▤Study Index ← Overview

Math · Index

The Math column of the University Plan · click any entry to open

Free legal textbooksfree from the authors

Released free by the authors or publisher; already downloaded locally.

▤
Boyd · Convex Optimization
Free from Stanford (Boyd) · core ML mathematics · 6.6 MB
↗
▤
MacKay · Information Theory, Inference, and Learning Algorithms
Free from the author's site · entropy / mutual information / KL divergence · computational theories of cognition · 11 MB
↗
▤
Jurafsky · Speech and Language Processing (SLP3)
Free draft from Stanford · the NLP bible · 25 MB
↗
▤
Dive into Deep Learning (D2L)
Free and open source · Chinese edition · by Mu Li's team · 31 MB
↗

Official free alternativesneeds internet

Copyrighted textbooks (Strang / Blitzstein / Strogatz and others) cannot be downloaded pirated — these are their official free alternatives: open courses and lecture notes.

◈
MIT 18.06 Linear Algebra (Strang)
Use alongside the Linear Algebra notes · builds the "a matrix is a linear transformation" intuition
↗
◈
MIT 18.065 Matrix Methods in Data Analysis / ML
Taught by Strang · advanced linear algebra · leads straight into machine learning
↗
◈
Harvard Stat 110 Probability (Blitzstein)
Widely considered the best probability course · use alongside the Probability notes
↗
◈
3Blue1Brown · Essence of Calculus
Builds intuition in a few hours · his Essence of Linear Algebra is recommended too
↗

Cheatsheetlook up as needed

Covers every formula from stages 01–16. Each line comes with a sympy verification command you can paste straight into a terminal.

Q
Math Formula Cheatsheet Chinese
Limits / derivatives / integrals / determinants / matrices / eigenvalues / SVD · one sympy check per line · plus 12 common mistakes
→

Correctness self-checkre-runnable

Every numerical claim and formula in the notes was verified by actually computing it. You can re-run the whole thing yourself at any time.

✓
自检脚本.py Chinese output
12 numerical checks: SVD reconstruction · normal equations · base-rate fallacy · MLE bias · multiple comparisons · gradient descent · logistic chaos …
→
How to use: read the original notes first to build the frame (20–30 minutes each) → then take an open course for systematic study → go deeper with the downloaded PDFs → the code in the notes runs in the Jupyter environment under 数学工具/.

Verified (2026-09-10): all 12 numerical claims pass; several wording problems were corrected at the same time (the condition on the base-rate fallacy, the "non-zero" eigenvalues in SVD, real vs complex equilibria, the approximation in Poisson, the distinction between Markov uniqueness and convergence; also that AAᵀ's zero eigenvalue corresponds to the left null space N(Aᵀ), and that a single action potential is a transient rather than a limit cycle).

Honest limitation: what code can verify is formulas and numbers; terminology conventions (e.g. calling something a "focus" vs a "node") and mathematical proofs cannot be fully checked by computation. These notes are a study aid — for systematic learning, follow the open course and the original textbook.