Matrix Methods in Data Analysis, Signal Processing, and Machine Learning
Linear algebra for data science and deep learning: SVD, least squares, low rank, gradient descent, neural nets and CNNs.
Linear algebra is the language of machine learning, and this course by Gilbert Strang reviews it with applications to probability, statistics and optimization, and above all to a full explanation of deep learning. In recorded lectures it moves through matrix factorizations, eigenvalues and the singular value decomposition, low-rank approximation and least squares, then calculus of matrices, gradient descent and its stochastic and accelerated forms, and finally the structure of neural networks, backpropagation, convolutional networks and clustering in graphs.
What You Will Achieve
- Factor matrices and use eigenvalues and the singular value decomposition
- Solve least squares problems and find low-rank approximations
- Apply gradient descent, momentum and stochastic gradient descent to optimization
- Explain how neural networks are structured and trained by backpropagation
Program Overview
Highlights of Linear Algebra
Column spaces, factorizations and the singular value decomposition.
- Multiplying and factoring matrices
- Eigenvalues, eigenvectors and positive definite matrices
- The SVD and the Eckart-Young theorem
- Norms and four ways to solve least squares
Computation and Optimization
Computing with large matrices and minimizing functions.
- Randomized matrix multiplication and low rank changes
- Derivatives of matrices and singular values
- Gradient descent, momentum and linear programming
- Stochastic gradient descent
Learning from Data
Neural networks and the matrices inside them.
- Structure of neural nets and backpropagation
- Circulant matrices and the Fourier matrix
- Convolutional neural networks
- Distance matrices and clustering in graphs
Course Completion
Job Outlook
Matrix methods are the working mathematics of machine learning engineering, data science, signal processing and scientific computing.
Good for: Learners heading into machine learning who want the linear algebra behind it explained properly.
Keep in mind: The problem sets refer to sections of a textbook that is not included, and two in-class computing sessions have no video.
Bottom line: A clear, well-built course that shows how linear algebra drives deep learning.
Frequently Asked Questions
What does the group project involve?
It replaces the last three homework sets: in groups of two or three, you take one topic further, such as SVD for PCA, sparse solutions, or learning weights in a neural net.
Is there a podcast?
Yes. A Chalk Radio podcast episode and a video with Professor Gilbert Strang describe how he taught the course.
Does it touch on the Julia language?
Yes. The last session features Alan Edelman and the Julia language, after the sessions on distance matrices and clustering.



There are no reviews yet.