时间：2026-02-16T22:34:52Z
Adversarial training for LLMs is one of the most promising methods to reliably improve robustness against adversaries. However, despite significant progress, m…
- Adversarial training for LLMs is one of the most promising methods to reliably improve robustness against adversaries
- However, despite significant progress, models remain vulnerable to simple in-distribution exploits, such as rewriting prompts in the past t…
- We argue that this persistent fragility stems from a fundamental limitation in current adversarial training algorithms: they minimize adver…
- To bridge this gap, we propose Distributional Adversarial Training, DAT
- We leverage Diffusion LLMs to approximate the true joint distribution of prompts and responses, enabling generation of diverse, high-likeli…
- By combining optimization over the data distribution provided by the diffusion model with continuous adversarial training, DAT achieves sub…
#arXiv #paper #研究/论文
19. BindCLIP: A Unified Contrastive-Generative Representation Learning Framework for Virtual Screening
分类：研究/论文来源：arxiv_search分数：70作者：Anjie Qiao时间：2026-02-16T22:26:55Z
Virtual screening aims to efficiently identify active ligands from massive chemical libraries for a given target pocket. Recent CLIP-style models such as DrugC…
- Virtual screening aims to efficiently identify active ligands from massive chemical libraries for a given target pocket
- Recent CLIP-style models such as DrugCLIP enable scalable virtual screening by embedding pockets and ligands into a shared space
- However, our analyses indicate that such representations can be insensitive to fine-grained binding interactions and may rely on shortcut c…
- To address these issues, we propose BindCLIP, a unified contrastive-generative representation learning framework for virtual screening
- BindCLIP jointly trains pocket and ligand encoders using CLIP-style contrastive learning together with a pocket-conditioned diffusion objec…
- To further mitigate shortcut reliance, we introduce hard-negative augmentation and a ligand-ligand anchoring regularizer that prevents repr…
#arXiv #paper #研究/论文
20. tensorFM: Low-Rank Approximations of Cross-Order Feature Interactions
分类：研究/论文来源：arxiv_search分数：68作者：Alessio Mazzetto时间：2026-02-16T22:21:48Z
We address prediction problems on tabular categorical data, where each instance is defined by multiple categorical attributes, each taking values from a finite…
- We address prediction problems on tabular categorical data, where each instance is defined by multiple categorical attributes, each taking …
- These attributes are often referred to as fields, and their categorical values as features
- Such problems frequently arise in practical applications, including click-through rate prediction and social sciences
- We introduce and analyze {tensorFM}, a new model that efficiently captures high-order interactions between attributes via a low-rank tensor…
- Our model generalizes field-weighted factorization machines
- Empirically, tensorFM demonstrates competitive performance with state-of-the-art methods
#arXiv #paper #研究/论文