{"id":51511662,"url":"https://github.com/caciniep/math-to-deep-learning","last_synced_at":"2026-07-08T07:01:29.763Z","repository":{"id":369943810,"uuid":"1275308611","full_name":"CacinieP/math-to-deep-learning","owner":"CacinieP","description":"纯数学到深度学习：用应用数学串联数学知识与深度学习教程 — 同一个数学概念，从纸笔推导到GPU训练","archived":false,"fork":false,"pushed_at":"2026-07-07T14:24:41.000Z","size":201,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-07-07T15:23:44.577Z","etag":null,"topics":["calculus","deep-learning","linear-algebra","machine-learning","math","mathematics","neural-network","optimization","pytorch","tutorial"],"latest_commit_sha":null,"homepage":null,"language":null,"has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/CacinieP.png","metadata":{"files":{"readme":"README.md","changelog":"CHANGELOG.md","contributing":"CONTRIBUTING.md","funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-06-20T14:20:24.000Z","updated_at":"2026-07-07T14:25:20.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/CacinieP/math-to-deep-learning","commit_stats":null,"previous_names":["caciniep/math-to-deep-learning"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/CacinieP/math-to-deep-learning","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/CacinieP%2Fmath-to-deep-learning","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/CacinieP%2Fmath-to-deep-learning/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/CacinieP%2Fmath-to-deep-learning/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/CacinieP%2Fmath-to-deep-learning/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/CacinieP","download_url":"https://codeload.github.com/CacinieP/math-to-deep-learning/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/CacinieP%2Fmath-to-deep-learning/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":35255353,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-07-08T02:00:06.796Z","response_time":61,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["calculus","deep-learning","linear-algebra","machine-learning","math","mathematics","neural-network","optimization","pytorch","tutorial"],"created_at":"2026-07-08T07:01:28.836Z","updated_at":"2026-07-08T07:01:29.755Z","avatar_url":"https://github.com/CacinieP.png","language":null,"funding_links":[],"categories":[],"sub_categories":[],"readme":"# 纯数学到深度学习：用应用数学串联数学知识与深度学习教程\n\n\u003e 同一个数学概念，从纸笔推导（Mathematics-Universe）到 GPU 训练（PyTorch），是一条连续的理解线。\n\n---\n\n### 🪐 姊妹仓库：[CacinieP/Mathematics-Universe](https://github.com/CacinieP/Mathematics-Universe)\n\n本仓库的所有内容都以 **纯数学母仓库** 为地基。两个仓库配套使用，形成完整的「数学 → 深度学习」学习闭环：\n\n| 仓库 | 定位 | 回答的问题 |\n|------|------|-----------|\n| 🪐 **[Mathematics-Universe](https://github.com/CacinieP/Mathematics-Universe)** | 纯数学纵向知识图谱（高中→考研→超纲） | \"定理怎么证？概念怎么定义？\" |\n| 📡 **math-to-deep-learning**（本仓库） | 数学在 ML 中的应用棱镜 | \"这个概念怎么变成模型的一层？代码怎么写？\" |\n\n\u003e 📌 本仓库文章中的 `[[Mathematics-Universe/03-高等数学/...]]` 形式链接，均指向母仓库的详细推导。**建议同时 Star 两个仓库，对照阅读。**\n\n---\n\n### 📊 仓库状态\n\n![完成度](https://img.shields.io/badge/完成度-100%25-brightgreen)\n![正文](https://img.shields.io/badge/正文-39篇-blue)\n![更新](https://img.shields.io/badge/最后更新-2026--07--07-orange)\n\n**最新更新（[2026-07-07] 全量补全）**：完成度从 19% → 100%，README 规划的 PART-01 至 PART-05 + APPENDIX 全部填满。详见 [CHANGELOG](./CHANGELOG.md)。\n\n---\n\n## 设计理念\n\n本仓库是 [CacinieP/Mathematics-Universe](https://github.com/CacinieP/Mathematics-Universe) 的**应用侧延伸**。纯数学仓库回答\"它是什么、为什么成立\"，本仓库回答\"它怎么用、代码怎么写\"。\n\n### 三层叙事结构\n\n每篇文章遵循统一的叙事节奏：\n\n```\n纯数学层（\"它是什么\"）\n  → 引用 Mathematics-Universe 的详细推导\n应用映射层（\"它在ML中变成什么\"）\n  → 图示 + 类比 + 直觉\n工程实现层（\"代码怎么写\"）\n  → Python/PyTorch 最小可运行示例\n```\n\n## 目录结构\n\n### PART-01：数学基础回顾（机器学习视角）\n\n对纯数学仓库的每章内容做\"ML裁剪版\"——只保留深度学习需要的那部分，去除纯数学中不需要的证明细节。\n\n| 编号 | 内容 | 对应纯数学仓库 |\n|------|------|---------------|\n| 01 | [线性代数（机器学习视角）](./PART-01-数学基础回顾/01-线性代数（机器学习视角）/) | 04-线性代数 |\n| 02 | [概率论与统计（机器学习视角）](./PART-01-数学基础回顾/02-概率论与统计（机器学习视角）/) | 05-概率论与数理统计 |\n| 03 | [微积分与优化（机器学习视角）](./PART-01-数学基础回顾/03-微积分与优化（机器学习视角）/) | 03-高等数学 |\n| 04 | [信息论（机器学习视角）](./PART-01-数学基础回顾/04-信息论（机器学习视角）/) | 06-超纲拓展 |\n| 05 | [函数空间与逼近论（机器学习视角）](./PART-01-数学基础回顾/05-函数空间与逼近论（机器学习视角）/) | 03+06 跨章节 |\n\n### PART-02：深度学习核心\n\n从零构建深度学习的数学骨架，每节都明确标注其背后的纯数学来源。\n\n| 编号 | 内容 | 核心数学 |\n|------|------|----------|\n| 01 | [神经网络基础](./PART-02-深度学习核心/01-神经网络基础/) | 函数复合、线性映射 |\n| 02 | [反向传播](./PART-02-深度学习核心/02-反向传播/) | 链式法则（高数）→ Jacobi矩阵（线代） |\n| 03 | [优化算法](./PART-02-深度学习核心/03-优化算法/) | 梯度下降、凸优化、Taylor展开 |\n| 04 | [正则化](./PART-02-深度学习核心/04-正则化/) | 偏差-方差分解、Sobolev空间 |\n| 05 | [损失函数](./PART-02-深度学习核心/05-损失函数/) | 概率分布、KL散度、信息论 |\n\n### PART-03：专题映射（五条轴线）\n\n**核心部分**——按\"纯数学概念 → 深度学习应用\"的轴线组织，每条轴线是一条从理论到实践的完整理解链。\n\n| 轴线 | 主题 | 纯数学核心 | 深度学习终点 |\n|------|------|-----------|-------------|\n| **A** | [矩阵分解](./PART-03-专题映射/轴线A-矩阵分解/) | 特征值/SVD/QR/Cholesky | PCA → 自编码器 → Neural ODE |\n| **B** | [概率模型](./PART-03-专题映射/轴线B-概率模型/) | Bayes/MLE/正态/变分推断 | 贝叶斯网络 → VAE → 扩散模型 |\n| **C** | [优化理论](./PART-03-专题映射/轴线C-优化理论/) | 梯度/凸优化/Taylor/Lagrange | 反向传播 → Adam → 约束优化 |\n| **D** | [函数逼近](./PART-03-专题映射/轴线D-函数逼近/) | Weierstrass/Fourier/正交基 | 通用近似定理 → 注意力 → 残差 |\n| **E** | [信息论](./PART-03-专题映射/轴线E-信息论/) | 熵/KL散度/互信息 | 交叉熵 → 信息瓶颈 → RLHF |\n\n### PART-04：从理论到工程\n\n数学理论落地时的工程问题——为什么\"理论上成立\"的东西在GPU上会崩。\n\n| 编号 | 内容 | 涉及的数学 |\n|------|------|-----------|\n| 01 | [数值稳定性](./PART-04-从理论到工程/01-数值稳定性/) | 浮点误差、条件数 |\n| 02 | [梯度消失/爆炸的数学根源](./PART-04-从理论到工程/02-梯度消失爆炸的数学根源/) | 矩阵乘积范数、谱半径 |\n| 03 | [BatchNorm的统计视角](./PART-04-从理论到工程/03-BatchNorm的统计视角/) | 协方差、白化、分布对齐 |\n| 04 | [注意力机制的线性代数本质](./PART-04-从理论到工程/04-注意力机制的线性代数本质/) | Softmax=概率归一化、QKV=基变换 |\n| 05 | [Transformer的谱分析](./PART-04-从理论到工程/05-Transformer的谱分析/) | 注意力矩阵的特征值、SVD |\n\n### PART-05：前沿中的数学\n\n最新模型背后的数学前沿——持续更新。\n\n| 编号 | 内容 | 数学前沿 |\n|------|------|---------|\n| 01 | [Diffusion模型的随机微分方程](./PART-05-前沿中的数学/01-Diffusion模型的SDE视角/) | SDE、Ornstein-Uhlenbeck过程 |\n| 02 | [流模型与微分同胚](./PART-05-前沿中的数学/02-流模型与微分同胚/) | 微分几何、Jacobian行列式 |\n| 03 | [图神经网络的谱图理论](./PART-05-前沿中的数学/03-图神经网络的谱图理论/) | 图Laplacian、谱聚类 |\n| 04 | [大语言模型的缩放定律](./PART-05-前沿中的数学/04-大语言模型的缩放定律/) | 幂律、相变、临界现象 |\n| 05 | [RLHF的博弈论视角](./PART-05-前沿中的数学/05-RLHF的博弈论视角/) | 纳什均衡、潜在博弈 |\n\n### APPENDIX：工具箱\n\n- [公式速查](./APPENDIX-工具箱/公式速查.md) — 深度学习中最常用的数学公式速查表\n- [代码实现映射](./APPENDIX-工具箱/代码实现映射.md) — 数学概念 → PyTorch API 对照表\n- [常见陷阱与反例](./APPENDIX-工具箱/常见陷阱与反例.md) — 数学直觉在ML中失效的场景\n\n## 如何阅读\n\n### 如果你是深度学习初学者\n\n```\nPART-01（选你薄弱的那章）→ PART-02（01-05按顺序）→ PART-03 轴线A\n```\n\n### 如果你有ML基础，想深入理解\"为什么\"\n\n```\n直接进 PART-03，选你最感兴趣的一条轴线，顺着读到底\n```\n\n### 如果你要攻克具体难题\n\n```\nPART-04（梯度消失、数值稳定性等工程问题）→ PART-05（前沿模型）\n```\n\n### 如果你在备考或复习数学\n\n```\nMathematics-Universe（纯数学）→ PART-01（ML视角裁剪版）→ PART-03（轴线映射）\n```\n\n## 与 Mathematics-Universe 的关系\n\n| 仓库 | 定位 | 回答的问题 |\n|------|------|-----------|\n| [Mathematics-Universe](https://github.com/CacinieP/Mathematics-Universe) | 纯数学纵向知识图谱 | \"定理怎么证？概念怎么定义？\" |\n| **math-to-deep-learning** | 数学在ML中的应用棱镜 | \"这个概念怎么变成模型的一层？代码怎么写？\" |\n\n两个仓库通过双向链接互联：本文中的 `[[Mathematics-Universe/03-高等数学/...]]` 指向纯数学仓库的详细推导。\n\n## 难度标注\n\n- `[基础]` — 需要高中数学 + 一点编程经验\n- `[标准]` — 需要大学数学 + ML入门（能写MNIST分类器）\n- `[进阶]` — 需要扎实的数学 + 中等ML经验\n- `[前沿]` — 需要数学 + ML + 追踪论文的能力\n\n## 贡献\n\n见 [CONTRIBUTING.md](./CONTRIBUTING.md)\n\n## 许可证\n\nCC-BY-SA-4.0 — 与 Mathematics-Universe 一致\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fcaciniep%2Fmath-to-deep-learning","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fcaciniep%2Fmath-to-deep-learning","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fcaciniep%2Fmath-to-deep-learning/lists"}