微软 AI For Beginners：63k Stars 的 12 周 24 课 AI 入门课程

by Mycelium Protocol

---

AI 学习资源多，但系统性地从符号 AI 讲到深度学习、从神经网络讲到 Transformer、配有动手 Lab 且完全免费的——微软这套课程在 GitHub 上积累了 63k stars，是其中完整度最高的一个。

GitHub: https://github.com/microsoft/AI-For-Beginners | ⭐ 63,421 | MIT License

---

课程结构

12 周，24 课，5 个模块：

模块 I：AI 简介

• 第 1 课：AI 的历史与方法论

模块 II：符号 AI

• 第 2 课：知识表示与专家系统（含本体论和概念图 Notebook）

模块 III：神经网络基础

• 第 3 课：感知机
• 第 4 课：多层感知机与自制框架
• 第 5 课：PyTorch / TensorFlow / Keras 入门 + 过拟合

模块 IV：计算机视觉

• 第 6 课：OpenCV 计算机视觉基础
• 第 7 课：卷积神经网络 + CNN 架构
• 第 8 课：迁移学习与预训练网络
• 第 9 课：自编码器与 VAE
• 第 10 课：生成对抗网络 + 风格迁移
• 第 11 课：目标检测

模块 V：自然语言处理

• 第 12–17 课：词嵌入、RNN、LSTM → Transformer → 预训练语言模型

模块 VI：其他方法

• 第 18 课：遗传算法
• 第 19 课：多 Agent 系统
• 第 20–24 课：强化学习

---

学什么，不学什么

课程涵盖：

• “好旧”的符号 AI：知识表示和推理
• 神经网络与深度学习（PyTorch + TensorFlow 双轨）
• 计算机视觉的经典与现代模型
• 遗传算法与多 Agent 系统

课程不覆盖（有专项微软课程的领域）：

• AI 商业应用（商业场景）
• 经典机器学习（见 ML for Beginners 课程）
• Cognitive Services 实践（Azure 专项课）
• 云端 ML 平台（Azure ML / Fabric / Databricks）
• 对话 AI 和聊天机器人

---

快速开始

克隆（不含翻译文件，避免下载量过大）：

# macOS / Linux
git clone --filter=blob:none --sparse https://github.com/microsoft/AI-For-Beginners.git
cd AI-For-Beginners
git sparse-checkout set --no-cone '/*' '!translations' '!translated_images'

# Windows CMD
git clone --filter=blob:none --sparse https://github.com/microsoft/AI-For-Beginners.git
cd AI-For-Beginners
git sparse-checkout set --no-cone "/*" "!translations" "!translated_images"

仓库包含 55+ 语言的翻译，完整 clone 体积较大，建议用 sparse checkout。

在线运行：点击 README 里的 Binder 徽章，无需本地环境直接跑 Notebook。

---

为什么还值得学

在大模型普及的今天，这套课程依然有价值，原因在于它覆盖了符号 AI 和神经网络的历史演进脉络——理解为什么深度学习取代了专家系统、为什么 Transformer 又统一了序列建模，需要从头看这条路。这不是”速成调 API”，而是建立对 AI 系统的底层理解。

---

Mycelium Protocol — 追踪 AI 系统的底层演化

---

关于 Mycelium

菌丝协议。持续追踪 AI 工具、系统和实验的内容节点。

---

Microsoft AI For Beginners: A 12-Week, 24-Lesson Curriculum with 63k Stars

by Mycelium Protocol

---

AI learning resources are abundant, but a systematic curriculum that goes from symbolic AI through deep learning, neural networks to Transformers — with hands-on labs and completely free — is rare. Microsoft’s AI-For-Beginners repository has accumulated 63k stars and is one of the most complete options available.

GitHub: https://github.com/microsoft/AI-For-Beginners | ⭐ 63,421 | MIT License

---

Curriculum Structure

12 weeks, 24 lessons, 5 modules:

Module I: Introduction to AI

• Lesson 1: History and approaches to AI

Module II: Symbolic AI

• Lesson 2: Knowledge representation and expert systems (with Ontology and Concept Graph notebooks)

Module III: Neural Network Fundamentals

• Lesson 3: Perceptron
• Lesson 4: Multi-layered perceptron and building your own framework
• Lesson 5: Intro to PyTorch / TensorFlow / Keras + overfitting

Module IV: Computer Vision

• Lesson 6: OpenCV basics
• Lesson 7: Convolutional Neural Networks + CNN architectures
• Lesson 8: Transfer learning and pre-trained networks
• Lesson 9: Autoencoders and VAEs
• Lesson 10: GANs + artistic style transfer
• Lesson 11: Object detection

Module V: Natural Language Processing

• Lessons 12–17: Word embeddings → RNNs/LSTMs → Transformers → pre-trained language models

Module VI: Other Approaches

• Lessons 18–19: Genetic algorithms and multi-agent systems
• Lessons 20–24: Reinforcement learning

---

What It Covers and What It Doesn’t

Covered:

• “Good old” symbolic AI: knowledge representation and reasoning
• Neural networks and deep learning (PyTorch + TensorFlow dual track)
• Classic and modern computer vision models
• Genetic algorithms and multi-agent systems

Not covered (separate Microsoft courses exist for these):

• AI in business
• Classical machine learning (see ML for Beginners)
• Azure Cognitive Services hands-on
• Cloud ML platforms (Azure ML / Fabric / Databricks)
• Conversational AI and chatbots

---

Getting Started

Clone without translations (avoids large download):

# macOS / Linux
git clone --filter=blob:none --sparse https://github.com/microsoft/AI-For-Beginners.git
cd AI-For-Beginners
git sparse-checkout set --no-cone '/*' '!translations' '!translated_images'

The repo includes 55+ language translations, so the full clone is large. Sparse checkout gives you the full content without the translation files.

Run online: click the Binder badge in the README to run notebooks without any local setup.

---

Why It Still Matters

In an era of large models and API-first development, this curriculum remains valuable because it covers the historical arc from symbolic AI through neural networks — understanding why deep learning supplanted expert systems and why Transformers unified sequence modeling requires seeing that progression from the beginning. This isn’t a “call an API fast” tutorial; it builds foundational understanding of how AI systems actually work.

---

Mycelium Protocol — tracking the deep evolution of AI systems

© 2026 Mycelium Protocol. All rights reserved.