datawhalechina/leedl-tutorial

《李宏毅深度学习教程》（李宏毅老师推荐👍，苹果书🍎），PDF下载地址：https://github.com/datawhalechina/leedl-tutorial/releases

What it solves

Learning deep learning from Chinese materials often means juggling scattered videos, notes, and code. This project turns a well-known lecture series into a structured, book-like tutorial that lowers the entry barrier. It selects the essential parts of the course, explains difficult ideas with extra attention, and derives every formula step by step so readers can follow without getting stuck on math notation.

How it works

The tutorial is built mainly from Professor Hung-yi Lee's "Machine Learning" (Spring 2021) open course, with some content from his 2017 version. The authors reorganized the material, added original explanations, and included additional deep-learning topics beyond the original lectures. Every formula is shown with a detailed derivation, and tricky concepts are highlighted and explained more thoroughly. The repository provides the full text in PDF form (via releases), a free online reading version (linked through the project pages), and a printed book published by Posts & Telecom Press. A companion code folder (Homework) offers practical exercises that match the chapters.

Who it's for

This is for Chinese-speaking learners who want to get into deep learning but find typical math-heavy textbooks intimidating. It suits self-study students who prefer clear, friendly explanations with anime-related examples, as well as anyone preparing to explore more specialized AI fields. The authors also recommend it for readers who like a physical book for reference while following online videos.

Highlights

  • Based on a highly popular Chinese-language deep learning course, endorsed by Professor Lee himself.
  • Covers a wide range of topics: deep learning basics, CNNs, self-attention, RNNs, Transformers, GANs, self-supervised learning (BERT, GPT-3), autoencoders, diffusion models, adversarial attacks, explainable AI, transfer learning, deep reinforcement learning, lifelong learning, network compression, meta-learning, and ChatGPT.
  • Every formula is derived in detail to reduce math barriers.
  • Available in multiple formats: free PDF, online reading, and a printed book.
  • Includes a Homework folder with companion code for hands-on practice.
  • Licensed under CC BY-NC-SA 4.0, so it is free to share and adapt for non-commercial use.

Related

  • Project
    tracel-ai/burn

    Burn is an open‑source Rust library that combines a tensor engine with a full deep‑learning framework. It lets you write training code in Rust and run the exact same code for inference on GPUs, CPUs, WebGPU, or even remote machines, thanks to a unified `Device` API, JIT‑compiled kernels, automatic fusion, and optional autodiff. The ecosystem includes CubeCL (multi‑backend GPU compiler), ONNX import, weight loaders, domain‑specific crates (vision, RL, datasets), and tooling such as a terminal training dashboard and benchmark suite.

  • Project
    modelscope/ms-swift

    A scalable infrastructure for fine-tuning and deploying 1,000+ text and multimodal large models, supporting the full pipeline from training to quantization and inference.

  • Project
    AI4Finance-Foundation/FinRL

    An open-source framework for financial reinforcement learning that enables users to automate quantitative trading through a train-test-trade pipeline.

  • Project
    rasbt/deeplearning-models

    A collection of deep learning architectures and model implementations in PyTorch and TensorFlow, providing runnable Jupyter Notebooks for a wide variety of neural network types.