Matrix Methods in Machine Learning
Undergraduate course on matrix methods and their applications in machine learning.
Course Information
Instructor: Ulugbek S. Kamilov
Institution: University of Wisconsin–Madison
Term: Spring 2026
Course Description
This course introduces the linear-algebraic foundations of machine learning, with an emphasis on real-world applications of matrix methods in classification, clustering, denoising, and data analysis. Mathematical topics include systems of linear equations, least-squares regression, regularization, the singular value decomposition, and iterative algorithms. Machine learning topics include the lasso, support vector machines, kernel methods, clustering, dictionary learning, neural networks, and deep learning.
Lecture Topics
- Vectors and matrices in ML
- Matrix and vector operations
- Matrix and vector norms
- Linear independence
- Systems of linear equations
- Solutions to linear systems
- Orthogonality and subspaces
- Introduction to least squares
- Gradient descent for least squares
- Orthogonality, projections, and least squares
- Vector calculus
- Introduction to SVD
- Least-squares solutions via SVD
- Low-rank approximation and power method
- PCA and dimensionality reduction
- Linear autoencoders and ridge regression
- Review: SVD as a unifying tool in ML
- Beyond least squares
- Gradient descent
- Stochastic gradient descent (SGD)
- Subgradient and proximal methods
- Kernel Methods: From Features to Kernels
- Kernel Methods: Duality and RKHS
- Neural Networks