datawhalechina/key-book

What it solves

Reading machine learning theory can be tough: the textbook “An Introduction to Theoretical Machine Learning” is dense and skips many steps in proofs, assuming strong math fluency. This project is a companion notebook that helps readers get through the fog by adding three kinds of support: plain-language explanations of concepts the book mentions but doesn't elaborate, walkthroughs of proof ideas with the missing derivation steps filled in, and concrete examples to make abstract theory easier to grasp.

How it works

The project is organized as a chapter-by-chapter companion to the original book. It follows the same structure, covering eight core topics: learnability, complexity, generalization bounds, stability, consistency, convergence rates, and regret bounds. For each chapter, the companion provides additional explanation, step-by-step proof supplements, and illustrative case studies. The content is available for free online reading, and a PDF version can be downloaded directly. It also includes a preface that connects computational learnability to learnability, and an appendix with mathematical basics like norms, convex sets, and optimization.

Who it’s for

This is for anyone currently studying the machine learning theory textbook and wanting a reference to aid understanding. It's also useful for researchers and students interested in a systematic introduction to machine learning theory, and for engineers who already have practical experience but want to build a stronger theoretical foundation.

Highlights

• Provides three specific types of additions: concept explanations, proof supplements, and real examples
• Covers all seven core theoretical concepts from the original book: learnability, complexity, generalization bounds, stability, consistency, convergence rates, and regret bounds
• Free to read online, with a downloadable PDF version
• Actively welcomes contributions through issues and pull requests
• Community support via WeChat and QQ groups for reader discussion
• Created by the Datawhale open-source community, known for its companion “Pumpkin Book” for Zhou Zhihua's “Machine Learning” textbook
• Licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International