Turning a viral Chinese ML lecture series into a readable textbook

What it does

This repository hosts the open-source draft of LeeDL Tutorial—the “Apple Book”—a Chinese deep-learning textbook that grew out of Professor Hung-yi Lee’s beloved NTU video lectures. A team of graduate students from SJTU, Oxford, and PKU transcribed the lectures, added exhaustive formula derivations, and stitched in extra topics to turn episodic video content into a linear, self-contained primer. The result covers everything from CNNs and Transformers to diffusion models, network compression, and ChatGPT, backed by a folder of Jupyter Notebook homework.

The interesting bit

The project’s core trick is freezing Lee’s famously casual, anime-referencing teaching style into rigorous text that you can read at your own pace. It also carries the professor’s public Twitter endorsement, which is about as close to canon as a community-built textbook can get.

Key highlights

• Personally endorsed by Professor Hung-yi Lee on Twitter.
• Published in print by Posts & Telecom Press; the raw PDF draft remains free on GitHub Releases.
• Surprisingly broad topical range, including diffusion models, adversarial attacks, meta-learning, and the pre-training mechanics behind ChatGPT.
• Bundles a Homework folder with Jupyter Notebook companions.
• Maintains a public errata page and is licensed under CC BY-NC-SA 4.0.

Caveats

• The material is entirely in Chinese, so it is inaccessible to non-Chinese readers.
• The PDF is explicitly labeled as a first draft; the physical book contains the polished, edited version.
• Several chapters—such as auto-encoders and deep reinforcement learning—are listed in the README without detailed sub-sections, so their depth is harder to gauge from the table of contents alone.

Verdict

A strong pick for Chinese-speaking learners who want a readable, mathematically explicit bridge into deep learning without sitting through hours of video. If your bookshelf is already full of English ML textbooks, you are unlikely to find new theory here.