[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-gigwegbe--tinyml-papers-and-projects":3,"tool-gigwegbe--tinyml-papers-and-projects":64},[4,17,27,35,43,56],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},3808,"stable-diffusion-webui","AUTOMATIC1111\u002Fstable-diffusion-webui","stable-diffusion-webui 是一个基于 Gradio 构建的网页版操作界面，旨在让用户能够轻松地在本地运行和使用强大的 Stable Diffusion 图像生成模型。它解决了原始模型依赖命令行、操作门槛高且功能分散的痛点，将复杂的 AI 绘图流程整合进一个直观易用的图形化平台。\n\n无论是希望快速上手的普通创作者、需要精细控制画面细节的设计师，还是想要深入探索模型潜力的开发者与研究人员，都能从中获益。其核心亮点在于极高的功能丰富度：不仅支持文生图、图生图、局部重绘（Inpainting）和外绘（Outpainting）等基础模式，还独创了注意力机制调整、提示词矩阵、负向提示词以及“高清修复”等高级功能。此外，它内置了 GFPGAN 和 CodeFormer 等人脸修复工具，支持多种神经网络放大算法，并允许用户通过插件系统无限扩展能力。即使是显存有限的设备，stable-diffusion-webui 也提供了相应的优化选项，让高质量的 AI 艺术创作变得触手可及。",162132,3,"2026-04-05T11:01:52",[13,14,15],"开发框架","图像","Agent","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",138956,2,"2026-04-05T11:33:21",[13,15,26],"语言模型",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":23,"last_commit_at":33,"category_tags":34,"status":16},2271,"ComfyUI","Comfy-Org\u002FComfyUI","ComfyUI 是一款功能强大且高度模块化的视觉 AI 引擎，专为设计和执行复杂的 Stable Diffusion 图像生成流程而打造。它摒弃了传统的代码编写模式，采用直观的节点式流程图界面，让用户通过连接不同的功能模块即可构建个性化的生成管线。\n\n这一设计巧妙解决了高级 AI 绘图工作流配置复杂、灵活性不足的痛点。用户无需具备编程背景，也能自由组合模型、调整参数并实时预览效果，轻松实现从基础文生图到多步骤高清修复等各类复杂任务。ComfyUI 拥有极佳的兼容性，不仅支持 Windows、macOS 和 Linux 全平台，还广泛适配 NVIDIA、AMD、Intel 及苹果 Silicon 等多种硬件架构，并率先支持 SDXL、Flux、SD3 等前沿模型。\n\n无论是希望深入探索算法潜力的研究人员和开发者，还是追求极致创作自由度的设计师与资深 AI 绘画爱好者，ComfyUI 都能提供强大的支持。其独特的模块化架构允许社区不断扩展新功能，使其成为当前最灵活、生态最丰富的开源扩散模型工具之一，帮助用户将创意高效转化为现实。",107662,"2026-04-03T11:11:01",[13,14,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":23,"last_commit_at":41,"category_tags":42,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,26],{"id":44,"name":45,"github_repo":46,"description_zh":47,"stars":48,"difficulty_score":23,"last_commit_at":49,"category_tags":50,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",84991,"2026-04-05T10:45:23",[14,51,52,53,15,54,26,13,55],"数据工具","视频","插件","其他","音频",{"id":57,"name":58,"github_repo":59,"description_zh":60,"stars":61,"difficulty_score":10,"last_commit_at":62,"category_tags":63,"status":16},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,"2026-04-04T04:44:48",[15,14,13,26,54],{"id":65,"github_repo":66,"name":67,"description_en":68,"description_zh":69,"ai_summary_zh":70,"readme_en":71,"readme_zh":72,"quickstart_zh":73,"use_case_zh":74,"hero_image_url":75,"owner_login":76,"owner_name":77,"owner_avatar_url":78,"owner_bio":79,"owner_company":80,"owner_location":81,"owner_email":82,"owner_twitter":83,"owner_website":84,"owner_url":85,"languages":86,"stars":87,"forks":88,"last_commit_at":89,"license":90,"difficulty_score":91,"env_os":92,"env_gpu":93,"env_ram":93,"env_deps":94,"category_tags":97,"github_topics":98,"view_count":23,"oss_zip_url":86,"oss_zip_packed_at":86,"status":16,"created_at":105,"updated_at":106,"faqs":107,"releases":108},2091,"gigwegbe\u002Ftinyml-papers-and-projects","tinyml-papers-and-projects","This is a list of interesting papers and projects about TinyML. ","tinyml-papers-and-projects 是一个专注于 TinyML（微型机器学习）领域的精选资源库，旨在汇集该方向最具价值的学术论文、开源项目、技术文章及演讲资料。随着物联网设备对本地智能处理需求的激增，如何在内存仅几 KB 的微控制器上高效运行深度学习模型成为一大挑战，而该资源库正是为解决这一痛点而生。它系统性地整理了从模型压缩、量化训练到硬件感知架构搜索等关键技术的前沿成果，帮助从业者快速定位从理论突破到落地实践的核心资料。\n\n无论是希望深入探索算法优化的研究人员，还是需要在资源受限设备上部署 AI 应用的嵌入式开发者，都能从中获益。资源库按年份梳理了包括 SqueezeNet、CMSIS-NN、ProxylessNAS 等里程碑式论文，并收录了如 OpenMV、GesturePod 等具有参考价值的实战项目代码。其独特的亮点在于不仅涵盖纯理论研究，还特别关注软硬件协同设计与实际基准测试，提供了从书籍、课程到工具库的全方位学习路径。对于想要入门或深耕端侧智能的工程师而言，tinyml-papers-and-projects 就像一位博学的向导，让复杂的技术演进脉络变得","tinyml-papers-and-projects 是一个专注于 TinyML（微型机器学习）领域的精选资源库，旨在汇集该方向最具价值的学术论文、开源项目、技术文章及演讲资料。随着物联网设备对本地智能处理需求的激增，如何在内存仅几 KB 的微控制器上高效运行深度学习模型成为一大挑战，而该资源库正是为解决这一痛点而生。它系统性地整理了从模型压缩、量化训练到硬件感知架构搜索等关键技术的前沿成果，帮助从业者快速定位从理论突破到落地实践的核心资料。\n\n无论是希望深入探索算法优化的研究人员，还是需要在资源受限设备上部署 AI 应用的嵌入式开发者，都能从中获益。资源库按年份梳理了包括 SqueezeNet、CMSIS-NN、ProxylessNAS 等里程碑式论文，并收录了如 OpenMV、GesturePod 等具有参考价值的实战项目代码。其独特的亮点在于不仅涵盖纯理论研究，还特别关注软硬件协同设计与实际基准测试，提供了从书籍、课程到工具库的全方位学习路径。对于想要入门或深耕端侧智能的工程师而言，tinyml-papers-and-projects 就像一位博学的向导，让复杂的技术演进脉络变得清晰可循，极大地降低了学习与研发门槛。","## TinyML Papers and Projects\n>\n> TinyML is awesome.\n\n[![Awesome](https:\u002F\u002Fawesome.re\u002Fbadge.svg)](https:\u002F\u002Fawesome.re) [![Contributions](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fissues-pr-closed-raw\u002Fgigwegbe\u002Ftinyml-papers-and-projects.svg?label=contributions)](https:\u002F\u002Fgithub.com\u002Fgigwegbe\u002Ftinyml-papers-and-projects\u002Fpulls) [![Commits](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flast-commit\u002Fgigwegbe\u002Ftinyml-papers-and-projects.svg?label=last%20contribution)](https:\u002F\u002Fgithub.com\u002Fgigwegbe\u002Ftinyml-papers-and-projects\u002Fcommits\u002Fmain)\n\nThis is a list of interesting papers, projects, articles and talks about TinyML.\n\n- [Awesome Papers](#awesome-papers-): [2016](#2016) | [2017](#2017) | [2018](#2018) | [2019](#2019) | [2020](#2020) | [2022](#2022) | [2023](#2023) | [2024](#2024) | [2025](#2025)\n- [Awesome Projects](#awesome-tinyml-projects): [Projects Source code](#projects-source-code) | [Projects Articles](#projects-articles)\n- [Benchmarking](#benchmarking)\n- Resources\n  - [Articles](#articles)\n  - [Books](#books)\n  - [Libraries and Tools](#libraries-and-tools)\n  - [Courses](#courses)\n  - [TinyML Talks](#tinyml-talks--conferences)\n- [Contact & Feedback](#contact--feedback)\n\n## Awesome Papers\n\n### \u003Cins>**2016**\u003C\u002Fins>\n\n- DEEP COMPRESSION: COMPRESSING DEEP NEURAL NETWORKS WITH PRUNING, TRAINED QUANTIZATION AND HUFFMAN CODING | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1510.00149.pdf)\n- **[SQUEEZENET]** ALEXNET-LEVEL ACCURACY WITH50X FEWER PARAMETERS AND \u003C0.5MB MODEL SIZE | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1602.07360.pdf)\n\n### \u003Cins>**2017**\u003C\u002Fins>\n\n- Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1712.05877.pdf)\n- Resource-efficient Machine Learning in 2 KB RAM for the Internet of Things  | [`[pdf]`](https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fresearch\u002Fuploads\u002Fprod\u002F2017\u002F06\u002Fkumar17.pdf)\n- ProtoNN: Compressed and Accurate kNN for Resource-scarce Devices  | [`[pdf]`](https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fresearch\u002Fuploads\u002Fprod\u002F2017\u002F06\u002Fprotonn.pdf)\n- OPENMV: A PYTHON POWERED, EXTENSIBLE MACHINE VISION CAMERA | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1711.10464.pdf) [`[official code]`]( https:\u002F\u002Fgithub.com\u002Fopenmv\u002Fopenmv.git)\n\n### \u003Cins>**2018**\u003C\u002Fins>\n\n- **[AMC]** AutoML for Model Compression and Acceleration on Mobile Devices | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1802.03494.pdf) [`[official code]`](https:\u002F\u002Fgithub.com\u002Fmit-han-lab\u002Famc)\n- Mobile Machine Learning Hardware at ARM: A Systems-on-Chip (SoC) Perspective | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1801.06274.pdf)\n- **[HAQ]** Hardware-Aware Automated Quantization with Mixed Precision | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fabs\u002F1811.08886)\n- Efficient and Robust Machine Learning for Real-World Systems | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1812.02240.pdf)\n- **[GesturePod]** Gesture-based Interaction Cane for People with Visual Impairments | [`[pdf]`](https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fresearch\u002Fuploads\u002Fprod\u002F2018\u002F05\u002FCHI19_GesturePod.pdf)\n- **[YOLO-LITE]** A Real-Time Object Detection Algorithm Optimized for Non-GPU Computers | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1811.05588v1.pdf)\n- **[CMSIS-NN]** Efficient Neural Network Kernels for Arm Cortex-M CPUs | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1801.06601.pdf)\n- Quantizing deep convolutional networks for efficient inference: A whitepaper | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1806.08342.pdf)\n- **[Hello Edge]** Keyword Spotting on Microcontrollers | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1711.07128.pdf)\n\n  | ▲ [Top](#tinyml-papers-and-projects) |\n  | ------------------------------------ |\n\n### \u003Cins>**2019**\u003C\u002Fins>\n\n- FastGRNN: A Fast, Accurate, Stable and Tiny Kilobyte Sized Gated Recurrent Neural Network | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1901.02358.pdf)\n- Image Classification on IoT Edge Devices: Profiling and Modeling| [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1902.11119.pdf)\n- **[PROXYLESSNAS]** DIRECT NEURAL ARCHITECTURE SEARCH ON TARGET TASK AND HARDWARE |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1812.00332.pdf) [`[official code]`](https:\u002F\u002Fgithub.com\u002Fmit-han-lab\u002Fproxylessnas)\n- Energy Efficient Hardware for On-Device CNN Inference via Transfer Learning | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1812.01672.pdf)\n- Visual Wake Words Dataset | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1906.05721.pdf)\n- Compiling KB-Sized Machine Learning Models to Tiny IoT Devices | [`[pdf]`](microsoft.com\u002Fen-us\u002Fresearch\u002Fuploads\u002Fprod\u002F2018\u002F10\u002Fpldi19-SeeDot.pdf)\n- Reconfigurable Multitask Audio Dynamics Processing Scheme | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fabs\u002F1903.06392 )\n- Pushing the limits of RNN Compression | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1910.02558.pdf)\n- A low-power end-to-end hybrid neuromorphic framework for surveillance applications | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1910.09806.pdf)\n- Deep Learning at the Edge | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1910.10231.pdf)\n- Memory-Driven Mixed Low Precision Quantization For Enabling Deep Network Inference On Microcontrollers | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1905.13082.pdf) [`[official code]`](https:\u002F\u002Fgithub.com\u002FEEESlab\u002FCMix-NN)\n- **[SpArSe]** Sparse Architecture Search for CNNs on Resource-Constrained Microcontrollers  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1905.12107.pdf)\n- **[MobileNetV2]** Inverted Residuals and Linear Bottlenecks |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1801.04381.pdf)\n- Latent Weights Do Not Exist: Rethinking Binarized Neural Network Optimization  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1906.02107.pdf)\n- Low-Power Computer Vision: Status, Challenges, Opportunities |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1904.07714.pdf)\n\n  | ▲ [Top](#tinyml-papers-and-projects) |\n  | ------------------------------------ |\n\n### \u003Cins>**2020**\u003C\u002Fins>\n\n- COMPRESSING RNNS FOR IOT DEVICES BY 15-38X USING KRONECKER PRODUCTS |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1906.02876.pdf)\n- BENCHMARKING TINYML SYSTEMS: CHALLENGES AND DIRECTION |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2003.04821v3.pdf)\n- Lite Transformer with Long-Short Range Attention |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2004.11886.pdf)\n- **[FANN-on-MCU]** An Open-Source Toolkit for Energy-Efficient Neural Network Inference at the Edge of the Internet of Things |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1911.03314.pdf)\n- **[TENSORFLOW LITE MICRO]** EMBEDDED MACHINE LEARNING ON TINYML SYSTEMS |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2010.08678v2.pdf)\n- **[AttendNets]** Tiny Deep Image Recognition Neural Networks for the Edge via Visual Attention Condensers |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2009.14385v1.pdf)\n- **[TinySpeech]** Attention Condensers for Deep Speech Recognition Neural Networks on Edge Devices |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2008.04245v6.pdf)\n- Robust navigation with tinyML for autonomous mini-vehicles |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2007.00302v1.pdf) [`[official code]`](https:\u002F\u002Fgithub.com\u002Fpraesc\u002FRobust-navigation-with-TinyML)\n- **[MICRONETS]** NEURAL NETWORK ARCHITECTURES FOR DEPLOYING TINYML APPLICATIONS ON COMMODITY MICROCONTROLLERS |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2010.11267v2.pdf)\n- **[TinyLSTMs]** Efficient Neural Speech Enhancement for Hearing Aids |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2005.11138.pdf)\n- **[MCUNet]** Tiny Deep Learning on IoT Devices |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fabs\u002F2007.10319) [`[official code]`](https:\u002F\u002Fgithub.com\u002Fmit-han-lab\u002Fmcunet)\n- Efficient Residue Number System Based Winograd Convolution | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2007.12216.pdf)\n- INTEGER QUANTIZATION FOR DEEP LEARNING INFERENCE: PRINCIPLES AND EMPIRICAL EVALUATION  | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2004.09602)\n- On Front-end Gain Invariant Modeling for Wake Word Spotting | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2010.06676.pdf)\n- TOWARDS DATA-EFFICIENT MODELING FOR WAKE WORD SPOTTING | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2010.06659.pdf)\n- Accurate Detection of Wake Word Start and End Using a CNN | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2008.03790.pdf)\n- **[PoPS]** Policy Pruning and Shrinking for Deep Reinforcement Learning | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2001.05012.pdf)\n- Howl: A Deployed, Open-Source Wake Word Detection System | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2008.09606.pdf) [`[official code]`](https:\u002F\u002Fgithub.com\u002Fcastorini\u002Fhowl)\n- **[LeakyPick]** IoT Audio Spy Detector | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2007.00500.pdf)\n- On-Device Machine Learning: An Algorithms and Learning Theory Perspective  | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1911.00623.pdf)\n- Leveraging Automated Mixed-Low-Precision Quantization for tiny edge microcontrollers | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2008.05124.pdf)\n- OPTIMIZE WHAT MATTERS: TRAINING DNN-HMM KEYWORD SPOTTING MODEL USING END METRIC | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.01151.pdf)\n- **[RNNPool]** Efficient Non-linear Pooling for RAM Constrained Inference | [`[blog]`](https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fresearch\u002Fblog\u002Fseeing-on-tiny-battery-powered-microcontrollers-with-rnnpool\u002F?utm_medium=email&_hsmi=104017359&_hsenc=p2ANqtz-_DVkWnyh_NhAV6j4hTFngepUyiNjZ5GO5CYIQfpl5NzerjwxOBQcpdkilzGpt9ic4HglvgM80h7wIkFNX89xe-3_j7Kw&utm_content=104017359&utm_source=hs_email)  [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2002.11921.pdf) [`[official code]`](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FEdgeML\u002Fblob\u002Fmaster\u002Fpytorch\u002Fedgeml_pytorch\u002Fgraph\u002Frnnpool.py)\n- **[Shiftry]** RNN Inference in 2KB of RAM |[`[pdf]`](https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fresearch\u002Fuploads\u002Fprod\u002F2020\u002F10\u002Foopsla20main-p230-p-aba27a6-48263M-final.pdf)\n- **[Once for All]** Train One Network and Specialize it for Efficient Deployment  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1908.09791.pdf) [`[official code]`](https:\u002F\u002Fgithub.com\u002Fmit-han-lab\u002Fonce-for-all)\n- A Tiny CNN Architecture for Medical Face Mask Detection for Resource-Constrained Endpoints  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.14858.pdf)\n- Rethinking Generalization in American Sign Language Prediction for Edge Devices with Extremely Low Memory Footprint |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.13741.pdf) [`[presentation]`](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=bJ1vnhAbJ9o&feature=youtu.be)\n- **[ShadowNet]** A Secure and Efficient System for On-device Model Inference |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.05905.pdf)\n- Hardware Aware Training for Efficient Keyword Spotting on General Purpose and Specialized Hardware |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2009.04465.pdf)\n- Automated facial recognition for wildlife that lack unique markings: A deep learning approach for brown bears  |[`[pdf]`](https:\u002F\u002Fonlinelibrary.wiley.com\u002Fdoi\u002Fepdf\u002F10.1002\u002Fece3.6840)\n- **[HyNNA]**: Improved Performance for Neuromorphic Vision Sensor based Surveillance\nusing Hybrid Neural Network Architecture |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2003.08603.pdf)\n- The Hardware Lottery |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2009.06489.pdf)\n- MLPerf Inference Benchmark |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1911.02549.pdf)\n- MLPerf Mobile Inference Benchmark : Why Mobile AI Benchmarking Is Hard and What to Do About It |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2012.02328.pdf)\n- **[TinyRL]** Learning to Seek: Tiny Robot Learning for Source Seeking on a Nano Quadcopter |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1909.11236.pdf) [`[presentation]`](https:\u002F\u002Fyoutu.be\u002FwmVKbX7MOnU)\n- Pushing the Limits of Narrow Precision Inferencing at Cloud Scale with Microsoft Floating Point |[`[pdf]`](https:\u002F\u002Fproceedings.neurips.cc\u002F\u002Fpaper\u002F2020\u002Ffile\u002F747e32ab0fea7fbd2ad9ec03daa3f840-Paper.pdf)\n- **[TinyBERT]** Distilling BERT for Natural Language Understanding  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1909.10351.pdf)\n- **[Larq]** An Open-Source Library for Training Binarized Neural Networks |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.09398.pdf) [`[presentation]`](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=f9SNqDejOB0) [`[official code]`](https:\u002F\u002Fgithub.com\u002Flarq\u002Flarq)\n- **[FedML]** A Research Library and Benchmark for Federated Machine Learning  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2007.13518.pdf)\n- Survey of Machine Learning Accelerators  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2009.00993.pdf)\n\n  | ▲ [Top](#tinyml-papers-and-projects) |\n  | ------------------------------------ |\n\n### \u003Cins>**2021**\u003C\u002Fins>\n\n- **[I-BERT]** Integer-only BERT Quantization |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2101.01321.pdf)\n- **[TinyTL]** Reduce Memory, Not Parameters for Efficient On-Device Learning |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2007.11622.pdf) [`[official code]`](https:\u002F\u002Fgithub.com\u002Fmit-han-lab\u002Ftinyml\u002Ftree\u002Fmaster\u002Ftinytl)\n- ON THE QUANTIZATION OF RECURRENT NEURAL NETWORKS |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2101.05453.pdf)\n- **[TINY TRANSDUCER]** A HIGHLY-EFFICIENT SPEECH RECOGNITION MODEL ON EDGE DEVICES |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2101.06856.pdf)\n- LARQ COMPUTE ENGINE: DESIGN, BENCHMARK, AND DEPLOY STATE-OF-THE-ART BINARIZED NEURAL NETWORKS |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.09398.pdf)\n- **[LEAF]** A LEARNABLE FRONTEND FOR AUDIO CLASSIFICATION |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2101.08596.pdf)\n- Enabling Large NNs on Tiny MCUs with Swapping |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2101.08744.pdf)\n- Fixed-point Quantization of Convolutional Neural Networks for Quantized Inference on Embedded Platforms |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.02147.pdf)\n- Estimating indoor occupancy through low-cost BLE devices |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.03351.pdf)\n- **[Tiny Eats]** Eating Detection on a Microcontroller |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2003.06699.pdf)\n- **[DEVICETTS]** A SMALL-FOOTPRINT, FAST, STABLE NETWORK FOR ON-DEVICE TEXT-TO-SPEECH |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2010.15311.pdf)\n- A 0.57-GOPS\u002FDSP Object Detection PIM Accelerator on FPGA |[`[pdf]`](https:\u002F\u002Fdl.acm.org\u002Fdoi\u002Fpdf\u002F10.1145\u002F3394885.3431659)\n- Rethinking Co-design of Neural Architectures and Hardware Accelerators |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.08619.pdf)\n- Sparsity in Deep Learning: Pruning and growth for efficient inference and training in neural networks |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.00554.pdf)\n- **[Apollo]** Transferable Architecture Exploration |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.01723.pdf)\n- DEEP NEURAL NETWORK BASED COUGH DETECTION USING BED-MOUNTED ACCELEROMETER MEASUREMENTS |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.04997.pdf)\n- TapNet: The Design, Training, Implementation, and Applications of a Multi-Task Learning CNN for Off-Screen Mobile Input|[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.09087.pdf)\n- MEMORY-EFFICIENT SPEECH RECOGNITION ON SMART DEVICES |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.11531.pdf)\n- SWIS - Shared Weight bIt Sparsity for Efficient Neural Network Acceleration |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.01308.pdf)\n- Hardware Aware Training for Efficient Keyword Spotting on General Purpose and Specialized Hardware |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2009.04465.pdf)\n- Hypervector Design for Efficient Hyperdimensional Computing on Edge Devices  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.06709.pdf)\n- When Being Soft Makes You Tough:A Collision Resilient Quadcopter Inspired by Arthropod Exoskeletons |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.04423.pdf)\n- **[TinyOL]** TinyML with Online-Learning on Microcontrollers |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.08295.pdf)\n- Quantization-Guided Training for Compact TinyML Models |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.06231.pdf)\n- hls4ml: An Open-Source Codesign Workflow to Empower Scientific Low-Power Machine Learning Devices |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.05579.pdf)\n- Memory-Efficient, Limb Position-Aware Hand Gesture Recognition using Hyperdimensional Computing |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.05267.pdf)\n- Dynamically Throttleable Neural Networks(TNN) |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.02836.pdf)\n- A Comprehensive Survey on Hardware-Aware Neural Architecture Search |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2101.09336.pdf)\n- An Intelligent Bed Sensor System for Non-Contact Respiratory Rate Monitoring |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.13792.pdf)\n- Measuring what Really Matters: Optimizing Neural Networks for TinyML |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.10645.pdf)\n- Few-Shot Keyword Spotting in Any Language |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.01454.pdf)\n- DOPING: A TECHNIQUE FOR EXTREME COMPRESSION OF LSTM MODELS USING SPARSE STRUCTURED ADDITIVE MATRICES |[`[pdf]`](https:\u002F\u002Fproceedings.mlsys.org\u002Fpaper\u002F2021\u002Ffile\u002Fa3f390d88e4c41f2747bfa2f1b5f87db-Paper.pdf)\n- **[OutlierNets]** Highly Compact Deep Autoencoder Network Architectures for On-Device Acoustic Anomaly Detection |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.00528.pdf)\n- **[TENT]** Efficient Quantization of Neural Networks on the tiny Edge with Tapered FixEd PoiNT |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.02233.pdf)\n- A 1D-CNN Based Deep Learning Technique for Sleep Apnea Detection in IoT Sensors  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2105.00528.pdf)\n- ADAPTIVE TEST-TIME AUGMENTATION FOR LOW-POWER CPU |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2105.06183.pdf)\n- Compiler Toolchains for Deep Learning Workloads on Embedded Platforms |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.04576.pdf)\n- **[ProxiMic]** Convenient Voice Activation via Close-to-Mic Speech Detected by a Single Microphone |[`[pdf]`](https:\u002F\u002Fdl.acm.org\u002Fdoi\u002Fpdf\u002F10.1145\u002F3411764.3445687)\n- **[Fusion-DHL]** WiFi, IMU, and Floorplan Fusion for Dense History of Locations in Indoor Environments |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2105.08837.pdf)\n- **[µNAS]** Constrained Neural Architecture Search for Microcontrollers |[`[pdf]`](https:\u002F\u002Fdl.acm.org\u002Fdoi\u002Fpdf\u002F10.1145\u002F3437984.3458836?utm_content=167905304&utm_medium=social&utm_source=linkedin&hss_channel=lcp-19239958)\n- RaspberryPI for mosquito neutralization by power laser |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2105.14190.pdf)\n- Widening Access to Applied Machine Learning with TinyML |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2106.04008.pdf)\n- Using Machine Learning in Embedded Systems |[`[pdf]`](https:\u002F\u002Fwww.tiriasresearch.com\u002Fwp-content\u002Fuploads\u002F2021\u002F05\u002FUsing-Machine-Learning-in-Embedded-Systems.pdf)\n- **[FRILL]** A Non-Semantic Speech Embedding for Mobile Devices |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.04609.pdf)\n- Few-Shot Keyword Spotting in Any Language  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.01454.pdf)\n- MLPerf Tiny Benchmark |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2106.07597.pdf)\n- A Survey of Quantization Methods for Efficient Neural Network Inference  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.13630)\n- Efficient Deep Learning: A Survey on Making Deep Learning Models Smaller, Faster, and Better |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2106.08962.pdf)\n- AttendSeg: A Tiny Attention Condenser Neural Network for Semantic Segmentation on the Edge |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.14623.pdf)\n- RANDOMNESS IN NEURAL NETWORK TRAINING:CHARACTERIZING THE IMPACT OF TOOLING |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2106.11872.pdf)\n- TinyML: Analysis of Xtensa LX6 microprocessor for Neural Network Applications by ESP32 SoC |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2106.10652.pdf)\n- **[Keyword Transformer]**: A Self-Attention Model for Keyword Spotting |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.00769.pdf)\n- LB-CNN: An Open Source Framework for Fast Training of Light Binary Convolutional Neural Networks using Chainer and Cupy |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2106.15350.pdf)\n- **[Only Train Once]**: A One-Shot Neural Network Training And Pruning Framework |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2107.07467.pdf)\n- **[BEANNA]**: A Binary-Enabled Architecture for Neural Network Acceleration|[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2108.02313.pdf)\n- A TinyML Platform for On-Device Continual Learning with Quantized Latent Replays |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2110.10486.pdf)\n- CLASSIFICATION OF ANOMALOUS GAIT USING MACHINE LEARNING TECHNIQUES AND EMBEDDED SENSORS |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2110.06139.pdf)\n- **[MOBILEVIT]**: LIGHT-WEIGHT, GENERAL-PURPOSE, AND MOBILE-FRIENDLY VISION TRANSFORMER |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2110.02178.pdf?utm_medium=email&_hsmi=175723863&_hsenc=p2ANqtz--rVrB87u6eUrx_A5XM8m9kyJc-xTO-fwCZheo-n_Mx9IQ02upaLz87dMne5xsrlcFq5G0vxBHD_IzIXHOIvuR--axMLA&utm_content=175723863&utm_source=hs_email)\n- **[MCUNetV2]**: Memory-Efficient Patch-based Inference for Tiny Deep Learning |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fabs\u002F2110.15352#)\n- **[LCS]**: LEARNING COMPRESSIBLE SUBSPACES FOR ADAPTIVE NETWORK COMPRESSION AT INFERENCE TIME |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2110.04252.pdf)\n- Feature Augmented Hybrid CNN for Stress Recognition Using Wrist-based Photoplethysmography Sensor |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2108.03166.pdf)\n- **[ANALOGNETS]**: ML-HW CO-DESIGN OF NOISE-ROBUST TINYML MODELS AND ALWAYS-ON ANALOG COMPUTE-IN-MEMORY ACCELERATOR |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2111.06503.pdf)\n- **[BSC]**: Block-based Stochastic Computing to Enable Accurate and Efficient TinyML |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2111.06686.pdf)\n- **[TiWS-iForest]**: Isolation Forest in Weakly Supervised and Tiny ML scenarios |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2111.15432.pdf)\n- **[RadarNet]**: Efficient Gesture Recognition Technique Utilizing a Miniature Radar Sensor|[`[pdf]`](https:\u002F\u002Fdl.acm.org\u002Fdoi\u002Fpdf\u002F10.1145\u002F3411764.3445367)\n- The Synergy of Complex Event Processing and Tiny Machine Learning in Industrial IoT  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2105.03371.pdf)\n\n  | ▲ [Top](#tinyml-papers-and-projects) |\n  | ------------------------------------ |\n\n### \u003Cins>**2022**\u003C\u002Fins>\n\n- A Heterogeneous In-Memory Computing Cluster For Flexible End-to-End Inference of Real-World Deep Neural Networks |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2201.01089.pdf)\n- CFU Playground: Full-Stack Open-Source Framework for Tiny Machine Learning (tinyML) Acceleration on FPGAs |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2201.01863.pdf)\n- BottleFit: Learning Compressed Representations in Deep Neural Networks for Effective and Efficient Split Computing |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2201.02693.pdf)\n- **[UDC]**: Unified DNAS for Compressible TinyML Models |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2201.05842.pdf)\n- A VM\u002FContainerized Approach for Scaling TinyML Applications |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2202.05057.pdf)\n- A Fast Network Exploration Strategy to Profile Low Energy Consumption for Keyword Spotting |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2202.02361.pdf)\n- PocketNN: Integer-only Training and Inference of Neural Networks via Direct Feedback Alignment and Pocket Activations in Pure C++ |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2201.02863.pdf)\n- **[TinyMLOps]**: Operational Challenges for Widespread Edge AI Adoption |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2203.10923.pdf)\n- **[Auritus]**: An Open-Source Optimization Toolkit for Training and Development of Human Movement Models and Filters Using Earables |[`[pdf]`](https:\u002F\u002Fwww.researchgate.net\u002Fpublication\u002F359759183_Auritus_An_Open-Source_Optimization_Toolkit_for_Training_and_Development_of_Human_Movement_Models_and_Filters_Using_Earables) |[`[code]`](https:\u002F\u002Fgithub.com\u002Fnesl\u002Fauritus)\n- Enabling Hyperparameter Tuning of Machine Learning Classifiers in Production |[`[pdf]`](https:\u002F\u002Fwww.researchgate.net\u002Fpublication\u002F356911955_Enabling_Hyperparameter_Tuning_of_Machine_Learning_Classifiers_in_Production)\n- TinyOdom: Hardware-Aware Efficient Neural Inertial Navigation |[`[pdf]`](https:\u002F\u002Fwww.researchgate.net\u002Fpublication\u002F360075622_TinyOdom_Hardware-Aware_Efficient_Neural_Inertial_Navigation?fbclid=IwAR3F5LhoDiXD6tDhyE2PLFDB1hgy0IBM6V5YIUFwva7TvUvHYDi7C0ryTB8) |[`[code]`](https:\u002F\u002Fgithub.com\u002Fnesl\u002Ftinyodom?fbclid=IwAR1zbqrymVPxsVRHT6LZOtcwjJtzUYqfd0E8ChliklUvug-D6KKhWAAZ3dg)\n- Searching for Efficient Neural Architectures for On-Device ML on Edge TPUs |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2204.14007.pdf)\n- Green Accelerated Hoeffding Tree |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2205.03184.pdf)\n- tinyRadar: mmWave Radar based Human Activity Classification for Edge Computing |[`[pdf]`](https:\u002F\u002Flabs.dese.iisc.ac.in\u002Fneuronics\u002Fwp-content\u002Fuploads\u002Fsites\u002F16\u002F2022\u002F04\u002FmmWave_Radar_ver1.3.5.pdf)\n- MACHINE LEARNING SENSORS |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.03266.pdf)\n- Evaluating Short-Term Forecasting of Multiple Time Series in IoT Environments |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.07784.pdf)\n- How to train accurate BNNs for embedded systems? |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.12322.pdf)\n- Vildehaye: A Family of Versatile, Widely-Applicable, and Field-Proven Lightweight Wildlife Tracking and Sensing Tags  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.06171.pdf)\n- On-Device Training Under 256KB Memory  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.15472.pdf)\n- DEPTH PRUNING WITH AUXILIARY NETWORKS FOR TINYML  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2204.10546.pdf)\n- **[EdgeNeXt]**: Efficiently Amalgamated CNN-Transformer Architecture for Mobile Vision Applications |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.10589.pdf)\n- Tiny Robot Learning: Challenges and Directions for Machine Learning in Resource-Constrained Robots |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2205.05748.pdf)\n- **[POET]**: Training Neural Networks on Tiny Devices with Integrated Rematerialization and PagingPOET: Training Neural Networks on Tiny Devices |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2207.07697.pdf)\n- Two-stage Human Activity Recognition on\nMicrocontrollers with Decision Trees and CNNs  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.07652.pdf)\n- How to Manage Tiny Machine Learning at Scale – An Industrial Perspective |[`[pdf`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2202.09113.pdf)\n- **[SeLoC-ML]**: Semantic Low-Code Engineering for Machine Learning Applications in Industrial IoT|[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2207.08818.pdf)\n- **[IMU2Doppler]**: Cross-Modal Domain Adaptation for Doppler-based Activity Recognition Using IMU Data\" |[`[pdf]\n`](https:\u002F\u002Fsmashlab.io\u002Fpdfs\u002Fimu2dop.pdf)\n- **[Tiny-HR]**: Towards an interpretable machine learning\npipeline for heart rate estimation on edge devices  |[`[pdf]\n`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2208.07981.pdf)\n- [Enabling Fast Deep Learning on Tiny Energy-Harvesting IoT Devices]|[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2111.14051.pdf)\n- Extremely Simple Activation Shaping for\nOut-of-Distribution Detection |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2209.09858.pdf)\n- A processing‑in‑pixel‑in‑memory paradigm for resource‑constrained TinyML applications  |[`[pdf]`](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41598-022-17934-1.pdf)\n- **[tinySNN]**: Towards Memory- and Energy-Efficient Spiking Neural Networks |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.08656.pdf)\n- **[DeepPicarMicro]**: Applying TinyML to Autonomous Cyber Physical Systems |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2208.11212.pdf)\n- Incremental Online Learning Algorithms Comparison for Gesture and Visual Smart Sensors |[`[pdf`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2209.00591.pdf)\n-**[Protean]**: An Energy-Efficient and Heterogeneous Platform for\nAdaptive and Hardware-Accelerated Battery-free Computing |[`[pdf`](https:\u002F\u002Fdl.acm.org\u002Fdoi\u002Fpdf\u002F10.1145\u002F3560905.3568561?utm_medium=email&_hsmi=249309419&_hsenc=p2ANqtz--_ltviIYSpzfz9c4PiqHChBsEQkRDbr6treGEpYyBsHcwC5HX_R7JMp4ldGoydUfIlR-bOB-V2lKC0RIAhOcFen7daog&utm_content=249309419&utm_source=hs_email)\n- IN-SENSOR & NEUROMORPHIC COMPUTING ARE ALL YOU NEED FOR ENERGY\nEFFICIENT COMPUTER VISION  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2212.10881.pdf)\n- Energy Efficient Hardware Acceleration of\nNeural Networks with Power-of-Two\nQuantisation |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2209.15257.pdf)\n- Enabling ISP-less Low-Power Computer Vision |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2210.05451.pdf)\n- Rethinking Vision Transformers for MobileNet Size and Speed |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2212.08059.pdf)\n- Neuromorphic Computing and Sensing in Space |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2212.05236.pdf)\n- Joint Data Deepening-and-Prefetching for Energy-Efficient Edge Learning |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2211.07146.pdf)\n- PreMa: Predictive Maintenance of Solenoid Valve in Real-Time at Embedded Edge-Level | [`pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2211.12326.pdf)\n\n  | ▲ [Top](#tinyml-papers-and-projects) |\n  | ------------------------------------ |\n\n### \u003Cins>**2023**\u003C\u002Fins>\n\n- **[Coral NPU]**: A machine learning accelerator core designed for energy-efficient AI at the edge.  |[`[code]`](https:\u002F\u002Fgithub.com\u002Fgoogle-coral\u002Fcoralnpu)\n- **[cpp-transformer]**: A C++ implementation of Transformer without special library dependencies, including training and inference. |[`[code]`](https:\u002F\u002Fgithub.com\u002Ffreelw\u002Fcpp-transformer)\n- Exploring Automatic Gym Workouts Recognition Locally On Wearable Resource-Constrained Devices |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2301.05748.pdf)\n- **[MetaLDC]**: Meta Learning of Low-Dimensional Computing Classifiers for Fast On-Device Adaption |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2302.12347.pdf)\n- Faster Attention Is What You Need: A Fast\nSelf-Attention Neural Network Backbone\nArchitecture for the Edge via Double-Condensing\nAttention Condensers |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2208.06980.pdf)\n- **[TinyReptile]**: TinyML with Federated Meta-Learning |[`[pdf`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2304.05201.pdf)\n- **[TinyProp]** - Adaptive Sparse Backpropagation for Efficient TinyML On-device Learning |[`[pdf`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2308.09201.pdf)\n\n- **[LiteTrack]** - Layer Pruning with Asynchronous Feature Extraction\nfor Lightweight and Efficient Visual Tracking - Adaptive Sparse Backpropagation for Efficient TinyML On-device Learning |[`[pdf`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2409.00608v1)\n\n- **[MCUFormer]** - Deploying Vision Transformers on Microcontrollers with Limited Memory |[`[pdf`](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.16898)\n\n### \u003Cins>**2024**\u003C\u002Fins>\n\n- Model Compression in Practice: Lessons Learned from Practitioners Creating On-device Machine Learning Experiences | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2310.04621v2)\n- TinyAgent: Function Calling at the Edge | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2409.00608v1)\n- **[SENSORLLM]**: ALIGNING LARGE LANGUAGE MODELS WITH MOTION SENSORS FOR HUMAN ACTIVITY\nRECOGNITION | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2410.10624)\n- **[Penetrative AI]**: Making LLMs Comprehend the Physical World | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2310.09605v2)\n- **[MobileCLIP]**: Fast Image-Text Models through Multi-Modal Reinforced Training  | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2311.17049)\n- **[Zero-TPrune]**: Zero-Shot Token Pruning through Leveraging of the Attention\nGraph in Pre-Trained Transformers  | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2305.17328)\n- Towards Edge General Intelligence via Large Language Models: Opportunities and Challenges | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2410.18125)\n- **[TinyTTA]**: Efficient Test-time Adaptation via Early-exit Ensembles on Edge Devices | [`[pdf]`](https:\u002F\u002Fproceedings.neurips.cc\u002Fpaper_files\u002Fpaper\u002F2024\u002Ffile\u002F4c454d34f3a4c8d6b4ca85a918e5d7ba-Paper-Conference.pdf)\n\n\u003Cins>**2025**\u003C\u002Fins>\n\n- **[EdgeMark]**: An Automation and Benchmarking System for Embedded Artificial Intelligence Tools  | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2502.01700) | [`[official code]`](https:\u002F\u002Fgithub.com\u002FBlack3rror\u002FEdgeMark)\n- Optimizing Edge AI: A Comprehensive Survey on Data, Model, and System Strategies| [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2501.03265) | [`[GitHub]`](https:\u002F\u002Fgithub.com\u002Fwangxb96\u002FAwesome-EdgeAI)\n- Small Language Models are the Future of Agentic AI | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.02153) | [`[GitHub]`](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Flpr\u002Fslm-agents\u002F)\n- SensorLM: Learning the Language of Wearable Sensors | | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2506.09108)\n- A Comparative Survey of PyTorch vs TensorFlow for Deep Learning: Usability, Performance, and Deployment Trade-offss | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2508.04035v1)\n- **[MobileCLIP2]**: Improving Multi-Modal Reinforced Training  | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2508.20691) | [`[official code]`](https:\u002F\u002Fgithub.com\u002Fapple\u002Fml-mobileclip\u002Ftree\u002Fmain)\n- **[FastVLM]**: Efficient Vision Encoding for Vision Language Models | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2412.13303)\n- **[Flavors of Moonshine]**:Tiny Specialized ASR Models for Edge Devices | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2509.02523)\n- LLM Pruning and Distillation in Practice: The Minitron Approach | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2408.11796)\n- Real-Time Performance Benchmarking of TinyML Models in Embedded Systems (PICO: Performance of Inference, CPU, and Operations) | [`[pdf]`](https:\u002F\u002Fwww.arxiv.org\u002Fpdf\u002F2509.04721)\n- Small Language Models for Agentic Systems: A Survey of Architectures, Capabilities, and Deployment Trade-offs | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2510.03847)\n- AutoNeural: Co-Designing Vision–Language Models for NPU Inference | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2512.02924)\n\n  | ▲ [Top](#tinyml-papers-and-projects) |\n  | ------------------------------------ |\n\n## Awesome TinyML Projects\n\n### \u003Cins>**Projects Source code**\u003C\u002Fins>\n\n- TinyFederatedLearning | [`[official code]`](https:\u002F\u002Fgithub.com\u002Fkavyakvk\u002FTinyFederatedLearning) [`[presentation]`](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=KSaidr3ZN9M&feature=youtu.be) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fkavyakvk\u002FTinyFederatedLearning?style=social)\n- [TinyML Study Group](https:\u002F\u002Fgithub.com\u002Ftinyml-team\u002Fstudy-group) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Ftinyml-team\u002Fstudy-group?style=social)\n- [Arduino trash classification TinyML example](https:\u002F\u002Fgithub.com\u002Flightb0x\u002Farduino_trash_classification) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Flightb0x\u002Farduino_trash_classification?style=social)\n- [TinyML on Arduino](https:\u002F\u002Fgithub.com\u002Fsandeepmistry\u002Faimldevfest-workshop-2019) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fsandeepmistry\u002Faimldevfest-workshop-2019?style=social)\n- [Edge AI Anomaly Detection](https:\u002F\u002Fgithub.com\u002FShawnHymel\u002Ftinyml-example-anomaly-detection) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FShawnHymel\u002Ftinyml-example-anomaly-detection?style=social)\n- [Air Guitar CS249R](https:\u002F\u002Fgithub.com\u002FRobJMal\u002FAir-Guitar-CS249R) [`[presentation]`](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=PVk9RUW1Hwo) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FRobJMal\u002FAir-Guitar-CS249R?style=social)\n- [TinyML ESP32](https:\u002F\u002Fgithub.com\u002FHollowMan6\u002FTinyML-ESP32) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FHollowMan6\u002FTinyML-ESP32?style=social)\n- [MagicWand-TFLite-ESP32](https:\u002F\u002Fgithub.com\u002Fandriyadi\u002FMagicWand-TFLite-ESP32) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fandriyadi\u002FMagicWand-TFLite-ESP32?style=social)\n- [Localize your cat at home with BLE beacon, ESP32s, and Machine Learning](https:\u002F\u002Fgithub.com\u002FfilipsPL\u002Fcat-localizer) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FfilipsPL\u002Fcat-localizer?style=social)\n- [ESP32 Cam and Edge Impulse](https:\u002F\u002Fgithub.com\u002Fluisomoreau\u002FESP32-Cam-Edge-Impulse) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fluisomoreau\u002FESP32-Cam-Edge-Impulse?style=social)\n- [The C++ Neural Network and Machine Learning project](https:\u002F\u002Fgithub.com\u002Fintel\u002Fcppnnml) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fintel\u002Fcppnnml?style=social)\n- [Water Meter System Complete](https:\u002F\u002Fgithub.com\u002Fjomjol\u002Fwater-meter-system-complete) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fjomjol\u002Fwater-meter-system-complete?style=social)\n- [Number recognition with MNIST on Raspberry Pi Pico](https:\u002F\u002Fgithub.com\u002Fiwatake2222\u002Fpico-mnist) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fiwatake2222\u002Fpico-mnist?style=social)\n- [HallSensor RPM meter using Machine Learning](https:\u002F\u002Fgithub.com\u002FMiguelest07\u002FHallSensor_ML_EdgeImpulse) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FMiguelest07\u002FHallSensor_ML_EdgeImpulse?style=social)\n- [Weather forcasting with TinyML](https:\u002F\u002Fgithub.com\u002FBaptisteZloch\u002FWeather_forcasting) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FBaptisteZloch\u002FWeather_forcasting?style=social)\n- [TinyML using different frameworks applied to STM32F407 uC](https:\u002F\u002Fgithub.com\u002Ffjpolo\u002FSTM32F407TinyML) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Ffjpolo\u002FSTM32F407TinyML?style=social)\n- [CurrentSense-TinyML](https:\u002F\u002Fgithub.com\u002FSantandersecurityresearch\u002FCurrentSense-TinyML) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FSantandersecurityresearch\u002FCurrentSense-TinyML?style=social)\n- [Tensorflow Lite for Microcontrollers in Micropython](https:\u002F\u002Fgithub.com\u002Fmocleiri\u002Ftensorflow-micropython-examples) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fmocleiri\u002Ftensorflow-micropython-examples?style=social)\n- [TensorFlow Lite Micro for Espressif Chipsets](https:\u002F\u002Fgithub.com\u002Fespressif\u002Ftflite-micro-esp-examples) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fespressif\u002Ftflite-micro-esp-examples?style=social)\n- [ML Audio Classifier Example for Pico](https:\u002F\u002Fgithub.com\u002FArmDeveloperEcosystem\u002Fml-audio-classifier-example-for-pico) ![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FArmDeveloperEcosystem\u002Fml-audio-classifier-example-for-pico?style=social)\n- [Handwritten digit classification using Raspberry Pi Pico and Machine Learning](https:\u002F\u002Fgithub.com\u002Fcode2k13\u002Frpipico_digit_classification)![GitHub stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fcode2k13\u002Frpipico_digit_classification?style=social)\n\n### \u003Cins>**Projects Articles**\u003C\u002Fins>\n\n- `2020-09` [Autonomous embedded driving using computer vision](https:\u002F\u002Fwww.edgeimpulse.com\u002Fblog\u002Fautonomous-driving-using-computer-vision)\n- `2020-10` [EleTect - TinyML and IoT Based Smart Wildlife Tracker](https:\u002F\u002Fwww.hackster.io\u002Fdhruvsheth_\u002Feletect-tinyml-and-iot-based-smart-wildlife-tracker-c03e5a)\n- `2020-03` [Handwriting Recognition](https:\u002F\u002Fwww.hackster.io\u002Fnaveenbskumar\u002Fhandwriting-recognition-7583e3)\n- `2021-01` [Why Benchmarking TinyML Systems Is Challenging](https:\u002F\u002Fanalyticsindiamag.com\u002Fwhy-benchmarking-tinyml-systems-is-challenging\u002F)\n- `2021-01` [Build your own Google Assistant using tinyML](https:\u002F\u002Fmjrobot.org\u002F2021\u002F01\u002F27\u002Fbuilding-an-intelligent-voice-assistant-from-scratch\u002F)\n- `2021-02` [Fall detection and heart rate monitoring using AVR-IoT](https:\u002F\u002Fwww.hackster.io\u002Fnaveenbskumar\u002Ffall-detection-and-heart-rate-monitoring-using-avr-iot-75fb16)\n- `2021-02` [The Maker Show: TinyML for wildlife conservation](https:\u002F\u002Fdev.to\u002Ffordevs-community\u002Fthe-maker-show-tinyml-for-wildlife-conservation-idg)\n- `2021-05` [Under $100 and Less Than 1mW: Pneumonia Detection Solution for Everyone](https:\u002F\u002Fwww.edgeimpulse.com\u002Fblog\u002Funder-dollar100-and-less-than-1mw-pneumonia-detection-solution-for-everyone)\n- `2021-06` [Early Pigs' Respiratory Disease Detection Using Edge Impulse](https:\u002F\u002Fwww.hackster.io\u002Fclinton_oduor\u002Fearly-pigs-respiratory-disease-detection-using-edge-impulse-2ab039)\n- `2021-06` [Posture Watchdog](https:\u002F\u002Fwww.hackster.io\u002Fnaveenbskumar\u002Fposture-watchdog-c03f77?utm_campaign=Advanced%20Wearables%20Contest%20Hackster.io&utm_source=twitter&utm_medium=social&utm_content=Dream%20Smart%20Wearables%20winner:%20posture%20watchdog)\n- `2021-07` [Localized Environmental Sensing With TinyML](https:\u002F\u002Fhighdemandskills.com\u002Flocalized-monitoring-tinyml\u002F)\n- `2021` [Wireless Quarter: Edge Intelligence](https:\u002F\u002Fwww.nordicsemi.com\u002F-\u002Fmedia\u002FPublications\u002FWireless-Quarter-pdf\u002F2021\u002FWQ_Issue_2_2021.pdf?la=en&hash=A58D1AB12248E18E465658CE3CDFE33F9187692F#page=8)\n- [Arduino Machine Learning: Build a Tensorflow lite model to control robot-car](https:\u002F\u002Fwww.survivingwithandroid.com\u002Farduino-machine-learning-tensorflow-lite\u002F)\n- [TinyML ESP32-CAM: Edge Image classification with Edge Impulse](https:\u002F\u002Fwww.survivingwithandroid.com\u002Ftinyml-esp32-cam-edge-image-classification-with-edge-impulse\u002F)\n- [Predictive Maintenance with TinyAutomator](https:\u002F\u002Fwww.waylay.io\u002Farticles\u002Fpredictive-maintenance-with-tinyautomator)\n- [TinyML Person Detection with Arduino and Arducam](https:\u002F\u002Fwww.thetinymlbook.com\u002Fresources\u002Ftinyml-person-detection)\n- [Object Detection and Spatial Understanding with VLMs ft. Qwen2.5-VL](https:\u002F\u002Flearnopencv.com\u002Fobject-detection-with-vlms-ft-qwen2-5-vl\u002F)\n- [VLM on Edge: Worth the Hype or Just a Novelty?](https:\u002F\u002Flearnopencv.com\u002Fvlm-on-edge-devices\u002F)\n\n| ▲ [Top](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## Benchmarking and Others\n\n- [EEMBCs EnergyRunner](https:\u002F\u002Fgithub.com\u002Feembc\u002Fenergyrunner): The EEMBC EnergyRunner application framework for the MLPerf Tiny benchmark.\n- [MLPerf - Tiny](https:\u002F\u002Fmlcommons.org\u002Fen\u002Finference-tiny-05\u002F): is an ML benchmark suite for extremely low-power systems such as microcontrollers. [`[GitHub]`](https:\u002F\u002Fgithub.com\u002Fmlcommons\u002Ftiny\u002Ftree\u002Fv0.5)\n- [FedML](https:\u002F\u002Ffedml.ai\u002F): A Research Library and Benchmark for Federated Machine Learning. [`[GitHub]`](https:\u002F\u002Fgithub.com\u002FFedML-AI\u002FFedML)\n- [FogML](https:\u002F\u002Fgithub.com\u002Ftszydlo\u002FFogML): A Research Library for source code generation of the inferencing functions for embedded devices [`[GitHub]`]()\n- [Benchmarking Machine Learning on the Edge](https:\u002F\u002Fgithub.com\u002Faallan\u002Fbenchmarking-ml-on-the-edge)\n\n| ▲ [Top](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## Books\n\n- `[2022-12]` **AI at the Edge** (D. Situnayake & J. Plunkett, 2022. O'Reilly): [`[Book]`](https:\u002F\u002Fwww.oreilly.com\u002Flibrary\u002Fview\u002Fai-at-the\u002F9781098120191\u002F)\n- `[2022-10]` **Machine Learning on Commodity Tiny Devices** (S. Guo & Q. Zhou, 2022. CRC Press): [`[Book]`](https:\u002F\u002Fwww.routledge.com\u002FMachine-Learning-on-Commodity-Tiny-Devices-Theory-and-Practice\u002FGuo-Zhou\u002Fp\u002Fbook\u002F9781032374239)\n- `[2022-07]` **Introduction to TinyML** (Rohit Sharma, 2022, AITS): [`[Book]`](https:\u002F\u002Fwww.thetinymlbook.com\u002F) | [`[GitHub]`](https:\u002F\u002Fgithub.com\u002Fai-techsystems\u002FdeepC)\n- `[2022-04]` **TinyML Cookbook** (Gian Marco Iodice, 2022. Packt): [`[Book]`](https:\u002F\u002Fwww.packtpub.com\u002Fproduct\u002Ftinyml-cookbook\u002F9781801814973) | [`[GitHub]`](https:\u002F\u002Fgithub.com\u002FPacktPublishing\u002FTinyML-Cookbook)\n- `[2021-03]` **Artificial Intelligence for IoT Cookbook** (Michael Roshak, 2021. Packt): [`[Book]`](https:\u002F\u002Fwww.packtpub.com\u002Fproduct\u002Fartificial-intelligence-for-iot-cookbook\u002F9781838981983) | [`[GitHub]`](https:\u002F\u002Fgithub.com\u002FPacktPublishing\u002FArtificial-Intelligence-for-IoT-Cookbook)\n- `[2020-04]` **Mobile Deep Learning with TensorFlow Lite, ML Kit and Flutter**: Build scalable real-world projects to implement end-to-end neural networks on Android and iOS (Anubhav Singh, Rimjhim Bhadani, 2020. Packt): [`[Book]`](https:\u002F\u002Fwww.amazon.com\u002FMobile-Deep-Learning-TensorFlow-Flutter\u002Fdp\u002F1789611210)\n- `[2020-01]` **TinyML: Machine Learning with TensorFlow Lite on Arduino and Ultra-Low-Power Microcontrollers** (Pete Warden. O'Reilly Media): [`[Book]`](https:\u002F\u002Fwww.amazon.com\u002FTinyML-Learning-TensorFlow-Ultra-Low-Power-Microcontrollers\u002Fdp\u002F1492052043)\n\n| ▲ [Top](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## Articles\n\n- `2019-12` [TinyML as-a-Service: What is it and what does it mean for the IoT Edge?](https:\u002F\u002Fwww.ericsson.com\u002Fen\u002Fblog\u002F2019\u002F12\u002Ftinyml-as-a-service-iot-edge)\n- `2019-12` [TinyML as a Service and the challenges of machine learning at the edge](https:\u002F\u002Fwww.ericsson.com\u002Fen\u002Fblog\u002F2019\u002F12\u002Ftinyml-as-a-service)\n- `2020-05` [Model Quantization Using TensorFlow Lite](https:\u002F\u002Fmedium.com\u002Fsclable\u002Fmodel-quantization-using-tensorflow-lite-2fe6a171a90d)\n- `2020-09` [TinyML is breathing life into billions of devices](https:\u002F\u002Fthenextweb.com\u002Fneural\u002F2020\u002F09\u002F03\u002Ftinyml-is-breathing-life-into-billions-of-devices\u002F)\n- `2020-12` [Predictions for Embedded Machine Learning for IoT in 2021](https:\u002F\u002Fwww.iotworldtoday.com\u002F2020\u002F12\u002F10\u002Fpredictions-for-embedded-machine-learning-for-iot-in-2021\u002F)\n- `2020-12` [Matthew Mattina: Life-Saving Models in Your Pocket](https:\u002F\u002Fread.deeplearning.ai\u002Fthe-batch\u002Fissue-72\u002F)\n- `2020-12` [Tiny four-bit computers are now all you need to train AI](https:\u002F\u002Fwww.technologyreview.com\u002F2020\u002F12\u002F11\u002F1014102\u002Fai-trains-on-4-bit-computers\u002F)\n- `2021-01` [How predictive maintenance is changing the industrial enterprise for good](https:\u002F\u002Ftechhq.com\u002F2021\u002F01\u002Fhow-predictive-maintenance-is-changing-the-industrial-enterprise-for-good\u002F)\n- `2021-02` [What is TinyML?](https:\u002F\u002Fwww.fierceelectronics.com\u002Felectronics\u002Fwhat-tinyml)\n- `2021-02` [How AI is Taking on Sensors](https:\u002F\u002Fwww.electropages.com\u002Fblog\u002F2021\u002F02\u002Fhow-ai-taking-sensors)\n- `2021-04` [MLCommons™ Releases MLPerf™ Inference v1.0 Results with First Power Measurements](https:\u002F\u002Fmlcommons.org\u002Fen\u002Fnews\u002Fmlperf-inference-v10\u002F)\n- `2021-05` [TapLock - A bike lock with machine learning](https:\u002F\u002Fwww.hackster.io\u002Ftaplock\u002Ftaplock-a-bike-lock-with-machine-learning-85641c)\n- `2021-05` [Taking Back Control](https:\u002F\u002Fwww.hackster.io\u002Fnews\u002Ftaking-back-control-14068dbb0bb7?fbclid=IwAR0QGucom06pzd7K5SJIdYByZr67xd29YlqTdbnK78OU7GqW540vJPeD534)\n- `2021-06` [Neural network architectures for deploying TinyML applications on commodity microcontrollers](https:\u002F\u002Fcommunity.arm.com\u002Fdeveloper\u002Fresearch\u002Fb\u002Farticles\u002Fposts\u002Fneural-network-architectures-for-deploying-tinyml-applications-on-commodity-microcontrollers)\n- `2021-06` [TinyML in MicroCosmos](https:\u002F\u002Fwww.hackster.io\u002FCHA_RAN\u002Ftinyml-in-microcosmos-c1161c)\n- `2021-06` [‘Small Data’ Are Also Crucial for Machine Learning](https:\u002F\u002Fwww.hackster.io\u002FCHA_RAN\u002Ftinyml-in-microcosmos-c1161c)\n- `2021-07` [A natively flexible 32-bit Arm microprocessor](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41586-021-03625-w)\n- `2021-07` [Wearable Devices Can Reduce Collision Risk in Blind and Visually Impaired People](https:\u002F\u002Fmasseyeandear.org\u002Fnews\u002Fpress-releases\u002F2021\u002F07\u002Fwearable-devices-can-reduce-collision-risk-in-blind-and-visually-impaired-people)\n- `2021-09` [AI Inspection Using Analog Gauge as an Example](https:\u002F\u002Findatalabs.com\u002Fblog\u002Fai-inspection)\n- `2025-10` [Top Small Language Models for Agentic AI Solutions Development](https:\u002F\u002Fthirdeyedata.ai\u002Ftop-small-language-models-for-agentic-ai-solutions-development\u002F)\n\n| ▲ [Top](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## Libraries and Tools\n\n- [Edge Impulse](https:\u002F\u002Fwww.edgeimpulse.com\u002F) - Interactive platform to generate models that can run in microcontrollers. They are also quite active on social netwoks talking about recent news on EdgeAI\u002FTinyML.\n- [EVE is Edge Virtualization Engine](https:\u002F\u002Fgithub.com\u002Flf-edge\u002Feve\u002Fblob\u002Fmaster\u002FREADME.md)\n- [microTVM](https:\u002F\u002Ftvm.apache.org\u002Fdocs\u002Fmicrotvm\u002Findex.html) - is an open source tool to optimize tensor programs.\n- [Larq](https:\u002F\u002Fgithub.com\u002Flarq\u002Flarq) - An Open-Source Library for Training Binarized Neural Networks.\n- [Neural Network on Microcontroller (NNoM)](https:\u002F\u002Fgithub.com\u002Fmajianjia\u002Fnnom) - Higher-level layer-based Neural Network library specifically for microcontrollers. Support for CMSIS-NN.\n- [BerryNet](https:\u002F\u002Fgithub.com\u002FDT42\u002FBerryNet) - Deep learning gateway on Raspberry Pi and other edge devices.\n- [Rune](https:\u002F\u002Fgithub.com\u002Fhotg-ai\u002Frune) - provides containers to encapsulate and deploy edgeML pipelines and applications.\n- [Onnxruntime](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fonnxruntime) - cross-platform, high performance ML inferencing and training accelerator.\n- [deepC](https:\u002F\u002Fgithub.com\u002Fai-techsystems\u002FdeepC) - vendor independent TinyML deep learning library, compiler and inference framework microcomputers and micro-controllers\n- [deepC for Arduino](https:\u002F\u002Fgithub.com\u002Fai-techsystems\u002Farduino) - TinyML deep learning library customized for Arduiono IDE\n- [emlearn](https:\u002F\u002Fgithub.com\u002Femlearn\u002Femlearn) - Machine learning for microcontroller and embedded systems. Train in Python, then do inference on any device with a C99 compiler.\n\n| ▲ [Top](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## Courses\n\n- **11-767: On-Device Machine Learning Fall** - by CMU | [`[website]`](https:\u002F\u002Fcmu-odml.github.io)\n- **TinyML4D: UNIFEI-IESTI01-TinyML-2023.1** - by UNIFEI | [`[website]`](https:\u002F\u002Fgithub.com\u002FMjrovai\u002FUNIFEI-IESTI01-TinyML-2023.1)\n- **Introduction to Embedded Deep Learning** - by CMU | [`[website]`](https:\u002F\u002Fz4ziad.github.io\u002Fembed-dl-s23\u002FEmbeddedDL_S23\u002F)\n- **TinyML and Efficient Deep Learning** - by MIT | [`[website]`](https:\u002F\u002Fefficientml.ai\u002F)\n- **Machine Learning at the Edge on Arm: A Practical Introduction** - by  ARM | [`[edx]`](https:\u002F\u002Fwww.edx.org\u002Fcourse\u002Fmachine-learning-at-the-edge-a-practical-introduction-from-arm)\n- **CS249r: Tiny Machine Learning (TinyML)** - *Harvard University* by Vijay Janapa Reddi: [sites.google.com](https:\u002F\u002Fsites.google.com\u002Fg.harvard.edu\u002Ftinyml\u002Fhome?authuser=0) | [`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fchannel\u002FUCLv1K6OaYHP44hXFd5rNqyA) | [`[edx]`](https:\u002F\u002Fwww.edx.org\u002Fprofessional-certificate\u002Fharvardx-tiny-machine-learning) | [`[GitHub]`](https:\u002F\u002Fgithub.com\u002FtinyMLx\u002Fcolabs)\n- **MLOps for Scaling TinyML** - *Harvard University* by Vijay Janapa Reddi: [`[edX]`](https:\u002F\u002Fwww.edx.org\u002Fcourse\u002Fmlops-for-scaling-tinyml)\n- **Introduction to Embedded Machine Learning** - *Edge Impulse* by Shawn Hymel: [`[Coursera]`](https:\u002F\u002Fwww.coursera.org\u002Flearn\u002Fintroduction-to-embedded-machine-learning)\n- **Embedded and Distributed AI** - *Jonkoping University, Sweden* by  Beril Sirmacek: [`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=OTXqT00MmPA&list=PLyulI6o7oOtycIT15i_I2_mhuLxnNvPvX)\n- **MLT Artificial Intelligence - EdgeAI** - Machine Learning Tokyo: [`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fplaylist?list=PLaPdEEY26UXxrxn-82sqe9cTTvWC0y-rM)\n- TinyML @ UPenn | [`[website]`](https:\u002F\u002Fgithub.com\u002FMjrovai\u002FUNIFEI-IESTI01-TinyML-2023.1) | [`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fplaylist?list=PL7rtKJAz_mPe6kAbiH6Ucq02Vpa95qvBJ)\n\n| ▲ [Top](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## TinyML Talks & Conferences\n\n- TinyML Talks, Summit & Research Symposium: [`[Website]`](https:\u002F\u002Fconf.researchr.org\u002Fseries\u002Ftinyml-symp) | [`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fchannel\u002FUC9iWqvsWjhowkHWVJquHwkg)\n- Embedded Vision Summit - [Edge AI & Vision Alliance](https:\u002F\u002Fwww.edge-ai-vision.com\u002F): [`[Website]`](https:\u002F\u002Fembeddedvisionsummit.com) | [`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FEdgeAIandVisionAlliance)\n- Low-Power Computer Vision Challenge (LPCV): [`[Website]`](https:\u002F\u002Flpcv.ai) | [`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fchannel\u002FUCAeAbQsRUZ8KWmGUKejtgBg)\n\n|                                                     Title                                                     |    Speaker    | Published Date |                                                  Link                                                   |\n| :-----------------------------------------------------------------------------------------------------------: | :-----------: | :------------: | :-----------------------------------------------------------------------------------------------------: |\n|           [Challenges for Large Scale Deployment of Tiny ML Devices](https:\u002F\u002Fyoutu.be\u002FbwjHLrLGkOY)            |  G. Raghavan  |   2022-04-29   |            [slide](https:\u002F\u002Fcms.tinyml.org\u002Fwp-content\u002Fuploads\u002Fsummit2022\u002FRaghavan-Gopal.pdf)             |\n|            [Building data-centric AI tooling for embedded engineers](https:\u002F\u002Fyoutu.be\u002F9rnzM-C7QdA)            | D. Situnayake |   2022-04-29   |           [slide](https:\u002F\u002Fcms.tinyml.org\u002Fwp-content\u002Fuploads\u002Fsummit2022\u002FSitunayake-Daniel.pdf)           |\n|                    [Sensors and ML: waking smarter for less](https:\u002F\u002Fyoutu.be\u002FVXpQlOouBqU)                    |   A. Ataya    |   2022-05-04   |              [slide](https:\u002F\u002Fcms.tinyml.org\u002Fwp-content\u002Fuploads\u002Fsummit2022\u002FAbbas-Ataya.pdf)              |\n| [MLOps for TinyML: Challenges & Directions in Operationalizing TinyML at Scale](https:\u002F\u002Fyoutu.be\u002FyydnTSH0R2I) |  V.J. Reddi   |   2022-05-24   | [slide](https:\u002F\u002Fcms.tinyml.org\u002Fwp-content\u002Fuploads\u002Ftalks2022\u002FtinyML_Talks_Vijay_Janapa_Reddi_220524.pdf) |\n|              [Vibration Monitoring Machine Learning Demonstration](https:\u002F\u002Fyoutu.be\u002F2iInOo0Lkfs)              |  J. Edwards   |   2020-12-22   |                            [github](https:\u002F\u002Fgithub.com\u002FNumerix-DSP\u002Fsiglib\u002F)                             |\n|     [Moving From AI To IntelligentAI To Reduce The Cost Of AI At The Edge](https:\u002F\u002Fyoutu.be\u002FmYy4Zv80tXQ)      |  J. Edwards   |   2020-12-22   |                                 [web](https:\u002F\u002Fwww.numerix-dsp.com\u002Fai\u002F)                                  |\n\n| ▲ [Top](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## Competitions\n\n- **[LPCV]**: Low-Power Computer Vision Challenge |[`[website]`](https:\u002F\u002Flpcv.ai\u002F)\n\n## Other Awesome Repos\n\n- [Awesome Human Activity Recognition](https:\u002F\u002Fgithub.com\u002FJie-su\u002FAwesome_Human_Activity_Recognition#2-Paper-with-code)\n\n## Contact & Feedback\n\nIf you have any suggestions about TinyML papers and projects, feel free to mail me :)\n\n- [e-mail](mailto:gigwegbe@gmail.com)\n- [pull request](https:\u002F\u002Fgithub.com\u002Fgigwegbe\u002Ftinyml-papers-and-projects\u002Fpulls)\n","## TinyML 论文与项目\n>\n> TinyML 太棒了。\n\n[![Awesome](https:\u002F\u002Fawesome.re\u002Fbadge.svg)](https:\u002F\u002Fawesome.re) [![贡献](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fissues-pr-closed-raw\u002Fgigwegbe\u002Ftinyml-papers-and-projects.svg?label=contributions)](https:\u002F\u002Fgithub.com\u002Fgigwegbe\u002Ftinyml-papers-and-projects\u002Fpulls) [![提交](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flast-commit\u002Fgigwegbe\u002Ftinyml-papers-and-projects.svg?label=last%20contribution)](https:\u002F\u002Fgithub.com\u002Fgigwegbe\u002Ftinyml-papers-and-projects\u002Fcommits\u002Fmain)\n\n这是一个关于 TinyML 的有趣论文、项目、文章和演讲的列表。\n\n- [精彩论文](#awesome-papers-): [2016](#2016) | [2017](#2017) | [2018](#2018) | [2019](#2019) | [2020](#2020) | [2022](#2022) | [2023](#2023) | [2024](#2024) | [2025](#2025)\n- [精彩项目](#awesome-tinyml-projects): [项目源代码](#projects-source-code) | [项目文章](#projects-articles)\n- [基准测试](#benchmarking)\n- 资源\n  - [文章](#articles)\n  - [书籍](#books)\n  - [库与工具](#libraries-and-tools)\n  - [课程](#courses)\n  - [TinyML 演讲与会议](#tinyml-talks--conferences)\n- [联系与反馈](#contact--feedback)\n\n## 精彩论文\n\n### \u003Cins>**2016**\u003C\u002Fins>\n\n- DEEP COMPRESSION: 使用剪枝、量化训练和霍夫曼编码压缩深度神经网络 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1510.00149.pdf)\n- **[SQUEEZENET]** 以比 AlexNet 少 50 倍的参数和小于 0.5MB 的模型大小实现 AlexNet 级别的准确率 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1602.07360.pdf)\n\n### \u003Cins>**2017**\u003C\u002Fins>\n\n- 用于仅使用整数运算高效推理的神经网络量化与训练 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1712.05877.pdf)\n- 物联网中仅需 2 KB 内存的资源高效机器学习 | [`[pdf]`](https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fresearch\u002Fuploads\u002Fprod\u002F2017\u002F06\u002Fkumar17.pdf)\n- ProtoNN: 面向资源匮乏设备的压缩且精确的 kNN | [`[pdf]`](https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fresearch\u002Fuploads\u002Fprod\u002F2017\u002F06\u002Fprotonn.pdf)\n- OPENMV: 一款由 Python 驱动、可扩展的机器视觉相机 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1711.10464.pdf) [`[官方代码]`]( https:\u002F\u002Fgithub.com\u002Fopenmv\u002Fopenmv.git)\n\n### \u003Cins>**2018**\u003C\u002Fins>\n\n- **[AMC]** 用于移动设备上模型压缩与加速的 AutoML | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1802.03494.pdf) [`[官方代码]`](https:\u002F\u002Fgithub.com\u002Fmit-han-lab\u002Famc)\n- ARM 上的移动机器学习硬件：片上系统（SoC）视角 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1801.06274.pdf)\n- **[HAQ]** 具有混合精度的硬件感知自动化量化 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fabs\u002F1811.08886)\n- 面向真实世界系统的高效且鲁本的机器学习 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1812.02240.pdf)\n- **[GesturePod]** 针对视障人士的手势交互手杖 | [`[pdf]`](https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fresearch\u002Fuploads\u002Fprod\u002F2018\u002F05\u002FCHI19_GesturePod.pdf)\n- **[YOLO-LITE]** 针对非 GPU 计算机优化的实时目标检测算法 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1811.05588v1.pdf)\n- **[CMSIS-NN]** 适用于 Arm Cortex-M CPU 的高效神经网络内核 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1801.06601.pdf)\n- 为高效推理而对深度卷积网络进行量化：白皮书 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1806.08342.pdf)\n- **[Hello Edge]** 微控制器上的关键词检测 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1711.07128.pdf)\n\n  | ▲ [返回顶部](#tinyml-papers-and-projects) |\n  | ------------------------------------ |\n\n### \u003Cins>**2019**\u003C\u002Fins>\n\n- FastGRNN: 一种快速、准确、稳定且体积小巧的千字节级门控循环神经网络 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1901.02358.pdf)\n- 物联网边缘设备上的图像分类：性能分析与建模 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1902.11119.pdf)\n- **[PROXYLESSNAS]** 直接在目标任务和硬件上进行神经架构搜索 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1812.00332.pdf) [`[官方代码]`](https:\u002F\u002Fgithub.com\u002Fmit-han-lab\u002Fproxylessnas)\n- 通过迁移学习实现设备端 CNN 推理的节能硬件 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1812.01672.pdf)\n- Visual Wake Words 数据集 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1906.05721.pdf)\n- 将 KB 级机器学习模型编译到微型物联网设备 | [`[pdf]`](microsoft.com\u002Fen-us\u002Fresearch\u002Fuploads\u002Fprod\u002F2018\u002F10\u002Fpldi19-SeeDot.pdf)\n- 可重构的多任务音频动态处理方案 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fabs\u002F1903.06392 )\n- 推进 RNN 压缩的极限 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1910.02558.pdf)\n- 一种低功耗的端到端混合神经形态框架，用于监控应用 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1910.09806.pdf)\n- 边缘计算中的深度学习 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1910.10231.pdf)\n- 基于内存驱动的混合低精度量化，以在微控制器上实现深度网络推理 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1905.13082.pdf) [`[官方代码]`](https:\u002F\u002Fgithub.com\u002FEEESlab\u002FCMix-NN)\n- **[SpArSe]** 面向资源受限微控制器的 CNN 稀疏架构搜索 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1905.12107.pdf)\n- **[MobileNetV2]** 反转残差与线性瓶颈 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1801.04381.pdf)\n- 隐式权重并不存在：重新思考二值化神经网络优化 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1906.02107.pdf)\n- 低功耗计算机视觉：现状、挑战与机遇 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1904.07714.pdf)\n\n  | ▲ [返回顶部](#tinyml-papers-and-projects) |\n  | ------------------------------------ |\n\n### \u003Cins>**2020**\u003C\u002Fins>\n\n- 使用克罗内克积将 RNN 在物联网设备上压缩 15 至 38 倍 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1906.02876.pdf)\n- 轻量级机器学习系统的基准测试：挑战与方向 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2003.04821v3.pdf)\n- 具有长短程注意力的轻量级 Transformer |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2004.11886.pdf)\n- **[FANN-on-MCU]** 用于物联网边缘端节能神经网络推理的开源工具包 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1911.03314.pdf)\n- **[TENSORFLOW LITE MICRO]** 轻量级机器学习系统上的嵌入式机器学习 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2010.08678v2.pdf)\n- **[AttendNets]** 通过视觉注意力浓缩器实现的面向边缘的微型深度图像识别神经网络 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2009.14385v1.pdf)\n- **[TinySpeech]** 用于边缘设备上深度语音识别神经网络的注意力浓缩器 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2008.04245v6.pdf)\n- 使用轻量级机器学习实现自主小型车辆的鲁棒导航 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2007.00302v1.pdf) [`[官方代码]`](https:\u002F\u002Fgithub.com\u002Fpraesc\u002FRobust-navigation-with-TinyML)\n- **[MICRONETS]** 用于在通用微控制器上部署轻量级机器学习应用的神经网络架构 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2010.11267v2.pdf)\n- **[TinyLSTMs]** 面向助听器的高效神经语音增强 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2005.11138.pdf)\n- **[MCUNet]** 物联网设备上的微型深度学习 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fabs\u002F2007.10319) [`[官方代码]`](https:\u002F\u002Fgithub.com\u002Fmit-han-lab\u002Fmcunet)\n- 基于残数数系的高效 Winograd 卷积 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2007.12216.pdf)\n- 深度学习推理中的整数量化：原理与经验评估  | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2004.09602)\n- 关于唤醒词检测前端增益不变建模的研究 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2010.06676.pdf)\n- 向数据高效的唤醒词检测建模迈进 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2010.06659.pdf)\n- 使用 CNN 精确检测唤醒词的起始与结束 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2008.03790.pdf)\n- **[PoPS]** 面向深度强化学习的策略剪枝与缩减 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2001.05012.pdf)\n- Howl：一个已部署的开源唤醒词检测系统 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2008.09606.pdf) [`[官方代码]`](https:\u002F\u002Fgithub.com\u002Fcastorini\u002Fhowl)\n- **[LeakyPick]** 物联网音频间谍探测器 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2007.00500.pdf)\n- 设备端机器学习：算法与学习理论视角  | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1911.00623.pdf)\n- 利用自动化混合低精度量化技术优化小型边缘微控制器 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2008.05124.pdf)\n- 优化关键指标：使用最终指标训练 DNN-HMM 关键字检测模型 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.01151.pdf)\n- **[RNNPool]** 面向内存受限推理的高效非线性池化 | [`[博客]`](https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fresearch\u002Fblog\u002Fseeing-on-tiny-battery-powered-microcontrollers-with-rnnpool\u002F?utm_medium=email&_hsmi=104017359&_hsenc=p2ANqtz-_DVkWnyh_NhAV6j4hTFngepUyiNjZ5GO5CYIQfpl5NzerjwxOBQcpdkilzGpt9ic4HglvgM80h7wIkFNX89xe-3_j7Kw&utm_content=104017359&utm_source=hs_email)  [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2002.11921.pdf) [`[官方代码]`](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002FEdgeML\u002Fblob\u002Fmaster\u002Fpytorch\u002Fedgeml_pytorch\u002Fgraph\u002Frnnpool.py)\n- **[Shiftry]** 在 2KB 内存中进行 RNN 推理 |[`[pdf]`](https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fresearch\u002Fuploads\u002Fprod\u002F2020\u002F10\u002Foopsla20main-p230-p-aba27a6-48263M-final.pdf)\n- **[Once for All]** 训练一个网络并针对高效部署进行特化 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1908.09791.pdf) [`[官方代码]`](https:\u002F\u002Fgithub.com\u002Fmit-han-lab\u002Fonce-for-all)\n- 面向资源受限终端的医疗口罩检测微型 CNN 架构  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.14858.pdf)\n- 重新思考极低内存占用边缘设备上的美国手语预测泛化问题 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.13741.pdf) [`[演示文稿]`](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=bJ1vnhAbJ9o&feature=youtu.be)\n- **[ShadowNet]** 一种安全高效的设备端模型推理系统 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.05905.pdf)\n- 面向通用及专用硬件的高效关键字检测硬件感知训练 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2009.04465.pdf)\n- 针对缺乏独特标记的野生动物的自动化面部识别：基于深度学习的棕熊识别方法 |[`[pdf]`](https:\u002F\u002Fonlinelibrary.wiley.com\u002Fdoi\u002Fepdf\u002F10.1002\u002Fece3.6840)\n- **[HyNNA]**：利用混合神经网络架构提升基于神经形态视觉传感器的监控性能 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2003.08603.pdf)\n- 硬件彩票 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2009.06489.pdf)\n- MLPerf 推理基准测试 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1911.02549.pdf)\n- MLPerf 移动推理基准测试：为什么移动 AI 基准测试困难以及如何应对 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2012.02328.pdf)\n- **[TinyRL]** 学习寻找：纳米四旋翼无人机上的微型机器人源点搜索学习 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1909.11236.pdf) [`[演示文稿]`](https:\u002F\u002Fyoutu.be\u002FwmVKbX7MOnU)\n- 使用 Microsoft 浮点数在云端规模下推动窄精度推理的极限 |[`[pdf]`](https:\u002F\u002Fproceedings.neurips.cc\u002F\u002Fpaper\u002F2020\u002Ffile\u002F747e32ab0fea7fbd2ad9ec03daa3f840-Paper.pdf)\n- **[TinyBERT]** 用于自然语言理解的 BERT 知识蒸馏 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F1909.10351.pdf)\n- **[Larq]** 一个用于训练二值化神经网络的开源库 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.09398.pdf) [`[演示文稿]`](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=f9SNqDejOB0) [`[官方代码]`](https:\u002F\u002Fgithub.com\u002Flarq\u002Flarq)\n- **[FedML]** 一个用于联邦学习的研究库和基准测试平台 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2007.13518.pdf)\n- 机器学习加速器综述  |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2009.00993.pdf)\n\n  | ▲ [返回顶部](#tinyml-papers-and-projects) |\n  | ------------------------------------ |\n\n### \u003Cins>**2021**\u003C\u002Fins>\n\n- **[I-BERT]** 仅整数的 BERT 量化 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2101.01321.pdf)\n- **[TinyTL]** 减少内存而非参数，实现高效的设备端学习 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2007.11622.pdf) [`[官方代码]`](https:\u002F\u002Fgithub.com\u002Fmit-han-lab\u002Ftinyml\u002Ftree\u002Fmaster\u002Ftinytl)\n- 循环神经网络的量化研究 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2101.05453.pdf)\n- **[TINY TRANSDUCER]** 一种在边缘设备上运行的高效率语音识别模型 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2101.06856.pdf)\n- LARQ 计算引擎：设计、基准测试及部署最先进的二值化神经网络 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.09398.pdf)\n- **[LEAF]** 用于音频分类的可学习前端 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2101.08596.pdf)\n- 通过交换技术在微型微控制器上运行大型神经网络 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2101.08744.pdf)\n- 用于嵌入式平台上量化推理的卷积神经网络定点量化 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.02147.pdf)\n- 通过低成本 BLE 设备估计室内人员占用情况 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.03351.pdf)\n- **[Tiny Eats]** 在微控制器上进行进食检测 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2003.06699.pdf)\n- **[DEVICETTS]** 一种占用空间小、速度快、稳定的设备端文本转语音网络 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2010.15311.pdf)\n- 基于 FPGA 的 0.57-GOPS\u002FDSP 目标检测 PIM 加速器 |[`[pdf]`](https:\u002F\u002Fdl.acm.org\u002Fdoi\u002Fpdf\u002F10.1145\u002F3394885.3431659)\n- 重新思考神经架构与硬件加速器的协同设计 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.08619.pdf)\n- 深度学习中的稀疏性：剪枝与生长在神经网络高效推理和训练中的应用 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.00554.pdf)\n- **[Apollo]** 可迁移的架构探索 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.01723.pdf)\n- 基于深度神经网络的咳嗽检测，利用床边加速度计测量数据 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.04997.pdf)\n- TapNet：一款用于屏幕外移动输入的多任务学习 CNN 的设计、训练、实现及应用 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.09087.pdf)\n- 智能设备上的内存高效语音识别 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2102.11531.pdf)\n- SWIS——共享权重位稀疏性，用于高效神经网络加速 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.01308.pdf)\n- 面向通用及专用硬件的高效关键词检测的硬件感知训练 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2009.04465.pdf)\n- 边缘设备上高效超维度计算的超向量设计 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.06709.pdf)\n- 软弱有时反而更坚韧：受节肢动物外骨骼启发的抗碰撞四旋翼无人机 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.04423.pdf)\n- **[TinyOL]** 微控制器上的在线学习 TinyML |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.08295.pdf)\n- 针对紧凑型 TinyML 模型的量化引导训练 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.06231.pdf)\n- hls4ml：一个开源联合设计工作流，助力科学低功耗机器学习设备 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.05579.pdf)\n- 基于超维度计算的内存高效、肢体位置感知手势识别 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.05267.pdf)\n- 动态可调节吞吐量的神经网络（TNN） |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.02836.pdf)\n- 硬件感知神经架构搜索综合综述 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2101.09336.pdf)\n- 用于非接触式呼吸频率监测的智能床传感器系统 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.13792.pdf)\n- 衡量真正重要的东西：为 TinyML 优化神经网络 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.10645.pdf)\n- 任意语言下的少量样本关键词检测 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.01454.pdf)\n- DOPING：一种利用稀疏结构化加法矩阵实现 LSTM 模型极端压缩的技术 |[`[pdf]`](https:\u002F\u002Fproceedings.mlsys.org\u002Fpaper\u002F2021\u002Ffile\u002Fa3f390d88e4c41f2747bfa2f1b5f87db-Paper.pdf)\n- **[OutlierNets]** 用于设备端声学异常检测的高度紧凑的深度自编码器网络架构 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.00528.pdf)\n- **[TENT]** 使用锥形定点实现边缘设备上神经网络的高效量化 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.02233.pdf)\n- 基于 1D-CNN 的深度学习技术，用于物联网传感器中的睡眠呼吸暂停检测 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2105.00528.pdf)\n- 适用于低功耗 CPU 的自适应测试时增强 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2105.06183.pdf)\n- 面向嵌入式平台深度学习工作负载的编译工具链 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.04576.pdf)\n- **[ProxiMic]** 通过单个麦克风检测近场语音实现便捷的语音激活功能 |[`[pdf]`](https:\u002F\u002Fdl.acm.org\u002Fdoi\u002Fpdf\u002F10.1145\u002F3411764.3445687)\n- **[Fusion-DHL]** WiFi、IMU 和楼层平面图融合，用于室内环境中的密集位置历史记录 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2105.08837.pdf)\n- **[µNAS]** 面向微控制器的约束神经架构搜索 |[`[pdf]`](https:\u002F\u002Fdl.acm.org\u002Fdoi\u002Fpdf\u002F10.1145\u002F3437984.3458836?utm_content=167905304&utm_medium=social&utm_source=linkedin&hss_channel=lcp-19239958)\n- 使用树莓派结合功率激光中和蚊子 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2105.14190.pdf)\n- 通过 TinyML 扩大应用机器学习的普及范围 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2106.04008.pdf)\n- 嵌入式系统中的机器学习应用 |[`[pdf]`](https:\u002F\u002Fwww.tiriasresearch.com\u002Fwp-content\u002Fuploads\u002F2021\u002F05\u002FUsing-Machine-Learning-in-Embedded-Systems.pdf)\n- **[FRILL]** 一种面向移动设备的非语义语音嵌入 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2011.04609.pdf)\n- 任意语言下的少量样本关键词检测 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.01454.pdf)\n- MLPerf Tiny 基准测试 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2106.07597.pdf)\n- 高效神经网络推理的量化方法综述 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2103.13630.pdf)\n- 高效深度学习：关于如何使深度学习模型更小、更快、更好的综述 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2106.08962.pdf)\n- AttendSeg：一种用于边缘语义分割的微型注意力浓缩神经网络 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.14623.pdf)\n- 神经网络训练中的随机性：工具对影响的表征 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2106.11872.pdf)\n- TinyML：ESP32 SoC 中 Xtensa LX6 微处理器在神经网络应用中的分析 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2106.10652.pdf)\n- **[Keyword Transformer]**：一种用于关键词检测的自注意力模型 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2104.00769.pdf)\n- LB-CNN：一个使用 Chainer 和 Cupy 快速训练轻量级二值卷积神经网络的开源框架 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2106.15350.pdf)\n- **[Only Train Once]**：一种一次性神经网络训练与剪枝框架 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2107.07467.pdf)\n- **[BEANNA]**：一种支持二值化的神经网络加速架构 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2108.02313.pdf)\n- 一个基于量化潜在重放的设备端持续学习 TinyML 平台 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2110.10486.pdf)\n- 利用机器学习技术和嵌入式传感器对异常步态的分类 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2110.06139.pdf)\n- **[MOBILEVIT]**：一种轻量级、通用且适合移动端的视觉 Transformer |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2110.02178.pdf?utm_medium=email&_hsmi=175723863&_hsenc=p2ANqtz--rVrB87u6eUrx_A5XM8m9kyJc-xTO-fwCZheo-n_Mx9IQ02upaLz87dMne5xsrlcFq5G0vxBHD_IzIXHOIvuR--axMLA&utm_content=175723863&utm_source=hs_email)\n- **[MCUNetV2]**：面向小型深度学习的内存高效补丁式推理 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fabs\u002F2110.15352#)\n- **[LCS]**：学习可压缩子空间，用于推理时的自适应网络压缩 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2110.04252.pdf)\n- 基于腕部光电容积脉搏波传感器的特征增强混合 CNN，用于压力识别 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2108.03166.pdf)\n- **[ANALOGNETS]**：噪声鲁棒型 TinyML 模型与始终开启的模拟存内计算加速器的软硬件协同设计 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2111.06503.pdf)\n- **[BSC]**：基于块的随机计算，以实现准确高效的 TinyML |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2111.06686.pdf)\n- **[TiWS-iForest]**：弱监督和 Tiny ML 场景下的孤立森林 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2111.15432.pdf)\n- **[RadarNet]**：利用微型雷达传感器的高效手势识别技术 |[`[pdf]`](https:\u002F\u002Fdl.acm.org\u002Fdoi\u002Fpdf\u002F10.1145\u002F3411764.3445367)\n- 工业物联网中复杂事件处理与 TinyML 的协同作用 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2105.03371.pdf)\n\n| ▲ [返回顶部](#tinyml-papers-and-projects) |\n  | ------------------------------------ |\n\n\n\n### \u003Cins>**2022**\u003C\u002Fins>\n\n- 用于灵活端到端推理真实世界深度神经网络的异构存内计算集群 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2201.01089.pdf)\n- CFU Playground：面向FPGA上TinyML加速的全栈开源框架 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2201.01863.pdf)\n- BottleFit：在深度神经网络中学习压缩表示，以实现高效分割计算 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2201.02693.pdf)\n- **[UDC]**：适用于可压缩TinyML模型的统一DNA结构 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2201.05842.pdf)\n- 一种基于虚拟机\u002F容器的方法，用于扩展TinyML应用 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2202.05057.pdf)\n- 一种快速网络探索策略，用于为关键词检测任务进行低功耗性能分析 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2202.02361.pdf)\n- PocketNN：通过直接反馈对齐和Pocket激活函数，在纯C++中实现仅整数运算的神经网络训练与推理 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2201.02863.pdf)\n- **[TinyMLOps]**：大规模边缘AI部署面临的运营挑战 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2203.10923.pdf)\n- **[Auritus]**：一个开源优化工具包，用于使用可穿戴设备训练和开发人体运动模型及滤波器 |[`[pdf]`](https:\u002F\u002Fwww.researchgate.net\u002Fpublication\u002F359759183_Auritus_An_Open-Source_Optimization_Toolkit_for_Training_and_Development_of_Human_Movement_Models_and_Filters_Using_Earables) |[`[code]`](https:\u002F\u002Fgithub.com\u002Fnesl\u002Fauritus)\n- 在生产环境中实现机器学习分类器的超参数调优 |[`[pdf]`](https:\u002F\u002Fwww.researchgate.net\u002Fpublication\u002F356911955_Enabling_Hyperparameter_Tuning_of_Machine_Learning_Classifiers_in_Production)\n- TinyOdom：硬件感知的高效神经惯性导航 |[`[pdf]`](https:\u002F\u002Fwww.researchgate.net\u002Fpublication\u002F360075622_TinyOdom_Hardware-Aware_Efficient_Neural_Inertial_Navigation?fbclid=IwAR3F5LhoDiXD6tDhyE2PLFDB1hgy0IBM6V5YIUFwva7TvUvHYDi7C0ryTB8) |[`[code]`](https:\u002F\u002Fgithub.com\u002Fnesl\u002Ftinyodom?fbclid=IwAR1zbqrymVPxsVRHT6LZOtcwjJtzUYqfd0E8ChliklUvug-D6KKhWAAZ3dg)\n- 针对边缘TPU上的设备端ML搜索高效神经架构 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2204.14007.pdf)\n- 绿色加速霍夫丁树 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2205.03184.pdf)\n- tinyRadar：基于毫米波雷达的人体活动分类，用于边缘计算 |[`[pdf]`](https:\u002F\u002Flabs.dese.iisc.ac.in\u002Fneuronics\u002Fwp-content\u002Fuploads\u002Fsites\u002F16\u002F2022\u002F04\u002FmmWave_Radar_ver1.3.5.pdf)\n- 机器学习传感器 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.03266.pdf)\n- 评估物联网环境中的多时间序列短期预测 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.07784.pdf)\n- 如何为嵌入式系统训练高精度二值神经网络？ |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.12322.pdf)\n- Vildehaye：一系列多功能、适用广泛且经过实地验证的轻量级野生动物追踪与传感标签 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.06171.pdf)\n- 设备端内存小于256KB下的在线训练 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.15472.pdf)\n- 基于辅助网络的TinyML深度剪枝 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2204.10546.pdf)\n- **[EdgeNeXt]**：面向移动视觉应用的高效融合CNN-Transformer架构 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.10589.pdf)\n- 小型机器人学习：资源受限机器人中机器学习的挑战与方向 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2205.05748.pdf)\n- **[POET]**：在微型设备上训练神经网络，结合集成重化与分页 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2207.07697.pdf)\n- 基于决策树和CNN的微控制器上两阶段人体活动识别 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.07652.pdf)\n- 如何规模化管理小型机器学习——工业视角 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2202.09113.pdf)\n- **[SeLoC-ML]**：面向工业物联网中机器学习应用的语义低代码工程 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2207.08818.pdf)\n- **[IMU2Doppler]**：利用IMU数据进行基于多普勒的活动识别的跨模态域适应 |[`[pdf]`](https:\u002F\u002Fsmashlab.io\u002Fpdfs\u002Fimu2dop.pdf)\n- **[Tiny-HR]**：迈向可解释的边缘设备心率估计机器学习流水线 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2208.07981.pdf)\n- [在微型能量采集物联网设备上实现快速深度学习] |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2111.14051.pdf)\n- 极其简单的激活形状设计，用于分布外检测 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2209.09858.pdf)\n- 面向资源受限的TinyML应用的像素内存储处理范式 |[`[pdf]`](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41598-022-17934-1.pdf)\n- **[tinySNN]**：迈向内存与能耗高效的脉冲神经网络 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2206.08656.pdf)\n- **[DeepPicarMicro]**：将TinyML应用于自主网络物理系统 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2208.11212.pdf)\n- 面向手势与视觉智能传感器的增量式在线学习算法比较 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2209.00591.pdf)\n- **[Protean]**：一个节能且异构的平台，用于自适应和硬件加速的无电池计算 |[`[pdf]`](https:\u002F\u002Fdl.acm.org\u002Fdoi\u002Fpdf\u002F10.1145\u002F3560905.3568561?utm_medium=email&_hsmi=249309419&_hsenc=p2ANqtz--_ltviIYSpzfz9c4PiqHChBsEQkRDbr6treGEpYyBsHcwC5HX_R7JMp4ldGoydUfIlR-bOB-V2lKC0RIAhOcFen7daog&utm_content=249309419&utm_source=hs_email)\n- 传感器内计算与类脑计算足以实现节能计算机视觉 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2212.10881.pdf)\n- 基于二的幂量化实现神经网络的节能硬件加速 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2209.15257.pdf)\n- 实现无ISP的低功耗计算机视觉 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2210.05451.pdf)\n- 重新思考视觉Transformer，使其达到MobileNet的尺寸与速度 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2212.08059.pdf)\n- 类脑计算与传感技术在太空中的应用 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2212.05236.pdf)\n- 联合数据深化与预取，实现边缘学习的节能 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2211.07146.pdf)\n- PreMa：在嵌入式边缘层面实时预测并维护电磁阀 | [`pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2211.12326.pdf)\n\n  | ▲ [返回顶部](#tinyml-papers-and-projects) |\n  | ------------------------------------ |\n\n### \u003Cins>**2023**\u003C\u002Fins>\n\n- **[Coral NPU]**：专为边缘端节能型AI设计的机器学习加速器核心。|[`[代码]`](https:\u002F\u002Fgithub.com\u002Fgoogle-coral\u002Fcoralnpu)\n- **[cpp-transformer]**：一种无需特殊库依赖的Transformer C++实现，包含训练与推理功能。|[`[代码]`](https:\u002F\u002Fgithub.com\u002Ffreelw\u002Fcpp-transformer)\n- 探索在资源受限的可穿戴设备上本地自动识别健身锻炼 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2301.05748.pdf)\n- **[MetaLDC]**：用于设备端快速适配的低维计算分类器元学习 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2302.12347.pdf)\n- 更快的注意力机制正是你需要的：通过双重压缩注意力浓缩器构建面向边缘的快速自注意力神经网络骨干架构 |[`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2208.06980.pdf)\n- **[TinyReptile]**：结合联邦元学习的TinyML |[`[pdf](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2304.05201.pdf)`\n- **[TinyProp]** - 面向高效TinyML设备端学习的自适应稀疏反向传播 |[`[pdf](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2308.09201.pdf)`\n\n- **[LiteTrack]** - 基于异步特征提取的层剪枝技术，用于轻量级高效视觉跟踪 - 自适应稀疏反向传播以实现高效的TinyML设备端学习 |[`[pdf](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2409.00608v1)`\n\n- **[MCUFormer]** - 在内存有限的微控制器上部署视觉Transformer |[`[pdf](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.16898)`\n\n### \u003Cins>**2024**\u003C\u002Fins>\n\n- 模型压缩实践：来自创建设备端机器学习体验的从业者的经验教训 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2310.04621v2)\n- TinyAgent：在边缘端进行函数调用 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2409.00608v1)\n- **[SENSORLLM]**：将大型语言模型与运动传感器对齐，用于人类活动识别 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2410.10624)\n- **[Penetrative AI]**：让大语言模型理解物理世界 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2310.09605v2)\n- **[MobileCLIP]**：通过多模态强化训练实现快速图文模型 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2311.17049)\n- **[Zero-TPrune]**：利用预训练Transformer中的注意力图进行零样本标记剪枝 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2305.17328)\n- 通过大型语言模型迈向边缘通用智能：机遇与挑战 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2410.18125)\n- **[TinyTTA]**：通过边缘设备上的早退出集成实现高效测试时适配 | [`[pdf]`](https:\u002F\u002Fproceedings.neurips.cc\u002Fpaper_files\u002Fpaper\u002F2024\u002Ffile\u002F4c454d34f3a4c8d6b4ca85a918e5d7ba-Paper-Conference.pdf)\n\n\u003Cins>**2025**\u003C\u002Fins>\n\n- **[EdgeMark]**：嵌入式人工智能工具的自动化与基准测试系统 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2502.01700) | [`[官方代码]`](https:\u002F\u002Fgithub.com\u002FBlack3rror\u002FEdgeMark)\n- 优化边缘AI：数据、模型和系统策略的综合综述 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2501.03265) | [`[GitHub]`](https:\u002F\u002Fgithub.com\u002Fwangxb96\u002FAwesome-EdgeAI)\n- 小型语言模型是代理型AI的未来 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.02153) | [`[GitHub]`](https:\u002F\u002Fresearch.nvidia.com\u002Flabs\u002Flpr\u002Fslm-agents\u002F)\n- SensorLM：学习可穿戴传感器的语言 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2506.09108)\n- PyTorch与TensorFlow在深度学习中的比较研究：易用性、性能与部署权衡 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2508.04035v1)\n- **[MobileCLIP2]**：改进多模态强化训练 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2508.20691) | [`[官方代码]`](https:\u002F\u002Fgithub.com\u002Fapple\u002Fml-mobileclip\u002Ftree\u002Fmain)\n- **[FastVLM]**：面向视觉语言模型的高效视觉编码 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2412.13303)\n- **[Flavors of Moonshine]**：面向边缘设备的小型专用ASR模型 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2509.02523)\n- 大语言模型剪枝与蒸馏实践：Minitron方法 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2408.11796)\n- 嵌入式系统中TinyML模型的实时性能基准测试（PICO：推理、CPU及操作性能） | [`[pdf]`](https:\u002F\u002Fwww.arxiv.org\u002Fpdf\u002F2509.04721)\n- 用于代理系统的小型语言模型：架构、能力与部署权衡综述 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2510.03847)\n- AutoNeural：为NPU推理协同设计视觉-语言模型 | [`[pdf]`](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2512.02924)\n\n  | ▲ [返回顶部](#tinyml-papers-and-projects) |\n  | ------------------------------------ |\n\n## 优秀的TinyML项目\n\n### \u003Cins>**项目源码**\u003C\u002Fins>\n\n- TinyFederatedLearning | [`[官方代码]`](https:\u002F\u002Fgithub.com\u002Fkavyakvk\u002FTinyFederatedLearning) [`[演示视频]`](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=KSaidr3ZN9M&feature=youtu.be) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fkavyakvk\u002FTinyFederatedLearning?style=social)\n- [TinyML 学习小组](https:\u002F\u002Fgithub.com\u002Ftinyml-team\u002Fstudy-group) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Ftinyml-team\u002Fstudy-group?style=social)\n- [Arduino 垃圾分类 TinyML 示例](https:\u002F\u002Fgithub.com\u002Flightb0x\u002Farduino_trash_classification) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Flightb0x\u002Farduino_trash_classification?style=social)\n- [Arduino 上的 TinyML](https:\u002F\u002Fgithub.com\u002Fsandeepmistry\u002Faimldevfest-workshop-2019) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fsandeepmistry\u002Faimldevfest-workshop-2019?style=social)\n- [边缘 AI 异常检测](https:\u002F\u002Fgithub.com\u002FShawnHymel\u002Ftinyml-example-anomaly-detection) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FShawnHymel\u002Ftinyml-example-anomaly-detection?style=social)\n- [Air Guitar CS249R](https:\u002F\u002Fgithub.com\u002FRobJMal\u002FAir-Guitar-CS249R) [`[演示视频]`](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=PVk9RUW1Hwo) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FRobJMal\u002FAir-Guitar-CS249R?style=social)\n- [TinyML ESP32](https:\u002F\u002Fgithub.com\u002FHollowMan6\u002FTinyML-ESP32) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FHollowMan6\u002FTinyML-ESP32?style=social)\n- [MagicWand-TFLite-ESP32](https:\u002F\u002Fgithub.com\u002Fandriyadi\u002FMagicWand-TFLite-ESP32) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fandriyadi\u002FMagicWand-TFLite-ESP32?style=social)\n- [利用 BLE 信标、ESP32 和机器学习在家中定位你的猫](https:\u002F\u002Fgithub.com\u002FfilipsPL\u002Fcat-localizer) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FfilipsPL\u002Fcat-localizer?style=social)\n- [ESP32 Cam 与 Edge Impulse](https:\u002F\u002Fgithub.com\u002Fluisomoreau\u002FESP32-Cam-Edge-Impulse) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fluisomoreau\u002FESP32-Cam-Edge-Impulse?style=social)\n- [C++ 神经网络与机器学习项目](https:\u002F\u002Fgithub.com\u002Fintel\u002Fcppnnml) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fintel\u002Fcppnnml?style=social)\n- [完整水表系统](https:\u002F\u002Fgithub.com\u002Fjomjol\u002Fwater-meter-system-complete) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fjomjol\u002Fwater-meter-system-complete?style=social)\n- [在 Raspberry Pi Pico 上使用 MNIST 进行数字识别](https:\u002F\u002Fgithub.com\u002Fiwatake2222\u002Fpico-mnist) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fiwatake2222\u002Fpico-mnist?style=social)\n- [基于机器学习的霍尔传感器转速计](https:\u002F\u002Fgithub.com\u002FMiguelest07\u002FHallSensor_ML_EdgeImpulse) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FMiguelest07\u002FHallSensor_ML_EdgeImpulse?style=social)\n- [使用 TinyML 进行天气预测](https:\u002F\u002Fgithub.com\u002FBaptisteZloch\u002FWeather_forcasting) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FBaptisteZloch\u002FWeather_forcasting?style=social)\n- [应用于 STM32F407 微控制器的不同框架的 TinyML](https:\u002F\u002Fgithub.com\u002Ffjpolo\u002FSTM32F407TinyML) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Ffjpolo\u002FSTM32F407TinyML?style=social)\n- [CurrentSense-TinyML](https:\u002F\u002Fgithub.com\u002FSantandersecurityresearch\u002FCurrentSense-TinyML) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FSantandersecurityresearch\u002FCurrentSense-TinyML?style=social)\n- [MicroPython 中的 TensorFlow Lite for Microcontrollers](https:\u002F\u002Fgithub.com\u002Fmocleiri\u002Ftensorflow-micropython-examples) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fmocleiri\u002Ftensorflow-micropython-examples?style=social)\n- [适用于 Espressif 芯片组的 TensorFlow Lite Micro](https:\u002F\u002Fgithub.com\u002Fespressif\u002Ftflite-micro-esp-examples) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fespressif\u002Ftflite-micro-esp-examples?style=social)\n- [Pico 平台的 ML 音频分类器示例](https:\u002F\u002Fgithub.com\u002FArmDeveloperEcosystem\u002Fml-audio-classifier-example-for-pico) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FArmDeveloperEcosystem\u002Fml-audio-classifier-example-for-pico?style=social)\n- [使用 Raspberry Pi Pico 和机器学习进行手写数字分类](https:\u002F\u002Fgithub.com\u002Fcode2k13\u002Frpipico_digit_classification) ![GitHub 星标](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fcode2k13\u002Frpipico_digit_classification?style=social)\n\n### \u003Cins>**项目文章**\u003C\u002Fins>\n\n- `2020-09` [基于计算机视觉的自主嵌入式自动驾驶](https:\u002F\u002Fwww.edgeimpulse.com\u002Fblog\u002Fautonomous-driving-using-computer-vision)\n- `2020-10` [EleTect - 基于TinyML和物联网的智能野生动物追踪器](https:\u002F\u002Fwww.hackster.io\u002Fdhruvsheth_\u002Feletect-tinyml-and-iot-based-smart-wildlife-tracker-c03e5a)\n- `2020-03` [手写识别](https:\u002F\u002Fwww.hackster.io\u002Fnaveenbskumar\u002Fhandwriting-recognition-7583e3)\n- `2021-01` [为什么对TinyML系统进行基准测试具有挑战性](https:\u002F\u002Fanalyticsindiamag.com\u002Fwhy-benchmarking-tinyml-systems-is-challenging\u002F)\n- `2021-01` [使用TinyML构建属于你自己的Google Assistant](https:\u002F\u002Fmjrobot.org\u002F2021\u002F01\u002F27\u002Fbuilding-an-intelligent-voice-assistant-from-scratch\u002F)\n- `2021-02` [利用AVR-IoT进行跌倒检测和心率监测](https:\u002F\u002Fwww.hackster.io\u002Fnaveenbskumar\u002Ffall-detection-and-heart-rate-monitoring-using-avr-iot-75fb16)\n- `2021-02` [创客秀：用于野生动物保护的TinyML](https:\u002F\u002Fdev.to\u002Ffordevs-community\u002Fthe-maker-show-tinyml-for-wildlife-conservation-idg)\n- `2021-05` [低于100美元且功耗低于1mW：面向所有人的肺炎检测解决方案](https:\u002F\u002Fwww.edgeimpulse.com\u002Fblog\u002Funder-dollar100-and-less-than-1mw-pneumonia-detection-solution-for-everyone)\n- `2021-06` [使用Edge Impulse早期检测猪呼吸道疾病](https:\u002F\u002Fwww.hackster.io\u002Fclinton_oduor\u002Fearly-pigs-respiratory-disease-detection-using-edge-impulse-2ab039)\n- `2021-06` [姿势看护者](https:\u002F\u002Fwww.hackster.io\u002Fnaveenbskumar\u002Fposture-watchdog-c03f77?utm_campaign=Advanced%20Wearables%20Contest%20Hackster.io&utm_source=twitter&utm_medium=social&utm_content=Dream%20Smart%20Wearables%20winner:%20posture%20watchdog)\n- `2021-07` [基于TinyML的本地化环境感知](https:\u002F\u002Fhighdemandskills.com\u002Flocalized-monitoring-tinyml\u002F)\n- `2021` [无线季度：边缘智能](https:\u002F\u002Fwww.nordicsemi.com\u002F-\u002Fmedia\u002FPublications\u002FWireless-Quarter-pdf\u002F2021\u002FWQ_Issue_2_2021.pdf?la=en&hash=A58D1AB12248E18E465658CE3CDFE33F9187692F#page=8)\n- [Arduino机器学习：构建TensorFlow Lite模型来控制机器人小车](https:\u002F\u002Fwww.survivingwithandroid.com\u002Farduino-machine-learning-tensorflow-lite\u002F)\n- [TinyML ESP32-CAM：使用Edge Impulse进行边缘图像分类](https:\u002F\u002Fwww.survivingwithandroid.com\u002Ftinyml-esp32-cam-edge-image-classification-with-edge-impulse\u002F)\n- [使用TinyAutomator进行预测性维护](https:\u002F\u002Fwww.waylay.io\u002Farticles\u002Fpredictive-maintenance-with-tinyautomator)\n- [使用Arduino和Arducam的TinyML人体检测](https:\u002F\u002Fwww.thetinymlbook.com\u002Fresources\u002Ftinyml-person-detection)\n- [使用VLM结合Qwen2.5-VL进行目标检测与空间理解](https:\u002F\u002Flearnopencv.com\u002Fobject-detection-with-vlms-ft-qwen2-5-vl\u002F)\n- [边缘设备上的VLM：值得 hype 还是只是噱头？](https:\u002F\u002Flearnopencv.com\u002Fvlm-on-edge-devices\u002F)\n\n| ▲ [返回顶部](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## 基准测试及其他\n\n- [EEMBC的EnergyRunner](https:\u002F\u002Fgithub.com\u002Feembc\u002Fenergyrunner)：EEMBC为MLPerf Tiny基准测试设计的能量运行应用框架。\n- [MLPerf - Tiny](https:\u002F\u002Fmlcommons.org\u002Fen\u002Finference-tiny-05\u002F)：这是一个针对微控制器等极低功耗系统的机器学习基准测试套件。[`[GitHub]`](https:\u002F\u002Fgithub.com\u002Fmlcommons\u002Ftiny\u002Ftree\u002Fv0.5)\n- [FedML](https:\u002F\u002Ffedml.ai\u002F)：一个用于联邦学习的研究库及基准测试平台。[`[GitHub]`](https:\u002F\u002Fgithub.com\u002FFedML-AI\u002FFedML)\n- [FogML](https:\u002F\u002Fgithub.com\u002Ftszydlo\u002FFogML)：一个用于生成嵌入式设备推理函数源代码的研究库。[`[GitHub]`]()\n- [边缘端机器学习基准测试](https:\u002F\u002Fgithub.com\u002Faallan\u002Fbenchmarking-ml-on-the-edge)\n\n| ▲ [返回顶部](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## 书籍\n\n- `[2022-12]` **边缘人工智能**（D. Situnayake & J. Plunkett，2022年，O'Reilly）：[`[图书]`](https:\u002F\u002Fwww.oreilly.com\u002Flibrary\u002Fview\u002Fai-at-the\u002F9781098120191\u002F)\n- `[2022-10]` **商品级小型设备上的机器学习**（S. Guo & Q. Zhou，2022年，CRC Press）：[`[图书]`](https:\u002F\u002Fwww.routledge.com\u002FMachine-Learning-on-Commodity-Tiny-Devices-Theory-and-Practice\u002FGuo-Zhou\u002Fp\u002Fbook\u002F9781032374239)\n- `[2022-07]` **TinyML入门**（Rohit Sharma，2022年，AITS）：[`[图书]`](https:\u002F\u002Fwww.thetinymlbook.com\u002F) | [`[GitHub]`](https:\u002F\u002Fgithub.com\u002Fai-techsystems\u002FdeepC)\n- `[2022-04]` **TinyML烹饪书**（Gian Marco Iodice，2022年，Packt）：[`[图书]`](https:\u002F\u002Fwww.packtpub.com\u002Fproduct\u002Ftinyml-cookbook\u002F9781801814973) | [`[GitHub]`](https:\u002F\u002Fgithub.com\u002FPacktPublishing\u002FTinyML-Cookbook)\n- `[2021-03]` **物联网人工智能烹饪书**（Michael Roshak，2021年，Packt）：[`[图书]`](https:\u002F\u002Fwww.packtpub.com\u002Fproduct\u002Fartificial-intelligence-for-iot-cookbook\u002F9781838981983) | [`[GitHub]`](https:\u002F\u002Fgithub.com\u002FPacktPublishing\u002FArtificial-Intelligence-for-IoT-Cookbook)\n- `[2020-04]` **使用TensorFlow Lite、ML Kit和Flutter进行移动深度学习**：构建可扩展的真实世界项目，在Android和iOS上实现端到端神经网络（Anubhav Singh、Rimjhim Bhadani，2020年，Packt）：[`[图书]`](https:\u002F\u002Fwww.amazon.com\u002FMobile-Deep-Learning-TensorFlow-Flutter\u002Fdp\u002F1789611210)\n- `[2020-01]` **TinyML：在Arduino和超低功耗微控制器上使用TensorFlow Lite进行机器学习**（Pete Warden，O'Reilly Media）：[`[图书]`](https:\u002F\u002Fwww.amazon.com\u002FTinyML-Learning-TensorFlow-Ultra-Low-Power-Microcontrollers\u002Fdp\u002F1492052043)\n\n| ▲ [返回顶部](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## 文章\n\n- `2019-12` [TinyML即服务：它是什么，对物联网边缘计算意味着什么？](https:\u002F\u002Fwww.ericsson.com\u002Fen\u002Fblog\u002F2019\u002F12\u002Ftinyml-as-a-service-iot-edge)\n- `2019-12` [TinyML即服务与边缘端机器学习的挑战](https:\u002F\u002Fwww.ericsson.com\u002Fen\u002Fblog\u002F2019\u002F12\u002Ftinyml-as-a-service)\n- `2020-05` [使用TensorFlow Lite进行模型量化](https:\u002F\u002Fmedium.com\u002Fsclable\u002Fmodel-quantization-using-tensorflow-lite-2fe6a171a90d)\n- `2020-09` [TinyML正在为数十亿设备注入活力](https:\u002F\u002Fthenextweb.com\u002Fneural\u002F2020\u002F09\u002F03\u002Ftinyml-is-breathing-life-into-billions-of-devices\u002F)\n- `2020-12` [2021年物联网嵌入式机器学习预测](https:\u002F\u002Fwww.iotworldtoday.com\u002F2020\u002F12\u002F10\u002Fpredictions-for-embedded-machine-learning-for-iot-in-2021\u002F)\n- `2020-12` [马修·马蒂纳：口袋里的救生模型](https:\u002F\u002Fread.deeplearning.ai\u002Fthe-batch\u002Fissue-72\u002F)\n- `2020-12` [如今，只需四比特微型计算机即可训练人工智能](https:\u002F\u002Fwww.technologyreview.com\u002F2020\u002F12\u002F11\u002F1014102\u002Fai-trains-on-4-bit-computers\u002F)\n- `2021-01` [预测性维护如何彻底改变工业企业](https:\u002F\u002Ftechhq.com\u002F2021\u002F01\u002Fhow-predictive-maintenance-is-changing-the-industrial-enterprise-for-good\u002F)\n- `2021-02` [什么是TinyML？](https:\u002F\u002Fwww.fierceelectronics.com\u002Felectronics\u002Fwhat-tinyml)\n- `2021-02` [人工智能如何接管传感器](https:\u002F\u002Fwww.electropages.com\u002Fblog\u002F2021\u002F02\u002Fhow-ai-taking-sensors)\n- `2021-04` [MLCommons™发布MLPerf™推理v1.0结果，并首次提供功耗测量数据](https:\u002F\u002Fmlcommons.org\u002Fen\u002Fnews\u002Fmlperf-inference-v10\u002F)\n- `2021-05` [TapLock——一款搭载机器学习的自行车锁](https:\u002F\u002Fwww.hackster.io\u002Ftaplock\u002Ftaplock-a-bike-lock-with-machine-learning-85641c)\n- `2021-05` [夺回控制权](https:\u002F\u002Fwww.hackster.io\u002Fnews\u002Ftaking-back-control-14068dbb0bb7?fbclid=IwAR0QGucom06pzd7K5SJIdYByZr67xd29YlqTdbnK78OU7GqW540vJPeD534)\n- `2021-06` [用于在通用微控制器上部署TinyML应用的神经网络架构](https:\u002F\u002Fcommunity.arm.com\u002Fdeveloper\u002Fresearch\u002Fb\u002Farticles\u002Fposts\u002Fneural-network-architectures-for-deploying-tinyml-applications-on-commodity-microcontrollers)\n- `2021-06` [MicroCosmos中的TinyML](https:\u002F\u002Fwww.hackster.io\u002FCHA_RAN\u002Ftinyml-in-microcosmos-c1161c)\n- `2021-06` [“小数据”对机器学习同样至关重要](https:\u002F\u002Fwww.hackster.io\u002FCHA_RAN\u002Ftinyml-in-microcosmos-c1161c)\n- `2021-07` [一种原生灵活的32位Arm微处理器](https:\u002F\u002Fwww.nature.com\u002Farticles\u002Fs41586-021-03625-w)\n- `2021-07` [可穿戴设备可降低盲人和视力障碍者的碰撞风险](https:\u002F\u002Fmasseyeandear.org\u002Fnews\u002Fpress-releases\u002F2021\u002F07\u002Fwearable-devices-can-reduce-collision-risk-in-blind-and-visually-impaired-people)\n- `2021-09` [以模拟仪表为例的AI检测](https:\u002F\u002Findatalabs.com\u002Fblog\u002Fai-inspection)\n- `2025-10` [用于构建代理型AI解决方案的顶级小型语言模型](https:\u002F\u002Fthirdeyedata.ai\u002Ftop-small-language-models-for-agentic-ai-solutions-development\u002F)\n\n| ▲ [返回顶部](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## 库与工具\n\n- [Edge Impulse](https:\u002F\u002Fwww.edgeimpulse.com\u002F) - 一个交互式平台，用于生成可在微控制器上运行的模型。他们在社交媒体上也非常活跃，经常分享关于边缘AI\u002FTinyML的最新动态。\n- [EVE是边缘虚拟化引擎](https:\u002F\u002Fgithub.com\u002Flf-edge\u002Feve\u002Fblob\u002Fmaster\u002FREADME.md)\n- [microTVM](https:\u002F\u002Ftvm.apache.org\u002Fdocs\u002Fmicrotvm\u002Findex.html) - 一个开源工具，用于优化张量程序。\n- [Larq](https:\u002F\u002Fgithub.com\u002Flarq\u002Flarq) - 一个用于训练二值化神经网络的开源库。\n- [微控制器上的神经网络（NNoM）](https:\u002F\u002Fgithub.com\u002Fmajianjia\u002Fnnom) - 一个专为微控制器设计的高级层式神经网络库，支持CMSIS-NN。\n- [BerryNet](https:\u002F\u002Fgithub.com\u002FDT42\u002FBerryNet) - 一个基于树莓派及其他边缘设备的深度学习网关。\n- [Rune](https:\u002F\u002Fgithub.com\u002Fhotg-ai\u002Frune) - 提供容器来封装和部署边缘ML流水线及应用程序。\n- [Onnxruntime](https:\u002F\u002Fgithub.com\u002Fmicrosoft\u002Fonnxruntime) - 一个跨平台、高性能的ML推理与训练加速器。\n- [deepC](https:\u002F\u002Fgithub.com\u002Fai-techsystems\u002FdeepC) - 一个与厂商无关的TinyML深度学习库、编译器和推理框架，适用于微型计算机和微控制器。\n- [针对Arduino的deepC](https:\u002F\u002Fgithub.com\u002Fai-techsystems\u002Farduino) - 一个专为Arduino IDE定制的TinyML深度学习库。\n- [emlearn](https:\u002F\u002Fgithub.com\u002Femlearn\u002Femlearn) - 面向微控制器和嵌入式系统的机器学习工具。可在Python中训练模型，然后在任何具备C99编译器的设备上进行推理。\n\n| ▲ [返回顶部](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## 课程\n\n- **11-767：设备端机器学习秋季课程** - 卡内基梅隆大学 | [`[官网]`](https:\u002F\u002Fcmu-odml.github.io)\n- **TinyML4D：UNIFEI-IESTI01-TinyML-2023.1** - UNIFEI | [`[官网]`](https:\u002F\u002Fgithub.com\u002FMjrovai\u002FUNIFEI-IESTI01-TinyML-2023.1)\n- **嵌入式深度学习导论** - 卡内基梅隆大学 | [`[官网]`](https:\u002F\u002Fz4ziad.github.io\u002Fembed-dl-s23\u002FEmbeddedDL_S23\u002F)\n- **TinyML与高效深度学习** - 麻省理工学院 | [`[官网]`](https:\u002F\u002Fefficientml.ai\u002F)\n- **ARM边缘端机器学习：实用入门** - ARM | [`[edX]`](https:\u002F\u002Fwww.edx.org\u002Fcourse\u002Fmachine-learning-at-the-edge-a-practical-introduction-from-arm)\n- **CS249r：微型机器学习（TinyML）** - *哈佛大学*，由Vijay Janapa Reddi主讲：[sites.google.com](https:\u002F\u002Fsites.google.com\u002Fg.harvard.edu\u002Ftinyml\u002Fhome?authuser=0) | [`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fchannel\u002FUCLv1K6OaYHP44hXFd5rNqyA) | [`[edX]`](https:\u002F\u002Fwww.edx.org\u002Fprofessional-certificate\u002Fharvardx-tiny-machine-learning) | [`[GitHub]`](https:\u002F\u002Fgithub.com\u002FtinyMLx\u002Fcolabs)\n- **TinyML规模化MLOps** - *哈佛大学*，由Vijay Janapa Reddi主讲：[`[edX]`](https:\u002F\u002Fwww.edx.org\u002Fcourse\u002Fmlops-for-scaling-tinyml)\n- **嵌入式机器学习导论** - *Edge Impulse*，由Shawn Hymel主讲：[`[Coursera]`](https:\u002F\u002Fwww.coursera.org\u002Flearn\u002Fintroduction-to-embedded-machine-learning)\n- **嵌入式与分布式人工智能** - *瑞典延雪平大学*，由Beril Sirmacek主讲：[`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=OTXqT00MmPA&list=PLyulI6o7oOtycIT15i_I2_mhuLxnNvPvX)\n- **MLT人工智能——边缘AI** - 东京机器学习：[`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fplaylist?list=PLaPdEEY26UXxrxn-82sqe9cTTvWC0y-rM)\n- TinyML @ 宾夕法尼亚大学 | [`[官网]`](https:\u002F\u002Fgithub.com\u002FMjrovai\u002FUNIFEI-IESTI01-TinyML-2023.1) | [`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fplaylist?list=PL7rtKJAz_mPe6kAbiH6Ucq02Vpa95qvBJ)\n\n| ▲ [返回顶部](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## TinyML 讲座与会议\n\n- TinyML 讲座、峰会及研究研讨会：[`[官网]`](https:\u002F\u002Fconf.researchr.org\u002Fseries\u002Ftinyml-symp) | [`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fchannel\u002FUC9iWqvsWjhowkHWVJquHwkg)\n- 嵌入式视觉峰会 - 边缘 AI 与视觉联盟 ([Edge AI & Vision Alliance](https:\u002F\u002Fwww.edge-ai-vision.com))：[`[官网]`](https:\u002F\u002Fembeddedvisionsummit.com) | [`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fc\u002FEdgeAIandVisionAlliance)\n- 低功耗计算机视觉挑战赛 (LPCV)：[`[官网]`](https:\u002F\u002Flpcv.ai) | [`[YouTube]`](https:\u002F\u002Fwww.youtube.com\u002Fchannel\u002FUCAeAbQsRUZ8KWmGUKejtgBg)\n\n|                                                     标题                                                     |    演讲者    | 发布日期 |                                                  链接                                                   |\n| :-----------------------------------------------------------------------------------------------------------: | :-----------: | :------------: | :-----------------------------------------------------------------------------------------------------: |\n|           [TinyML 设备大规模部署面临的挑战](https:\u002F\u002Fyoutu.be\u002FbwjHLrLGkOY)            |  G. Raghavan  |   2022-04-29   |            [幻灯片](https:\u002F\u002Fcms.tinyml.org\u002Fwp-content\u002Fuploads\u002Fsummit2022\u002FRaghavan-Gopal.pdf)             |\n|            [为嵌入式工程师构建以数据为中心的 AI 工具链](https:\u002F\u002Fyoutu.be\u002F9rnzM-C7QdA)            | D. Situnayake |   2022-04-29   |           [幻灯片](https:\u002F\u002Fcms.tinyml.org\u002Fwp-content\u002Fuploads\u002Fsummit2022\u002FSitunayake-Daniel.pdf)           |\n|                    [传感器与机器学习：用更少的资源实现更智能的唤醒](https:\u002F\u002Fyoutu.be\u002FVXpQlOouBqU)                    |   A. Ataya    |   2022-05-04   |              [幻灯片](https:\u002F\u002Fcms.tinyml.org\u002Fwp-content\u002Fuploads\u002Fsummit2022\u002FAbbas-Ataya.pdf)              |\n| [适用于 TinyML 的 MLOps：在规模化应用中落地 TinyML 的挑战与方向](https:\u002F\u002Fyoutu.be\u002FyydnTSH0R2I) |  V.J. Reddi   |   2022-05-24   | [幻灯片](https:\u002F\u002Fcms.tinyml.org\u002Fwp-content\u002Fuploads\u002Ftalks2022\u002FtinyML_Talks_Vijay_Janapa_Reddi_220524.pdf) |\n|              [振动监测机器学习演示](https:\u002F\u002Fyoutu.be\u002F2iInOo0Lkfs)              |  J. Edwards   |   2020-12-22   |                            [GitHub](https:\u002F\u002Fgithub.com\u002FNumerix-DSP\u002Fsiglib\u002F)                             |\n|     [从 AI 到 IntelligentAI：降低边缘端 AI 成本](https:\u002F\u002Fyoutu.be\u002FmYy4Zv80tXQ)      |  J. Edwards   |   2020-12-22   |                                 [官网](https:\u002F\u002Fwww.numerix-dsp.com\u002Fai\u002F)                                  |\n\n| ▲ [返回顶部](#tinyml-papers-and-projects) |\n| ------------------------------------ |\n\n## 竞赛\n\n- **[LPCV]**：低功耗计算机视觉挑战赛 |[`[官网]`](https:\u002F\u002Flpcv.ai\u002F)\n\n## 其他精彩仓库\n\n- [优秀的人类活动识别资源](https:\u002F\u002Fgithub.com\u002FJie-su\u002FAwesome_Human_Activity_Recognition#2-Paper-with-code)\n\n## 联系与反馈\n\n如果您对 TinyML 相关论文和项目有任何建议，欢迎随时给我发送邮件 :)\n\n- [电子邮箱](mailto:gigwegbe@gmail.com)\n- [拉取请求](https:\u002F\u002Fgithub.com\u002Fgigwegbe\u002Ftinyml-papers-and-projects\u002Fpulls)","# TinyML Papers and Projects 快速上手指南\n\n`tinyml-papers-and-projects` 并非一个需要编译安装的可执行软件或 Python 库，而是一个**精选资源列表仓库**。它汇集了关于 TinyML（微型机器学习）的学术论文、开源项目、基准测试工具及相关教程。\n\n本指南将指导开发者如何获取该资源库，并从中提取有价值的代码项目进行实践。\n\n## 环境准备\n\n由于本项目主要是文档和链接集合，无需特殊的系统运行环境，但为了运行列表中引用的具体项目（如 TensorFlow Lite Micro, MCUNet 等），建议准备以下基础环境：\n\n*   **操作系统**: Windows, macOS 或 Linux (推荐 Ubuntu 20.04+)\n*   **版本控制**: Git\n*   **开发语言**: Python 3.8+ (用于脚本处理和运行部分示例)\n*   **硬件目标 (可选)**: ARM Cortex-M 系列开发板 (如 Arduino Nano 33 BLE, STM32), ESP32, 或 Raspberry Pi Pico。\n*   **依赖管理**: `pip`, `virtualenv` 或 `conda`\n\n> **国内加速建议**：\n> *   克隆仓库时，若 GitHub 连接缓慢，可使用国内镜像源（如 Gitee 镜像，若有）或配置代理。\n> *   安装 Python 依赖时，推荐使用清华源或阿里源：\n>     ```bash\n>     pip install -r requirements.txt -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n>     ```\n\n## 安装步骤（获取资源）\n\n你只需要将该仓库克隆到本地即可开始浏览和使用其中的资源。\n\n1.  **打开终端** (Terminal \u002F CMD \u002F PowerShell)。\n2.  **克隆仓库**：\n    ```bash\n    git clone https:\u002F\u002Fgithub.com\u002Fgigwegbe\u002Ftinyml-papers-and-projects.git\n    ```\n    *(如果速度较慢，可尝试使用国内镜像地址，例如：`git clone https:\u002F\u002Fgitee.com\u002Fmirrors\u002Ftinyml-papers-and-projects.git`，若镜像不存在则直接使用原地址)*\n\n3.  **进入目录**：\n    ```bash\n    cd tinyml-papers-and-projects\n    ```\n\n4.  **查看内容**：\n    直接在浏览器中打开根目录下的 `README.md` 文件，或者在终端使用 `cat README.md` 查看分类索引。\n\n## 基本使用\n\n该仓库的核心价值在于**指引**。使用流程通常为：**查阅论文\u002F项目 -> 定位官方代码链接 -> 克隆具体项目进行实验**。\n\n### 1. 浏览与检索\n打开 `README.md`，根据年份（如 **2020**, **2023**）或类别（**Awesome Projects**, **Libraries and Tools**）查找感兴趣的内容。\n*   **找算法**: 查看 \"Awesome Papers\" 部分，点击 `[pdf]` 阅读论文。\n*   **找代码**: 查看带有 `[official code]` 标记的项目。\n\n### 2. 实战示例：运行一个具体的 TinyML 项目\n假设你在列表中对 **2020** 年的 **MCUNet** (Tiny Deep Learning on IoT Devices) 感兴趣，以下是将其拉取并探索的典型步骤：\n\n**步骤 A: 获取具体项目代码**\n从列表中复制 MCUNet 的官方代码地址并克隆：\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002Fmit-han-lab\u002Fmcunet.git\ncd mcunet\n```\n\n**步骤 B: 安装该项目特定依赖**\n大多数列出的项目都包含自己的 `requirements.txt` 或 `setup.py`。\n```bash\n# 推荐使用国内源加速安装\npip install -e . -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n# 或者安装具体要求\npip install -r requirements.txt -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n```\n\n**步骤 C: 运行示例推理**\n参考该项目自身的 README 进行演示（以下为伪代码示例，具体命令需参照 MCUNet 原文档）：\n```bash\n# 示例：运行一个简单的图像分类演示\npython demo.py --model mcunet --image test.jpg\n```\n\n### 3. 常用工具链推荐\n在仓库的 **Resources -> Libraries and Tools** 章节，你可以找到直接用于开发的工具，例如：\n*   **TensorFlow Lite Micro**: 部署模型到微控制器。\n*   **CMSIS-NN**: ARM Cortex-M 优化的神经网络内核。\n*   **Larq**: 训练二值化神经网络。\n\n直接访问这些工具的官方文档进行安装即可，本仓库提供了最权威的入口索引。","一家初创团队正致力于开发一款基于微控制器的智能助听器，需要在极低的功耗和有限的内存（仅几百 KB）下实现实时的噪音消除与人声增强。\n\n### 没有 tinyml-papers-and-projects 时\n- **文献检索如大海捞针**：团队成员需分别在 arXiv、IEEE 和各大会议网站手动搜索\"TinyML\"、“模型压缩”等关键词，耗时数周仍难以覆盖 2016 年至 2024 年的关键论文（如 SqueezeNet 或 HAQ）。\n- **技术选型缺乏依据**：面对量化、剪枝等多种优化方案，无法快速找到针对 ARM Cortex-M 架构的基准测试数据，导致在算法选择上盲目试错。\n- **重复造轮子风险高**：由于不了解开源社区已有的成熟项目（如 CMSIS-NN 内核或 OpenMV 案例），团队可能花费大量精力重新实现基础功能。\n- **硬件适配周期漫长**：缺乏针对特定硬件的资源清单，工程师在将模型部署到 2KB RAM 设备时，常因内存溢出问题陷入漫长的调试循环。\n\n### 使用 tinyml-papers-and-projects 后\n- **一站式获取核心成果**：团队直接通过该列表按年份筛选，迅速锁定了《Deep Compression》和《Hello Edge》等奠基性论文，半天内完成了技术背景调研。\n- **精准匹配硬件方案**：借助列表中\"Hardware-Aware\"相关的论文与项目链接，直接采用了经过验证的混合精度量化策略，显著提升了推理效率。\n- **复用成熟代码库**：通过\"Awesome Projects\"板块找到了针对微控制器优化的神经网络内核源码，直接集成而非从头编写，节省了两个月开发时间。\n- **规避已知陷阱**：参考列表中关于资源受限设备的案例分析，提前规划了内存管理策略，一次性通过了在低功耗芯片上的部署测试。\n\ntinyml-papers-and-projects 将原本分散且晦涩的学术资源转化为结构化的工程指南，极大地缩短了从理论验证到嵌入式落地的研发周期。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fgigwegbe_tinyml-papers-and-projects_c3afad95.png","gigwegbe","George Igwegbe","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fgigwegbe_db384e50.png","Train Deep Neural Nets on Edge Devices.","CMU","Japan","gigwegbe@gmail.com","iGeorge_i","https:\u002F\u002Fgeorgeigwegbe.com","https:\u002F\u002Fgithub.com\u002Fgigwegbe",null,994,160,"2026-04-05T07:12:27","MIT",1,"","未说明",{"notes":95,"python":93,"dependencies":96},"该仓库是一个关于 TinyML（微型机器学习）的论文、项目和资源列表（Awesome List），本身不是一个可执行的软件工具或代码库，因此没有特定的操作系统、硬件或依赖库安装需求。用户主要利用此列表查找相关研究论文和外部项目链接。部分列出的子项目（如 TensorFlow Lite Micro, CMSIS-NN）通常针对微控制器（MCU）等嵌入式设备，而非通用计算机环境。",[],[14,13],[99,100,101,102,103,104],"tinyml","embedded-systems","neural-architecture-search","wake-word","machine-learning","computer-vision","2026-03-27T02:49:30.150509","2026-04-06T05:36:49.065199",[],[]]