~37 min read

AI 时代专业课程设置参考资料

调研范围：全球独立大牛 / 个人 IP / 研究机构 / 公司 / 大学开源 / 工程社区开设的 AI 课程。

只列课表（每门课讲什么、有几节、时长、是否有 capstone），不写分析、评价、价格比较。

所有人名 / 课程名保留英文。

A 类 — 神级个人课程#

A1. Neural Networks: Zero to Hero — Andrej Karpathy#

• 平台：YouTube + GitHub
• URL：https://karpathy.ai/zero-to-hero.html
• 总长度：8 个视频，约 13 小时
• 覆盖范围：从 micrograd 手搓反向传播一路到从零构建 GPT 和 BPE Tokenizer

完整视频清单：

1. 1. The spelled-out intro to neural networks and backpropagation: building micrograd（145 分钟）
2. 2. The spelled-out intro to language modeling: building makemore（117 分钟）— bigram 字符级语言模型，PyTorch tensors 入门
3. 3. Building makemore Part 2: MLP（75 分钟）— 多层感知机、学习率、超参、过拟合
4. 4. Building makemore Part 3: Activations & Gradients, BatchNorm（115 分钟）
5. 5. Building makemore Part 4: Becoming a Backprop Ninja（56 分钟）— 不用 autograd 手动反传
6. 6. Building makemore Part 5: Building a WaveNet（56 分钟）
7. 7. Let's build GPT: from scratch, in code, spelled out（116 分钟）— 跟着 "Attention is All You Need" 和 GPT-2/GPT-3 paper
8. 8. Let's build the GPT Tokenizer（133 分钟）— BPE 实现

Capstone / 作业：无正式作业，每个 video 末尾有 suggested exercises。

A2. Software Is Changing (Again) — Andrej Karpathy（Y Combinator 2025 keynote）#

• 平台：YouTube / YC Startup Library
• URL：https://www.ycombinator.com/library/MW-andrej-karpathy-software-is-changing-again
• 总长度：1 个 keynote 演讲（不是课程）

章节：

1. 1. Software Evolution: From 1.0 to 3.0
2. 2. Programming in English: Rise of Software 3.0
3. 3. LLMs as Infrastructure（LLM OS, fabs, utilities）
4. 4. Psychology of LLMs（"people spirits", jagged intelligence, hallucinations）
5. 5. Partial Autonomy & Human-AI Collaboration
6. 6. Tesla Autopilot & Autonomy Sliders
7. 7. Vibe Coding（MenuGen demo）
8. 8. Building for Agents
9. 9. Future Outlook（mainframe era of LLMs）

A3. Stanford CS231n — Convolutional Neural Networks for Visual Recognition（Karpathy 早期版本，Winter 2016）#

• URL：https://cs231n.stanford.edu/2016/syllabus.html
• 总长度：15 个 lecture

Lecture 清单：

1. 1. Intro to Computer Vision, historical context
2. 2. Image Classification — k-NN, Linear Classification
3. 3. Linear classification II, optimization, SGD
4. 4. Backpropagation, Intro to Neural Networks
5. 5. Training Neural Networks Part 1（activations, init, batch norm）
6. 6. Training Neural Networks Part 2（parameter updates, dropout）+ CNN intro
7. 7. CNN architectures, convolution/pooling, ImageNet case studies
8. 8. Localization and Detection
9. 9. Visualization, DeepDream, Style Transfer, Adversarial Examples
10. 10. RNN, LSTM — Language modeling, Image captioning
11. 11. Practical ConvNet training（augmentation, transfer learning, distributed）
12. 12. Frameworks（Caffe, Torch, Theano, TensorFlow）
13. 13. Segmentation, Soft Attention, Spatial Transformer Networks
14. 14. Videos and Unsupervised Learning
15. 15. Invited talks / Conclusions

Lecture Notes（学生自学路径）：

• Module 0: Preparation（Python/Numpy）
• Module 1: Neural Networks（图像分类、线性分类、optimization、backprop、neural network 三部曲、case study）
• Module 2: Convolutional Neural Networks（架构、可视化、transfer learning）

Capstone：3 个 Assignment（kNN/SVM/Softmax/2-layer NN；CNN/Dropout/BN；RNN/LSTM/Captioning）+ Final Project。

A4. Stanford CS231n — Deep Learning for Computer Vision（最新 Spring 2026 版）#

• URL：https://cs231n.stanford.edu/schedule.html
• 总长度：18 个 lecture + Midterm + Final Project poster

Lecture 清单（按模块分）：

Module 1 — Deep Learning Basics

1. 1. Introduction
2. 2. Image Classification with Linear Classifiers
3. 3. Regularization and Optimization
4. 4. Neural Networks and Backpropagation

Module 2 — Perceiving and Understanding the Visual World

1. 5. Image Classification with CNNs
2. 6. CNN Architectures
3. 7. Recurrent Neural Networks
4. 8. Attention and Transformers
5. 9. Object Detection, Image Segmentation
6. 10. Video Understanding
7. 11. Large Scale Distributed Training

Module 3 — Generative and Interactive Visual Intelligence

1. 12. Self-supervised Learning
2. 13. Generative Models 1（VAEs, GANs, autoregressive）
3. 14. Generative Models 2（Diffusion）
4. 15. 3D Vision
5. 16. Vision and Language
6. 17. World Modeling
7. 18. Human-Centered AI

Capstone：3 个 Assignments + Project Proposal + Final Project + Poster Session。

A5. Practical Deep Learning for Coders — Fast.ai Part 1（Jeremy Howard）#

• URL：https://course.fast.ai/
• 总长度：9 个 lesson，每个约 90 分钟

Lesson 清单：

1. 1. Getting started
2. 2. Deployment
3. 3. Neural net foundations
4. 4. Natural Language (NLP)
5. 5. From-scratch model
6. 6. Random forests
7. 7. Collaborative filtering
8. 8. Convolutions (CNNs)
9. 9. Bonus: Data ethics

Capstone：无正式 capstone，每 lesson 有 summary 文档 + Kaggle / 自建项目。

工具栈：PyTorch, fastai, Hugging Face Transformers, Gradio。

A6. Fast.ai Part 2 — Deep Learning from the Foundations / Stable Diffusion（Jeremy Howard）#

• URL：https://course19.fast.ai/part2 + https://course.fast.ai/Lessons/part2.html
• 总长度：17 个 lesson（最新版包含 Stable Diffusion）

Lesson 清单（早期 "Deep Learning from the Foundations" 版本，lesson 8-14）：

1. 8. Matrix multiplication; forward and backward passes
2. 9. Loss functions, optimizers, and the training loop
3. 10. Looking inside the model（callbacks, hooks, batchnorm）
4. 11. Data Block API, and generic optimizer
5. 12. Advanced training techniques; ULMFiT from scratch
6. 13. Basics of Swift for Deep Learning
7. 14. C interop; Protocols; Putting it all together

最新版（"Deep Learning Foundations to Stable Diffusion"）扩展内容：matrix multiplication, clustering, backpropagation, autoencoders, attention, transformers, latent diffusion models。

A7. Deep Learning Specialization — Andrew Ng（DeepLearning.AI）#

• URL：https://www.coursera.org/specializations/deep-learning
• 总长度：5 门课，约 129 小时

课程清单：

1. 1. Neural Networks and Deep Learning（25h）— 全连接网络、向量化、架构参数
2. 2. Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization（24h）— Momentum, RMSprop, Adam, TensorFlow
3. 3. Structuring Machine Learning Projects（7h）— 错误诊断、bias/variance、transfer / multi-task learning
4. 4. Convolutional Neural Networks（36h）— CNN、neural style transfer、2D/3D 视觉
5. 5. Sequence Models（37h）— RNN/LSTM/GRU、HuggingFace tokenizers、Transformer、NER、QA

Capstone：每门课有 programming assignments，无单独 capstone。

A8. Machine Learning Specialization — Andrew Ng（DeepLearning.AI / Stanford Online，2022 重制版）#

• URL：https://www.coursera.org/specializations/machine-learning-introduction
• 总长度：3 门课，约 95 小时

课程清单：

1. 1. Supervised Machine Learning: Regression and Classification（33h）— 线性回归、逻辑回归、特征工程、训练优化
2. 2. Advanced Learning Algorithms（34h）— Neural networks with TensorFlow、多分类、决策树、ensemble、transfer learning
3. 3. Unsupervised Learning, Recommenders, Reinforcement Learning（28h）— 聚类、异常检测、降维、推荐系统、deep RL

A9. Machine Learning Engineering for Production (MLOps) Specialization — Andrew Ng / Robert Crowe（DeepLearning.AI）#

• URL：https://www.coursera.org/specializations/machine-learning-engineering-for-production-mlops
• 总长度：4 门课的专项；其中 Machine Learning in Production 单课为 3 周，~15 小时

Machine Learning in Production 单课课表：

• Week 1: Overview of the ML Lifecycle and Deployment（8 videos, 2 assignments, 2 labs）
• Week 2: Modeling Challenges and Strategies（16 videos, 2 assignments, 1 lab）
• Week 3: Data Definition and Baseline（17 videos, 2 assignments, 2 labs）

后续课程涵盖 ML Data Lifecycle, ML Modeling Pipelines, Deploying ML Models。

A10. Generative AI with Large Language Models — Andrew Ng / AWS（DeepLearning.AI）#

• URL：https://www.coursera.org/learn/generative-ai-with-llms
• 总长度：3 周，约 17 小时

Week 清单：

• Week 1: Generative AI Use Cases, Project Lifecycle, and Model Pre-training（17 videos / 116 min, 7 readings, 1 quiz, Lab 1: Summarize Dialogue）— Transformer 架构、pre-training、scaling laws
• Week 2: Fine-tuning and Evaluating Large Language Models（10 videos / 78 min, 1 quiz, Lab 2: Fine-tune for dialogue summarization）— Instruction fine-tuning、PEFT（LoRA, soft prompts）
• Week 3: Reinforcement Learning and LLM-powered Applications（21 videos / 141 min, 1 quiz, Lab 3: Fine-tune FLAN-T5 with RLHF）— RLHF、deployment、CoT、responsible AI

Capstone：3 个 AWS labs（无独立项目）。

A11. DeepLearning.AI 其他 Specializations / Professional Certificates#

• URL：https://www.deeplearning.ai/courses/

完整列表：

1. 1. Machine Learning Specialization
2. 2. Deep Learning Specialization
3. 3. Natural Language Processing Specialization
4. 4. TensorFlow Developer Professional Certificate
5. 5. PyTorch for Deep Learning
6. 6. Generative AI for Software Development
7. 7. Data Analytics
8. 8. Mathematics for Machine Learning and Data Science
9. 9. AI for Good
10. 10. AI for Medicine
11. 11. Machine Learning Engineering for Production (MLOps)

典型 short courses（每个 1-2 小时）：

• AI Prompting for Everyone（Andrew Ng）
• Agentic AI（Andrew Ng）
• AI for Everyone（Andrew Ng）
• Generative AI for Everyone（Andrew Ng）
• LangChain for LLM Application Development（Andrew Ng + Harrison Chase）
• AI Agents in LangGraph（Harrison Chase + Rotem Weiss）
• Multi AI Agent Systems with crewAI
• Retrieval Augmented Generation (RAG)
• Building with Llama 4（Amit Sangani / Meta）
• AI Agents for Image and Video Generation（Google）
• Build Interactive Agents with Generative UI（CopilotKit）
• Building Multimodal Data Pipelines（Snowflake）
• Spec-Driven Development with Coding Agents（JetBrains）
• Efficient Inference with SGLang
• Agent Memory（Oracle）
• Build and Train an LLM with JAX（Google）

A12. Build a Large Language Model From Scratch — Sebastian Raschka#

• URL：https://github.com/rasbt/LLMs-from-scratch
• 总长度：7 章 + 5 附录，528 页（Manning, 2024）

章节清单：

1. 1. Understanding Large Language Models（无 code）
2. 2. Working with Text Data — tokenization、data loading、preprocessing
3. 3. Coding Attention Mechanisms — single-head 和 multi-head attention
4. 4. Implementing a GPT Model from Scratch
5. 5. Pretraining on Unlabeled Data
6. 6. Finetuning for Text Classification
7. 7. Finetuning to Follow Instructions（instruction tuning, alignment）

附录：

• A. Introduction to PyTorch（含分布式训练）
• B. References
• C. Exercise Solutions
• D. Adding Bells and Whistles to the Training Loop
• E. Parameter-efficient Finetuning with LoRA

Bonus 材料：tokenizer 实现变种、attention 优化、Llama/Qwen/Gemma 架构、KV caching、MoE、preference tuning、evaluation frameworks。

配套课程：Manning Live Video Course "Master and Build Large Language Models" + 免费 YouTube playlist。

Capstone：每章末尾习题；完整端到端 GPT 实现就是项目。

A13. 其他 Sebastian Raschka 课程 / 书籍#

• Build a Reasoning Model (From Scratch)（in progress, 528 页）— reasoning LLMs、inference-time scaling、RL、distillation
• Machine Learning Q and AI（264 页, No Starch, 2024）— Q&A 形式：multi-GPU training、fine-tuning transformers、vision transformers
• Machine Learning with PyTorch and Scikit-Learn（770 页, Packt, 2022）— ML 基础到 deep learning

A14. Designing Machine Learning Systems — Chip Huyen#

• URL：https://huyenchip.com/books / O'Reilly 出版
• 总长度：11 章 + Epilogue

章节清单：

1. 1. Overview of Machine Learning Systems
2. 2. Introduction to Machine Learning Systems Design — business/ML objectives, reliability, scalability, maintainability, adaptability, ML problem framing
3. 3. Data Engineering Fundamentals — sources, formats, models, dataflow, batch vs stream
4. 4. Training Data — sampling, labeling, class imbalance, augmentation
5. 5. Feature Engineering — scaling, encoding, crossing, data leakage, generalization
6. 6. Model Development and Offline Evaluation — ensembles, tracking, distributed training, AutoML, baselines
7. 7. Model Deployment and Prediction Service — batch vs online, compression（factorization, distillation, pruning, quantization）, edge
8. 8. Data Distribution Shifts and Monitoring — shift detection、observability
9. 9. Continual Learning and Test in Production — stateless vs stateful、shadow、A/B、canary、interleaving、bandits
10. 10. Infrastructure and Tooling for MLOps — compute、dev env、containers、orchestrators、ML platform、feature store
11. 11. The Human Side of Machine Learning — UX、team structure、responsible AI

配套课程：Stanford CS329S（见 C 类）。

A15. AI Engineering — Chip Huyen#

• URL：https://github.com/chiphuyen/aie-book
• 总长度：10 章（O'Reilly, 2025）

章节清单：

1. 1. Introduction to Building AI Applications with Foundation Models — The Rise of AI Engineering, FM Use Cases, Planning AI Applications, The AI Engineering Stack
2. 2. Understanding Foundation Models — Training Data, Modeling, Post-Training, Sampling
3. 3. Evaluation Methodology — Challenges, Language Modeling Metrics, Exact Evaluation, AI as a Judge, Comparative Evaluation
4. 4. Evaluate AI Systems — Evaluation Criteria, Model Selection, Design Your Evaluation Pipeline
5. 5. Prompt Engineering — Introduction, Best Practices, Defensive Prompt Engineering
6. 6. RAG and Agents — RAG, Agents, Memory
7. 7. Finetuning — Overview, When to Finetune, Memory Bottlenecks, Techniques
8. 8. Dataset Engineering — Data Curation, Augmentation/Synthesis, Processing
9. 9. Inference Optimization — Understanding, Optimization
10. 10. AI Engineering Architecture and User Feedback — Architecture, User Feedback

A16. Eugene Yan — 自学路径文章系列#

• URL：https://eugeneyan.com/writing/

不是正式课程，但被广泛当成结构化自学材料，按主题：

LLMs：

• Prompting Fundamentals and How to Apply them Effectively
• Task-Specific LLM Evals that Do & Don't Work
• Evaluating the Effectiveness of LLM-Evaluators
• Patterns for Building LLM-based Systems & Products
• What We've Learned From A Year of Building with LLMs
• Evaluating Long-Context Question & Answer Systems

ML Systems：

• A Practical Guide to Maintaining Machine Learning in Production
• 6 Little-Known Challenges After Deploying Machine Learning
• Design Patterns in Machine Learning Code and Systems
• More Design Patterns For Machine Learning Systems
• How to Test Machine Learning Code and Systems
• Writing Robust Tests for Data & Machine Learning Pipelines

Recommender Systems：

• System Design for Recommendations and Search
• Patterns for Personalization in Recommendations and Search
• Real-time Machine Learning For Recommendations
• Improving Recommendation Systems & Search in the Age of LLMs

Data Science：

• How to get started in Data Science
• What does a Data Scientist really do?
• What I Do Before a Data Science Project to Ensure Success

A17. Mastering LLMs For Developers & Data Scientists — Hamel Husain + Dan Becker#

• URL：https://hamel.dev/blog/posts/course/ + https://parlance-labs.com/education/
• 总长度：40+ 小时开源（原 Maven 4 工作坊）

5 大主题（共 23+ sessions）：

1. 1. Evals（3 sessions）— LLM evaluation 框架与方法
2. 2. RAG（3 talks）— RAG 基础到系统改进
3. 3. Building Applications（4 resources）— 各种工具
4. 4. Prompt Engineering（1 session）
5. 5. Fine-Tuning（12+ sessions, 4 子分类）— Decision-making、Tools/Techniques（含 Wing Lian 讲 Axolotl、Zach Mueller 讲 HF Accelerate）、Deployment、Advanced

A18. AI Evals For Engineers & PMs — Hamel Husain + Shreya Shankar（Maven）#

• URL：https://maven.com/parlance-labs/evals
• 总长度：4 周，3-5 小时/周，77 lessons，10+ office hours，150+ 页 course reader

周清单（部分公开）：

• Week 1：Fundamentals & Lifecycle of Application-Centric Evals（Lesson 1）+ Systematic Error Analysis（Lesson 2-3）
• Week 2：Automated Evaluators（Lesson 4-5）
• Week 3-4：（完整 syllabus 课内）

Capstone：4 个 homework assignments + Recipe Bot Workflow 案例。

A19. The LLM Course — Maxime Labonne#

• URL：https://github.com/mlabonne/llm-course
• 总长度：3 轨道，几十个章节 + 大量 hands-on notebooks

轨道 1：🧩 LLM Fundamentals

1. 1. Mathematics for Machine Learning（线代、微积分、概率统计）
2. 2. Python for Machine Learning（基础、NumPy/Pandas/Matplotlib/Seaborn、预处理、Scikit-learn）
3. 3. Neural Networks（基础、训练优化、过拟合正则、MLP 实现）
4. 4. NLP（预处理、特征提取、Word Embeddings、RNN/LSTM/GRU）

轨道 2：🧑‍🔬 The LLM Scientist

1. 1. The LLM Architecture（encoder-decoder→decoder-only、tokenization、attention 变种、sampling）
2. 2. Pre-Training Models（数据准备、distributed training: DP/PP/TP、训练优化、监控）
3. 3. Post-Training Datasets（ShareGPT/ChatML/Alpaca、合成数据、增强、质量过滤）
4. 4. Supervised Fine-Tuning（full FT、LoRA、QLoRA、DeepSpeed、FSDP）
5. 5. Preference Alignment（rejection sampling、DPO、reward model、GRPO、PPO）
6. 6. Evaluation（自动 benchmarks、人评、model-based、反馈信号）
7. 7. Quantization（FP32/16/INT8、absmax、GGUF、llama.cpp、GPTQ、AWQ、SmoothQuant、ZeroQuant）
8. 8. New Trends（SLERP/DARE/TIES merging、CLIP/SD/LLaVA 多模态、可解释性 SAE/abliteration、test-time compute）

轨道 3：👷 The LLM Engineer

1. 1. Running LLMs（API: OpenAI/Google/Anthropic/OpenRouter/HF；开源: Ollama/LM Studio/llama.cpp；prompt engineering；结构化输出）
2. 2. Building a Vector Storage（loading, splitting, embeddings, Chroma/Pinecone/Milvus/FAISS/Annoy）
3. 3. RAG（LangChain/LlamaIndex/MCP、retriever 策略、记忆、Ragas/DeepEval）
4. 4. Advanced RAG（SQL/Cypher、tools、re-ranking、RAG-fusion、DSPy）
5. 5. Agents（基础、协议 MCP/A2A、OpenAI/Google ADK/Claude SDK、LangGraph/LlamaIndex/CrewAI/AutoGen）
6. 6. Inference Optimization（Flash Attention、KV cache、MQA/GQA、EAGLE-3 speculative decoding）
7. 7. Deploying LLMs（local、Gradio/Streamlit/HF Spaces、TGI/vLLM/SkyPilot、MLC LLM/mnn-llm）
8. 8. Securing LLMs（prompt hacking、backdoor、red teaming、garak、langfuse）

Tools：LLM AutoEval, LazyMergekit, LazyAxolotl, AutoQuant, Model Family Tree, ZeroSpace, AutoAbliteration, AutoDedup

Notebooks：Fine-tune Llama 3.1 with Unsloth、Llama 3 with ORPO、Mistral-7b with DPO/QLoRA、CodeLlama with Axolotl、Llama 2 with QLoRA、4-bit GPTQ Quantization、GGUF 量化、ExLlamaV2、MergeKit MoE、abliteration、Knowledge Graphs、Decoding Strategies。

B 类 — 研究机构 / 公司课程#

B1. Hugging Face — 全部课程清单#

平台：https://huggingface.co/learn

B2. Hugging Face LLM Course（NLP Course → LLM Course）#

• URL：https://huggingface.co/learn/llm-course
• 总长度：12 章，每章约 1 周 / 6-8 小时

章节清单：

• Chapter 1：NLP and LLMs intro、Transformer 基础（pipeline 用法、架构原理、encoder/decoder/encoder-decoder 适用任务、scaling laws、attention 变种、pretraining + instruction tuning）
• Chapter 2：Using Transformers — Tokenizers、Models、pipeline 内部、tokenization 算法
• Chapter 3：Fine-tuning a pretrained model — Trainer API、数据处理、整合
• Chapter 4：Sharing models and tokenizers — HF Hub
• Chapter 5：The 🤗 Datasets library — big data 处理
• Chapter 6：The 🤗 Tokenizers library — 从头训练 tokenizer
• Chapter 7：Main NLP Tasks — token classification、masked LM、translation、summarization、causal LM from scratch、QA、Mastering LLMs
• Chapter 8：How to ask for help
• Chapter 9：Building and sharing demos — Gradio
• Chapter 10：Curate high-quality datasets
• Chapter 11：Fine-tune LLMs
• Chapter 12：Build reasoning models

Capstone：每章末尾 quizzes；建议项目在 forum 上有清单。

B3. Hugging Face Agents Course#

• URL：https://huggingface.co/learn/agents-course
• 总长度：5 chapters + 3 bonus units，每章约 1 周 / 3-4 小时

Chapter 清单：

Bonus Units：

• Bonus 1：Fine-tuning an LLM for Function-calling
• Bonus 2：Agent Observability and Evaluation
• Bonus 3：Agents in Games with Pokemon

Capstone：Final benchmark agent + student leaderboard；有 certificate of fundamentals + certificate of completion 两种。

B4. Hugging Face Deep RL Course#

• URL：https://huggingface.co/learn/deep-rl-course
• 总长度：约 9 单元，每单元 3-4 小时/周

Unit 概要：

• 工具栈：Stable Baselines3、RL Baselines3 Zoo、Sample Factory、CleanRL
• 环境：SnowballFight、Huggy the Doggo、VizDoom、Space Invaders、PyBullet
• 包含 AI vs AI 挑战（Unit 7，目前停维护）
• Certificate of Completion（80% 作业）/ Certificate of Honors（100%）

B5. Hugging Face Audio Course#

• URL：https://huggingface.co/learn/audio-course
• 总长度：8 单元

Unit 清单：

• Unit 1：音频数据基础、音频处理、数据准备
• Unit 2：Audio pipelines for transformer 应用（分类、ASR）
• Unit 3：Audio transformer 架构
• Unit 4：Build your own music genre classifier
• Unit 5：Speech recognition — 转录会议录音
• Unit 6：Text-to-Speech 生成
• Unit 7：构建真实音频应用
• Unit 8：（高级 / 最终单元）

Capstone：每章有 hands-on 项目；80% / 100% 作业完成获 certificate。

B6. Hugging Face Diffusion Models Course#

• URL：https://huggingface.co/learn/diffusion-course
• 总长度：4 units，每单元约 1 周 / 6-8 小时

Unit 清单：

• Unit 1：Introduction to diffusion models — 🤗 Diffusers + 从零实现
• Unit 2：Finetuning and guidance — 微调 + guidance
• Unit 3：Stable Diffusion — text-conditioned latent diffusion
• Unit 4：Doing more with diffusion — 高级技术

每单元含 theory section + 2 notebooks。

B7. Hugging Face Community Computer Vision Course#

• URL：https://huggingface.co/learn/computer-vision-course
• 总长度：13 units

Unit 清单：

1. 1. Fundamentals of Computer Vision — 图像基础、形成、预处理、特征提取
2. 2. Convolutional Neural Networks (CNNs) — 架构、迁移学习
3. 3. Vision Transformers — Swin、DETR、CVT、transfer learning
4. 4. Multimodal Models — CLIP、GroupViT、BLIP、Owl-VIT
5. 5. Generative Models — GANs、VAEs、Diffusion、text-to-image、inpainting
6. 6. Basic Computer Vision Tasks — classification、object detection、segmentation、YOLO、SAM
7. 7. Video and Video Processing — temporal continuity、motion estimation
8. 8. 3D Vision, Scene Rendering, and Reconstruction — NeRF、GQN
9. 9. Model Optimization — compression、distillation、pruning、TinyML
10. 10. Synthetic Data Creation — point clouds、diffusion 合成数据
11. 11. Zero Shot Computer Vision
12. 12. Ethics and Biases in Computer Vision
13. 13. Outlook and Emerging Trends — Retentive Network、Hiera、Hyena、I-JEPA

B8. Cohere LLM University#

• URL：https://cohere.com/llmu
• 总长度：~7+ 模块，持续更新

Module 清单：

1. 1. What are Large Language Models? — LLM 基础、embeddings、attention、transformer 架构、semantic search、实战
2. 2. Text Representation with Cohere Endpoints — Classification / Embeddings / Semantic Search endpoints
3. 3. Text Generation with Cohere Endpoints — generative learning、generated endpoint、prompt engineering
4. 4. Deployment — AWS SageMaker、Streamlit、FastAPI
5. 5. Semantic Search（2023 新增）
6. 6. Prompt Engineering（2023 新增）
7. 7. The Cohere Platform（2023 新增）

Tool Use 章节（2024-2025）：

• From RAG to Tool Use（Part 1）— 构建查询销售/产品数据的 RAG assistant
• Tool Use Anatomy（Part 2）— 关键组件 + 代码示例
• Multi-Step Tool Use（Part 4）

AWS 集成模块：

• Introduction to Cohere on Amazon Bedrock
• Tool Use and Agents on Amazon Bedrock（Command R+）

单独 chapter：

• Introduction to Text Generation（Command + chatbot 工作机制 + Chat endpoint）
• Constructing Prompts
• Validating Outputs

B9. Anthropic Academy — 全部课程清单#

• URL：https://anthropic.skilljar.com/

完整课程列表：

1. 1. Claude 101
2. 2. Claude Code 101
3. 3. Introduction to Claude Cowork
4. 4. Claude Code in Action
5. 5. AI Fluency: Framework & Foundations
6. 6. Building with the Claude API
7. 7. Introduction to Model Context Protocol
8. 8. AI Fluency for Educators
9. 9. AI Fluency for Students
10. 10. Model Context Protocol: Advanced Topics
11. 11. Claude with Amazon Bedrock
12. 12. Claude with Google Cloud's Vertex AI
13. 13. Teaching AI Fluency
14. 14. AI Fluency for Nonprofits
15. 15. Introduction to Agent Skills
16. 16. Introduction to Subagents
17. 17. AI Capabilities and Limitations
18. 18. AI Fluency for Small Businesses

B10. Anthropic AI Fluency: Framework & Foundations（与 Joseph Feller / UCC 和 Rick Dakan / Ringling 合作）#

• URL：https://anthropic.skilljar.com/ai-fluency-framework-foundations
• AI Fluency Framework 站点：https://aifluencyframework.org/
• 核心 4D 框架：Delegation、Description、Discernment、Diligence

Module 清单：

1. 1. Introduction to AI Fluency
2. 2. The AI Fluency Framework — Why do we need AI Fluency? + The 4D Framework
3. 3. Deep Dive 1: What is Generative AI? (Part 1) — fundamentals、capabilities & limitations
4. 4. Delegation — A closer look + Project planning and Delegation
5. 5. Description — A closer look + Deep Dive 2: Effective prompting techniques
6. 6. Discernment — A closer look + Description-Discernment loop
7. 7. Diligence — A closer look
8. 8. Conclusion & Certificate — 总结 + 证书 + 额外活动

衍生课程：

• AI Fluency for Students：5 lessons，约 3 小时（含 30 分钟视频）
• AI Fluency for Educators：4 lessons，约 3 小时（含 35 分钟视频）

所有材料以 CC BY-NC-SA 许可发布，由爱尔兰 Higher Education Authority 通过 National Forum for Teaching and Learning 资助。

B11. OpenAI Academy#

• URL：https://academy.openai.com/

Collections：

• ChatGPT Fundamentals（resource）
• Prompting（resource）
• Codex 系列：Codex for Beginners（1h video, Aaron Wilkowitz）、Codex on Campus、Codex Fundamentals、Introduction to Codex、Codex for Admins and IT

Sector Collections：

• K-12 Education
• Higher Education（11 items）
• ChatGPT at Work
• Professors Teaching with OpenAI（83 items）
• Small Business（10+ items）
• Nonprofits（5 items）
• Government（12 items）
• Science（7 items）
• Healthcare

Special Topics：

• Prompt Packs、Custom GPTs、Workspace Agents、Advanced Features、Foundational AI Literacy

Stories & Case Studies：41 items

2025-2026 即将推出 AI fluency certifications（prompt engineering → AI-enabled work）。

B12. Google Machine Learning Crash Course#

• URL：https://developers.google.com/machine-learning/crash-course

模块清单：

ML Models

1. 1. Linear Regression — 线性模型、loss、SGD、超参
2. 2. Logistic Regression
3. 3. Classification — 二分类、thresholding、confusion matrix、accuracy/precision/recall/AUC

Data

1. 1. Working with Numerical Data
2. 2. Working with Categorical Data — one-hot、feature hashing、mean encoding、feature crosses
3. 3. Datasets, Generalization, and Overfitting

Advanced ML Models

1. 1. Neural Networks — perceptron、hidden layers、activation
2. 2. Embeddings
3. 3. Intro to Large Language Models — tokens、Transformer、训练

Real-World ML

1. 1. Production ML Systems
2. 2. AutoML
3. 3. ML Fairness

B13. Google Generative AI Learning Path#

• URL：https://www.skills.google/paths/118
• 5 activities

整体覆盖：generative AI 基础、LLMs、responsible AI。具体课程清单需登录 Google Skills 查看（搜索结果不公开详尽 list）。

B14. Microsoft AI for Beginners#

• URL：https://microsoft.github.io/AI-For-Beginners/ + https://github.com/microsoft/AI-For-Beginners
• 总长度：12 周，24 lessons

Section 清单：

I. Introduction to AI

• Lesson 1: Introduction and History of AI

II. Symbolic AI

• Lesson 2: Knowledge Representation and Expert Systems

III. Introduction to Neural Networks

• Lesson 3: Perceptron
• Lesson 4: Multi-Layered Perceptron and Creating our own Framework
• Lesson 5: Intro to Frameworks (PyTorch/TensorFlow) and Overfitting

IV. Computer Vision

• Lesson 6: Intro to Computer Vision. OpenCV
• Lesson 7: Convolutional Neural Networks & CNN Architectures
• Lesson 8: Pre-trained Networks and Transfer Learning
• Lesson 9: Autoencoders and VAEs
• Lesson 10: Generative Adversarial Networks & Artistic Style Transfer
• Lesson 11: Object Detection
• Lesson 12: Semantic Segmentation. U-Net

V. Natural Language Processing

• Lesson 13: Text Representation. Bow/TF-IDF
• Lesson 14: Semantic word embeddings. Word2Vec and GloVe
• Lesson 15: Language Modeling. Training your own embeddings
• Lesson 16: Recurrent Neural Networks
• Lesson 17: Generative Recurrent Networks
• Lesson 18: Transformers. BERT
• Lesson 19: Named Entity Recognition
• Lesson 20: Large Language Models, Prompt Programming and Few-Shot Tasks

VI. Other AI Techniques

• Lesson 21: Genetic Algorithms
• Lesson 22: Deep Reinforcement Learning
• Lesson 23: Multi-Agent Systems

VII. AI Ethics

• Lesson 24: AI Ethics and Responsible AI

B15. Microsoft Generative AI for Beginners#

• URL：https://microsoft.github.io/generative-ai-for-beginners/
• 总长度：22 lessons

Lesson 清单：

1. 0. Course Setup
2. 1. Introduction to Generative AI and LLMs
3. 2. Exploring and Comparing Different LLMs
4. 3. Using Generative AI Responsibly
5. 4. Understanding Prompt Engineering Fundamentals
6. 5. Creating Advanced Prompts
7. 6. Building Text Generation Applications
8. 7. Building Chat Applications
9. 8. Building Search Apps with Vector Databases
10. 9. Building Image Generation Applications
11. 10. Building Low Code AI Applications
12. 11. Integrating External Applications with Function Calling
13. 12. Designing UX for AI Applications
14. 13. Securing Your Generative AI Applications
15. 14. The Generative AI Application Lifecycle（LLMOps）
16. 15. Retrieval Augmented Generation and Vector Databases
17. 16. Open Source Models and Hugging Face
18. 17. AI Agents
19. 18. Fine-Tuning LLMs
20. 19. Building with SLMs
21. 20. Building with Mistral Models
22. 21. Building with Meta Models

B16. Microsoft Machine Learning for Beginners#

• URL：https://github.com/microsoft/ML-For-Beginners
• 总长度：25 lessons + 2 postscript

Lesson 清单：

1. Introduction（4 lessons）

• 01: Introduction to machine learning
• 02: The History of machine learning
• 03: Fairness and machine learning
• 04: Techniques for machine learning

2. Regression（4 lessons）

• 05: Introduction to regression
• 06: North American pumpkin prices 🎃
• 07: North American pumpkin prices 🎃 (continued)
• 08: North American pumpkin prices 🎃 (logistic regression)

3. Web App

• 09: A Web App 🔌

4. Classification（4 lessons）

• 10: Introduction to classification
• 11: Delicious Asian and Indian cuisines 🍜
• 12: Delicious Asian and Indian cuisines 🍜 (continued)
• 13: Delicious Asian and Indian cuisines 🍜 (web app)

5. Clustering（2 lessons）

• 14: Introduction to clustering
• 15: Exploring Nigerian Musical Tastes 🎧

6. NLP（5 lessons）

• 16: Introduction to natural language processing
• 17: Common NLP Tasks
• 18: Translation and sentiment analysis
• 19: Romantic hotels of Europe
• 20: Romantic hotels of Europe (continued)

7. Time Series（3 lessons）

• 21: Introduction to time series forecasting
• 22: World Power Usage - ARIMA
• 23: World Power Usage - SVR

8. Reinforcement Learning（2 lessons）

• 24: Introduction to reinforcement learning
• 25: Help Peter avoid the wolf!

9. Real-World（2 postscript）

• Real-World ML scenarios
• Model Debugging using RAI dashboard

B17. Meta AI / Llama Cookbook（公开教程）#

• URL：https://github.com/meta-llama/llama-recipes

不是正式课程，但是结构化教程集合：

Top-Level Categories：

1. 1. Getting Started — Inference、Fine-tuning、RAG
2. 2. End-to-End Use Cases — WhatsApp bot with Llama 4、Research paper analysis、Book character mind mapping
3. 3. Third-Party Integrations
4. 4. Latest Llama 4 Recipes — Llama API、Long context（5M tokens with Scout）、Maverick advanced analysis

主要为 Jupyter notebooks（93.7% 内容）。

与 DeepLearning.AI 合作课程：

• Building with Llama 4（Amit Sangani / Meta，DeepLearning.AI 平台）

B18. NVIDIA Deep Learning Institute（DLI）#

• URL：https://www.nvidia.com/en-us/training/self-paced-courses/
• 课程目录 PDF：https://cdn.dli.learn.nvidia.com/web-asset/nvidia-learning-training%20course-catalog.pdf

Tracks：

1. 1. Accelerated Computing
2. 2. Data Science
3. 3. Deep Learning
4. 4. Generative AI / LLM
5. 5. Infrastructure
6. 6. Simulation and Physical AI

课程形式：

• 2 小时短课（聚焦特定技术）
• 8 小时项目型课程（含 certificate）

已知具体课程：

• Fundamentals of Deep Learning（8h）
• Building Transformer-Based Natural Language Processing
• Generative AI with Diffusion Models
• Building RAG Agents with LLMs
• Efficient LLM Customization
• CUDA / Accelerated Computing 系列

（完整 catalog 需查 PDF 或 NVIDIA 在线目录）

C 类 — 大学开源 / 顶级开课#

C1. Stanford CS224N — Natural Language Processing with Deep Learning#

• URL：https://web.stanford.edu/class/cs224n/
• 总长度：10 周 / 约 20 lectures

Winter 2026 Schedule：

• Week 1: History of NLP / Word Vectors / Python Review
• Week 2: Backpropagation and Neural Network Basics / Language Models and RNNs / PyTorch Tutorial
• Week 3: Transformers / Final Projects: Custom and Default
• Week 4: Pretraining (Scaling, Systems, Data) / Post-training (RLHF, SFT, DPO)
• Week 5: Efficient Adaptation (Prompting + PEFT) / Agents, Tool Use, and RAG / Hugging Face Tutorial
• Week 6: Benchmarking and Evaluation / Reasoning 1
• Week 7: Reasoning 2 / Guest: Tokenization and Multilinguality (Julie Kallini)
• Week 8: Guest: Interpretability (Been Kim) / Social and Broader Impacts of NLP
• Week 9: Guest: Multimodality (Luke Zettlemoyer) / Guest: Tinker and LoRA Without Regret (John Schulman)
• Week 10: Open Questions in NLP 2026

Assignments：

• A1 (6%): Word Vectors
• A2 (14%): Neural networks, derivatives, dependency parsing
• A3 (14%): Self-attention and Transformers
• A4 (14%): LLM benchmarking and evaluation
• Final Project（Custom 或 Default）

C2. Stanford CS224U — Natural Language Understanding#

• URL：https://web.stanford.edu/class/cs224u/
• 总长度：~10 周

Spring 2023 Schedule：

• Apr 3: Course Introduction
• Apr 5-12: Contextual Word Representations & Language Models（Transformers、GPT、BERT、RoBERTa、ELECTRA、T5、BART、DistilBERT）
• Apr 17-24: Retrieval-Augmented In-Context Learning（neural IR、LLM prompting）
• Apr 26-May 3: Advanced Behavioral Evaluation
• May 8-15: Analysis Methods & NLP Evaluation（interpretability）
• May 31-Jun 7: Research Presentation & Project Completion

Assignments：

• A1: Domain adaptation sentiment classification
• A2: Open-domain question answering
• A3: ReCOGS compositional generalization
• 4 Quizzes（0-4）
• Literature Review
• Experimental Protocol
• Final Paper

C3. Stanford CS224W — Machine Learning with Graphs#

• URL：https://web.stanford.edu/class/cs224w/
• 总长度：19 lectures（Fall 2025）

Lecture 清单：

1. 1. Introduction
2. 2. Node Embeddings
3. 3. Graph Neural Networks
4. 4. A General Perspective on GNNs
5. 5. GNN Augmentation and Training
6. 6. Theory of GNNs
7. 7. Designing Powerful Graph Encoders
8. 8. Graph Transformers
9. 9. Heterogeneous Graphs
10. 10. Knowledge Graphs
11. 11. GNNs for Recommender Systems
12. 12. Relational Deep Learning
13. 13. Advanced Architectures in RDL
14. 14. Advanced Topics in GNNs
15. 15. Towards Foundation Models for Knowledge Graphs
16. 16. LLM + GNN
17. 17. Agents + Graphs
18. 18. Deep Generative Models for Graphs
19. 19. Conclusion

C4. Stanford CS234 — Reinforcement Learning#

• URL：https://web.stanford.edu/class/cs234/
• 总长度：11 周

Winter 2026 Schedule：

• Week 1: Introduction to RL / Tabular MDP Planning（A1 released）
• Week 2: Policy Evaluation / Q-learning and function approximation
• Week 3: Policy Search
• Week 4: Policy Search / Offline RL, Imitation Learning
• Week 5: Offline RL, RLHF / Midterm Exam（A3 released, Project Proposal due）
• Week 6: Offline RL / Bandits / Strategic data gathering / Exploration
• Week 7: Exploration
• Week 8: Exploration / RL and MCTS（Project Milestone due）
• Week 9: Guest Lecture / In-Class Quiz
• Week 10: Poster Session
• Week 11: Final Project Writeup

Capstone：3 个 Assignments + Final Project（proposal、milestone、poster、writeup）。

C5. Stanford CS330 — Deep Multi-Task and Meta Learning#

• URL：https://cs330.stanford.edu/
• 总长度：11 周（Fall 2023）

Week 清单：

• Week 1: Course introduction & multi-task learning fundamentals + PyTorch tutorial
• Week 2: Multi-task learning methods / Transfer learning & fine-tuning
• Week 3: Black-box meta-learning & in-context learning / Optimization-based meta-learning + MAML review
• Week 4: Few-shot learning via metric learning / Unsupervised pre-training contrastive
• Week 5: Unsupervised pre-training generative / Advanced meta-learning task construction
• Week 6: Variational inference / Bayesian meta-learning
• Week 7: Large-scale meta-optimization / Learned optimizers (guest)
• Week 8: Lifelong learning
• Week 10: LM advances (guest) / Domain adaptation and generalization
• Week 11: Frontiers / Project poster session

Grading：4 HW（HW0 5%, HW1-3 15% each）+ Project 50%。

C6. Stanford CS25 — Transformers United（seminar 系列）#

• URL：https://web.stanford.edu/class/cs25/

所有 Season + 嘉宾：

V1（Fall 2021）：

1. 1. Introduction to Transformers（instructors）
2. 2. Mark Chen (OpenAI) — "Transformers in Language: GPT-3, Codex"
3. 3. Lucas Beyer (Google Brain) — "Applications in Vision"
4. 4. Aditya Grover (FAIR) — "Transformers in RL & Universal Compute Engines"
5. 5. Barret Zoph + Irwan Bello + Liam Fedus (Google Brain) — "Scaling transformers"
6. 6. Andrew Jaegle (DeepMind) — "Perceiver"
7. 7. Aidan Gomez (Oxford) — "Self Attention & Non-Parametric Transformers"
8. 8. Geoffrey Hinton (Toronto) — "GLOM"
9. 9. Chris Olah (Anthropic) — "Interpretability with transformers"
10. 10. Prateek Verma (Stanford) — "Transformers for Audio, Speech and Music"

V2（Winter 2023）：

• Jan 10: Andrej Karpathy — Introduction to Transformers
• Jan 17: Jan Leike (OpenAI) — Language and Human Alignment
• Jan 24: Jason Wei (Google Brain) — Emergent Abilities and Scaling in LLMs
• Jan 31: Noam Brown (FAIR) — Strategic Games
• Feb 7: Ted Xiao (Google Brain) — Robotics and Imitation Learning
• Feb 14: Yejin Choi (UW / AI2) — Common Sense Reasoning
• Feb 21: Vivek Natarajan (Google Health AI) — Biomedical Transformers
• Feb 28: Stephanie Chan + Antonia Creswell (DeepMind) — In-Context Learning & Faithful Reasoning
• Mar 7: Trenton Bricken + Will Dorrell — Neuroscience-Inspired AI

V3（Fall 2023）：

• Oct 3: Sharan Narang (Meta AI) — Llama 2
• Oct 10: Fei Xia (Google DeepMind) — Low-level Embodied Intelligence with Foundation Models
• Oct 24: Jim Fan (NVIDIA) — Generalist Agents in Open-Ended Worlds
• Oct 31: Nazneen Rajani (HF) — Recipe for Training Helpful Chatbots
• Nov 7: Ashish Vaswani — How I Learned to Stop Worrying and Love the Transformer
• Nov 14: Angela Fan (Meta AI) — No Language Left Behind
• Nov 28: Instructors — Going Beyond LLMs: Agents, Emergent Abilities, BabyLM
• Dec 5: Douwe Kiela (Contextual AI) — Retrieval Augmented Language Models

V4（Spring 2024）：

1. 1. Instructors — Overview of Transformers
2. 2. Jason Wei (OpenAI) — Intuitions on Language Models
3. 3. Hyung Won Chung (OpenAI) — Shaping the Future of AI from the History of Transformer
4. 4. Nathan Lambert (AI2) — Aligning Open Language Models
5. 5. Albert Jiang (Mistral / Cambridge) — Demystifying Mixtral of Experts
6. 6. Jake Williams (Drexel) — Transformers that Transform Well Enough to Support Near-Shallow Architectures
7. 7. Ming Ding (Zhipu AI) — From LLMs to Large Multimodal Models
8. 8. Edward Hu (ex-OpenAI) — New training objective for LLMs
9. 9. Loubna Ben Allal (HF) — Behind the Scenes of LLM Pre-training: StarCoder

V6（Spring 2026）：

1. 1. Apr 9: Hazel Nam & Lucas Maes (Brown) — From Representation Learning to World Modeling through JEPAs
2. 2. Apr 16: Albert Gu (CMU, Cartesia AI) — Tradeoffs of State Space Models and Transformers
3. 3. Apr 23: Nouamane Tazi (HF) — Ultra-Scale Talk: Scaling Training to Thousands of GPUs
4. 4. Apr 30: Shrimai Prabhumoye (Mistral AI) — From Next-Token Prediction to Next-Generation Intelligence
5. 5. May 7: Andrew Lampinen (Anthropic) — Distinct Modes of Generalization from Parameters and Context
6. 6. May 14: Vivek Natarajan (DeepMind) — Advancing Science and Medicine with Collaborative AI Agents
7. 7. May 21: Victoria Lin (Thinking Machines) — From LMs to Native Multimodal Intelligence
8. 8. May 28: Charles Frye (Modal) — Serving Transformers: Lessons from the Trenches of Production Inference

C7. Stanford CS329S — Machine Learning Systems Design#

• URL：https://stanford-cs329s.github.io/
• 总长度：10 周（Winter 2022）

Schedule：

• Week 1: Understanding ML production / ML and Data Systems Fundamentals
• Week 2: Training Data / Feature engineering
• Week 3: Model selection, development, and training
• Week 4: Offline evaluation / Model evaluation tutorial (Goku Mohandas)
• Week 5: Deployment / Deployment tutorials (MLOps tools & Ray Serve)
• Week 6: Diagnosis of ML system failures, distribution shifts & monitoring / Monitoring & Continual Learning（streaming guest）
• Week 7: Model Deployment case study & experiment tracking / Monitoring tutorials (WhyLogs, Evidently)
• Week 8: Time series forecasting & fairness/security/governance (guests)
• Week 9: ML Infrastructure and Platform / Final project workshop
• Week 10: Integrating ML into business (guests) / Final project demo day

C8. Stanford CS336 — Language Modeling from Scratch（2025 新）#

• URL：http://cs336.stanford.edu/spring2025/（讲师：Percy Liang + Tatsu Hashimoto）
• 总长度：19 lectures

Lecture 清单：

1. 1. Overview, tokenization（Percy）
2. 2. PyTorch, resource accounting（Percy）
3. 3. Architectures, hyperparameters（Tatsu）
4. 4. Mixture of experts（Tatsu）
5. 5. GPUs（Tatsu）
6. 6. Kernels, Triton（Tatsu）
7. 7. Parallelism（Tatsu）
8. 8. Parallelism（Percy）
9. 9. Scaling laws（Tatsu）
10. 10. Inference（Percy）
11. 11. Scaling laws（Tatsu）
12. 12. Evaluation（Percy）
13. 13. Data（Percy）
14. 14. Data（Percy）
15. 15. Alignment - SFT/RLHF（Tatsu）
16. 16. Alignment - RL（Tatsu）
17. 17. Alignment - RL（Percy）
18. 18. Guest: Junyang Lin
19. 19. Guest: Mike Lewis

Assignments：

• A1: Basics
• A2: Systems
• A3: Scaling
• A4: Data
• A5: Alignment and Reasoning RL

C9. MIT 6.S191 — Introduction to Deep Learning#

• URL：http://introtodeeplearning.com/
• 总长度：9 lectures + 3 software labs + Final Project（2026 版）

Lecture 清单：

1. 1. Intro to Deep Learning
2. 2. Deep Sequence Modeling
3. 3. Deep Computer Vision
4. 4. Deep Generative Modeling
5. 5. Deep Reinforcement Learning
6. 6. New Frontiers
7. 7. The Three Laws of AI
8. 8. AI for Science
9. 9. Secrets to Massively Parallel Training

Software Labs：

1. 1. Deep Learning in Python; Music Generation
2. 2. Facial Detection Systems
3. 3. Fine-Tune an LLM, You Must!

Capstone：Final Project + Project Pitch + 颁奖。

C10. MIT 6.5940 — TinyML and Efficient Deep Learning Computing#

• URL：https://hanlab.mit.edu/courses/2024-fall-65940
• 总长度：26 lectures + 6 labs + Final Project

Lecture 清单（Fall 2024）：

1. 1. Introduction
2. 2. Basics of Deep Learning
3. 3. Pruning and Sparsity (Part I)
4. 4. Pruning and Sparsity (Part II)
5. 5. Quantization (Part I)
6. 6. Quantization (Part II)
7. 7. Neural Architecture Search (Part I)
8. 8. Neural Architecture Search (Part II)
9. 9. Knowledge Distillation
10. 10. MCUNet: TinyML on Microcontrollers
11. 11. TinyEngine and Parallel Processing
12. 12. Transformer and LLM
13. 13. Efficient LLM Deployment
14. 14. LLM Post Training
15. 15. Long Context LLM
16. 16. Vision Transformer
17. 17. GAN, Video, and Point Cloud
18. 18. Diffusion Model
19. 19. Distributed Training (Part I)
20. 20. Distributed Training (Part II)
21. 21. On-Device Training and Transfer Learning
22. 22. Course Summary + Quantum ML I
23. 23. Quantum ML II

24-26. Final Project Presentations

Labs：

• Lab 0: Basics
• Lab 1: Pruning
• Lab 2: Quantization
• Lab 3: Neural Architecture Search
• Lab 4: LLM Compression
• Lab 5: LLM Deployment

Capstone：Final Project（4-5 人 / proposal + final report）。

C11. CMU 10-714 — Deep Learning Systems#

• URL：https://dlsyscourse.org/
• 总长度：每周 2 lectures（TR 11:00-12:20）

5 Homework Assignments：

• HW0: Background review and softmax regression
• HW1: Automatic differentiation framework
• HW2: Neural network modules and optimizers
• HW3: NDArray library with CPU/CUDA backends
• HW4: Convolutions, RNNs, and advanced architectures

Capstone：Final Project（在 Needle 框架中实现新 feature + 模型）。

C12. CMU 11-711 — Advanced NLP（Graham Neubig，Spring 2025 / Fall 2024）#

• URL：https://cmu-l3.github.io/anlp-spring2025/ + https://phontron.com/class/anlp-fall2024/
• 总长度：27 lectures

Spring 2025 Lecture 清单：

1. 1. Introduction and Basics
2. 2. Word Representation and Text Classification
3. 3. Language Modeling Fundamentals
4. 4. Recurrent Neural Networks
5. 5. Attention and Transformers
6. 6. Pretraining
7. 7. Decoding and Generation（Guest: Amanda Bertsch）
8. 8. Prompting
9. 9. Fine-Tuning
10. 10. Retrieval and RAG（Guest: Akari Asai）
11. 11. Reinforcement Learning
12. 12. Evaluating Language Generators（Guest: Seungone Kim）
13. 13. Experimental Design
14. 14. Agents
15. 15. Quantization（Guest: Tim Dettmers）
16. 16. Advanced Pretraining: Parallelism and Scaling
17. 17. Project Discussion
18. 18. Long Sequence Models
19. 19. Advanced Inference Strategies
20. 20. Efficient Inference
21. 21. Advanced Post-Training
22. 22. Multimodal Models I
23. 23. AI for Mathematics
24. 24. Multimodal Models II

25-26. Poster Presentations

Assignments：

• A1: 个人（01/28-02/14）
• A2: 团队（02/13-03/14）
• A3: 团队 — Literature survey & baseline reproduction（02/20-04/07）
• A4: 团队 — Final project（02/20-04/27）

Grading：Quizzes 20%（drop 3）/ A1 15% / A2 15% / A3 20% / A4 (Project) 30%。

C13. CMU 17-645 — Machine Learning in Production / AI Engineering（Christian Kästner）#

• URL：https://mlip-cmu.github.io/s2025/ + https://github.com/mlip-cmu
• 配套教材：Machine Learning in Production: From Models to Products

Spring 2025 Schedule（14 周）：

• Week 1: Introduction and Motivation / From Models to AI-Enabled Systems / Lab 1: Flask APIs
• Week 2: Gathering Requirements / Lab 2: Kafka stream processing（I1 ML Product due）
• Week 3: Planning for Mistakes / Model Quality / Lab 3: git collaboration（I2 Requirements due）
• Week 4: Fostering Interdisciplinary Teams / Behavioral Model Testing / Lab 4: Model testing
• Week 5: Toward Architecture and Design / Deploying a Model / Lab 5: Docker containers
• Week 6: Testing and Experimenting in Production / Data Quality / Lab 6: Jenkins CI（M1 Modeling and First Deployment due）
• Week 7: Automating and Testing ML Pipelines / Midterm 1
• Week 8: Scaling the System / Planning for Operations / Lab 7: Prometheus + Grafana
• Week 9: Versioning / Process & Technical Debt / Lab 8: MLflow（M2 Infrastructure Quality due; I3 MLOps Tools due）
• Week 10: Intro to Ethics + Fairness / Measuring Fairness / Lab 9: Kubernetes
• Week 11: Building Fairer Systems / Guest Lecture
• Week 12: Explainability / Transparency & Accountability / Lab 10: Model Explainability（M3 Monitoring and CD due）
• Week 13: Safety / Security and Privacy / Lab 11: LLM Jailbreaking（I4 Explainability due）
• Week 14: Discussion / Summary / Review / Midterm 2（M4 Fairness, Security and Feedback Loops due）
• Final: Final report and video（May 5）

Grading：Individual 35% / Group project 30% / Midterms 15% / Participation 5% / Labs 10% / Reading quizzes 5%。

C14. Berkeley CS285 — Deep Reinforcement Learning（Sergey Levine）#

• URL：https://rail.eecs.berkeley.edu/deeprlcourse/
• 总长度：25 lectures

Lecture 模块：

• Weeks 1-2: Introduction / Behavioral Cloning / RL Basics
• Weeks 3-4: Policy Gradients / Actor Critic / Value-Based RL / Q-learning in Practice
• Week 5: Advanced Policy Gradients（2 部分）
• Weeks 6-7: Variational Inference / VI in RL / Control as Inference / LLM RL
• Week 8: Model-Based RL（2 部分）
• Week 9: Offline RL（2 部分）
• Week 10: Exploration / RL Theory
• Weeks 11-12: Midterm review / Advanced Exploration / Multi-task RL
• Weeks 13-14: Challenges and Open Problems / Final guest

5 Homework Assignments：

1. 1. Imitation Learning
2. 2. Policy Gradients
3. 3. Q-Learning and Actor Critic
4. 4. LLM RL
5. 5. Offline RL

Final Project Options：Offline-to-Online RL / LLM RL。

C15. Princeton COS 597G — Understanding Large Language Models（Danqi Chen，Fall 2022）#

• URL：https://www.cs.princeton.edu/courses/archive/fall22/cos597G/
• 总长度：~23 lectures

Lecture 清单：

Grading：Participation 25%（pre-lecture readings + 3 questions）/ Presentations 30%（60-min student-led）/ Lecture feedback 5% / Final project 40%。

Project Options：Fine-tune medium LMs（BERT/RoBERTa/T5/GPT-2 via HF）or 评估大模型（GPT-3, Codex）。

C16. Princeton COS 597R — Deep Dive into Large Language Models（Danqi Chen + Sanjeev Arora，Fall 2024）#

• URL：https://princeton-cos597r.github.io/
• 总长度：~23 lectures

Lecture 清单：

Grading：Participation 30% / Debate 15% / Lecture scribing 10% / Final project 35% / Project presentations 10%。

Debate Panels：Weeks 4-9 共 12 debate panels（每 5 个学生为一组，每 paper 有 presenter / critic / proponent）。

D 类 — 实操 / 工程社区课程#

D1. Full Stack Deep Learning（Sergey Karayev / Pieter Abbeel / Josh Tobin）#

• URL：https://fullstackdeeplearning.com/course/

FSDL 2022 (Spring) — 9 Lectures + 6 Labs：

Lectures：

1. 1. Course Vision and When to Use ML
2. 2. Development Infrastructure & Tooling
3. 3. Troubleshooting & Testing
4. 4. Data Management
5. 5. Deployment
6. 6. Continual Learning
7. 7. Foundation Models
8. 8. ML Teams and Project Management
9. 9. Ethics

Labs：

• Lab 0: Overview
• Lab 4: Experiment Management
• Lab 5: Troubleshooting & Testing
• Lab 6: Data Annotation
• Lab 7: Web Deployment
• Lab 8: Model Monitoring

外加 Project Showcase。

早期版本：FSDL 2021（Berkeley + online）/ 2020（UW）/ 2019 / 2018。

D2. Full Stack LLM Bootcamp（Spring 2023）#

• URL：https://fullstackdeeplearning.com/llm-bootcamp/spring-2023/

8 个 Lectures：

1. 1. Learn to Spell: Prompt Engineering
2. 2. LLMOps
3. 3. UX for Language User Interfaces
4. 4. Augmented Language Models
5. 5. Launch an LLM App in One Hour
6. 6. LLM Foundations
7. 7. Project Walkthrough: askFSDL
8. 8. What's Next?

3 个 Invited Talks：

• Reza Shabani — Training custom LLMs
• Harrison Chase — Agents
• Peter Welinder (OpenAI VP) — Fireside Chat

Capstone：项目 askFSDL（Q&A 系统，Python/Modal/Gantry）。

D3. MLOps Zoomcamp（DataTalks.Club）#

• URL：https://github.com/DataTalksClub/mlops-zoomcamp

7 个 Module：

1. 1. Introduction — MLOps 定义、maturity model、NY Taxi 数据集、环境搭建 + Homework
2. 2. Experiment Tracking & Model Management — MLflow 基础、模型保存/加载/注册 + Homework
3. 3. Orchestration & ML Pipelines + Homework
4. 4. Model Deployment — online (web/streaming) vs offline (batch)、Flask、AWS Kinesis + Lambda、batch scoring + Homework
5. 5. Model Monitoring — Prometheus + Evidently + Grafana、Prefect + MongoDB + Homework
6. 6. Best Practices — 单元/集成测试、linting/pre-commit、GitHub Actions CI/CD、Terraform IaC + Homework
7. 7. Final Project — 端到端整合所有概念

D4. LLM Zoomcamp（DataTalks.Club）#

• URL：https://github.com/DataTalksClub/llm-zoomcamp

7 个 Module + Workshop：

1. 1. Introduction to LLMs & RAG — 基础 RAG pipeline + text search
2. 2. Vector Search — semantic embeddings 索引/检索
3. 3. Agents — 自治 tool use 和 function calling
4. 4. (Workshop) Data Ingestion — dlt 从外部源接入 RAG
5. 5. Evaluation — 离线 + 在线 retrieval / answer 质量
6. 6. Monitoring — 用户反馈 + live dashboards
7. 7. Best Practices — LangChain、hybrid search（vector+keyword）、reranking
8. 8. End-to-End Project — 健身助手 with LLMs

Capstone：完整 RAG 应用（知识库 + retrieval + LLM 集成 + evaluation + UI + 监控反馈）。Certification 需提交项目并 peer review 3 个其他学生项目。

D5. AI Engineer World's Fair 工作坊#

• URL：https://www.ai.engineer/

每年 6 月在旧金山，workshop 由各家公司主讲（Anthropic、OpenAI、LangChain、LlamaIndex、Modal、Weaviate、Pinecone、Together AI 等）。典型 workshop 包括：

• LLMs for the Working Programmer
• 2-hour workshop on Gemini 2.5
• Hamel Husain — How to Construct Domain Specific LLM Evaluation Systems

完整 schedule 见每年事件页（2024 / 2025）。

D6. Modular Mojo AI Engineer 培训#

• 官方文档：https://mojolang.org/docs/manual/
• MAX 平台：https://www.modular.com/max/mojo

不是正式课程，而是 docs + 工具链：

• Mojo 语法、static class system（structs、fn）
• Modular Engine（LLVM + MLIR）
• Modular Playground（Jupyter-style IDE）

第三方 Udemy 课程：

• Mojo Programming Training: Powering High-Performance AI
• Mojo Programming: Basic to Expert with AI Use cases

D7. Replit Learn#

• URL：https://learn.replit.com/

目前提供：

1. 1. AI foundations（30 min, 7 lessons）— 用 AI 构建 app
2. 2. Enterprise foundations（30 min, 6 lessons）
3. 3. Intro to Replit（45 min, 5 lessons）— coming soon
4. 4. Advanced vibes（30 min, 5 lessons）— coming soon

Replit 100 Days of Python（讲师：David Morgan）：

• 100 天 / 每天 15 分钟
• 覆盖：变量、math 函数、for/while/while-true 循环、libraries、subroutines、lists/2D lists、字典、文件读写、eval()、try/except、csv 操作、os library、recursion、debugging、数据库函数、HTML/CSS Crash Course、Flask
• 项目：tip calculator、grade calculator、multiplayer games、login systems、web scraping

D8. The AI Engineering Bootcamp（Greg Loughnane + Chris Alexiuk）#

• URL：https://maven.com/aimakerspace/ai-eng-bootcamp
• 总长度：10 周

涵盖 AI 原则、Python、NLP、LLMs、Transformers、LangChain orchestration。Build, ship, and share。

附录 — 总览整理#

所有课程汇总：~61 门具体课程 + N 个 specialization / track。

重复出现的结构模式（仅观察，不做评价）：

• 时长：单课程多在 3-12 周
• Capstone：约 70% 课程有 final project 或 end-to-end capstone
• Lecture 数：单课程多在 9-25 个 lecture
• 作业密度：高质量课程多有 3-5 个 assignment 横跨学期

主要信息源

• karpathy.ai / cs231n.stanford.edu
• course.fast.ai / course19.fast.ai
• coursera.org/specializations/deep-learning
• coursera.org/specializations/machine-learning-introduction
• coursera.org/learn/generative-ai-with-llms
• deeplearning.ai/courses
• github.com/rasbt/LLMs-from-scratch
• github.com/chiphuyen/aie-book
• github.com/mlabonne/llm-course
• eugeneyan.com
• hamel.dev/talks.html
• parlance-labs.com/education/
• maven.com/parlance-labs/evals
• huggingface.co/learn （NLP/LLM/Agents/RL/Audio/Diffusion/CV）
• cohere.com/llmu
• anthropic.skilljar.com / aifluencyframework.org
• academy.openai.com
• developers.google.com/machine-learning/crash-course
• skills.google
• github.com/microsoft/AI-For-Beginners
• github.com/microsoft/generative-ai-for-beginners
• github.com/microsoft/ML-For-Beginners
• github.com/meta-llama/llama-recipes
• nvidia.com/en-us/training
• web.stanford.edu/class/cs224n + cs224u + cs224w + cs234 + cs330 + cs25 + cs329s
• cs336.stanford.edu/spring2025
• introtodeeplearning.com
• hanlab.mit.edu/courses/2024-fall-65940
• dlsyscourse.org
• cmu-l3.github.io/anlp-spring2025
• phontron.com/class/anlp-fall2024
• mlip-cmu.github.io
• rail.eecs.berkeley.edu/deeprlcourse
• cs.princeton.edu/courses/archive/fall22/cos597G
• princeton-cos597r.github.io
• fullstackdeeplearning.com
• github.com/DataTalksClub/mlops-zoomcamp + llm-zoomcamp
• ai.engineer
• modular.com/max/mojo
• learn.replit.com