Linear Algebra and Learning from Data
Synopsis
Gilbert Strang wrote this textbook for MIT's course 18.065, Matrix Methods in Data Analysis, Signal Processing, and Machine Learning, and it reads like his lectures: informal in tone, built on worked examples, and focused on why each idea matters for data. It opens with a fresh look at matrix multiplication by columns and the four fundamental subspaces, then makes the singular value decomposition the center of the book, using it to explain principal component analysis, low-rank approximation and least squares. Later parts cover special matrices, computation with very large matrices including randomized methods, compressed sensing, probability and statistics, and optimization from gradient descent to stochastic gradient descent.
The final part turns to the construction of deep neural networks, explaining backpropagation as the chain rule organized by matrices and showing how piecewise linear activation functions let networks carve up their input space. The payoff for readers is fluency in the mathematics underneath machine learning code: what a weight matrix does, why low-rank structure appears everywhere, and how optimization behaves. The book assumes a first course in linear algebra and some calculus, and its free companion videos on MIT OpenCourseWare make self-study practical. It does not cover transformers or attention, so readers must extend its tools to modern architectures themselves.
Scorecard
Editorial score
How it scored
Editorial score: 74.8 / 1005 weighted criteria
- 21.0/30
Influence30%
- 22.5/25
Authority25%
- 13.0/20
Practical value20%
- 11.3/15
Currency15%
- 7.0/10
Readability10%
- 74.8/100
Editorial score
How the rank was computed
Rank No. 275 of 500 comes from a composite of 53.9: observable signals 20.1 of 55 points plus editorial review 33.8 of 45 points.
All 11 inputs
| Input | Points |
|---|---|
| Observable signals | 20.1/55 |
| Reach (Amazon ratings volume) | 8.9/16.5 |
| Consensus across best-of lists | 2.8/13.8 |
| Awards and bestseller evidence | 0.0/8.3 |
| Reader rating, sample-adjusted | 7.1/8.3 |
| Amazon sales rank | 1.3/5.5 |
| Recency | 0.0/2.8 |
| Editorial review | 33.8/45 |
| Author authority | 10.1/11.3 |
| Influence | 9.5/13.5 |
| Practical value | 5.9/9.0 |
| Readability | 3.2/4.5 |
| Currency | 5.1/6.8 |
Each criterion is scored from 1 to 10 and weighted, and the points add up to the editorial score. The rank combines observable signals (55 percent) with this score (45 percent). Read the methodology.
Verdict
Strang's rigorous yet conversational bridge from matrix methods to deep learning, best used alongside his free MIT 18.065 lectures.
Strengths
- Clear, conversational explanations from a master teacher
- Paired with free MIT lecture videos
- Connects classical linear algebra to modern learning
Keep in mind
This is a mathematics text without code or framework guidance, and it omits attention and transformers, so practitioners must bridge to current architectures themselves.
Key takeaways
- The singular value decomposition is the central tool of data science, underlying principal component analysis, low-rank approximation and compression of large datasets.
- Viewing matrix multiplication as combinations of columns and rank-one pieces makes factorizations and many machine learning operations far easier to analyze.
- Stochastic gradient descent and its variants are the workhorse of training, and their behavior depends on step size, conditioning and problem structure.
- A deep network with ReLU activations computes a continuous piecewise linear function, and backpropagation is the chain rule applied efficiently through matrices.
Notable facts
- The book is the assigned textbook for MIT's 18.065, Matrix Methods in Data Analysis, Signal Processing, and Machine Learning, whose lectures are published free on MIT OpenCourseWare. Source (opens in a new tab)
- MIT News described it as Strang's 12th textbook and reported that his 18.06 linear algebra course had passed 10 million views on OpenCourseWare. Source (opens in a new tab)
About the author
Questions
What is Linear Algebra and Learning from Data about?
What background do I need for Linear Algebra and Learning from Data?
Is Linear Algebra and Learning from Data still useful in 2026?
Related books
- High-Dimensional Probability Roman Vershynin
- Probabilistic Graphical Models Daphne Koller and Nir Friedman
- Statistics, Data Mining, and Machine Learning in Astronomy Željko Ivezić, Andrew J. Connolly, Jacob T. VanderPlas and Alexander Gray
- Causal Inference Scott Cunningham
- High-Dimensional Statistics Martin J. Wainwright
- Learning From Data Yaser S. Abu-Mostafa, Malik Magdon-Ismail and Hsuan-Tien Lin
Sources
As an Amazon Associate, Best AI Books earns from qualifying purchases. Best AI Books is sponsored by Iternal.ai, whose founder and CEO, John Byron Hanby IV, wrote The AI Strategy Blueprint, our #1 ranked book. How we rank books.






