Mathematics of Machine Learning
A rigorous course on machine learning theory, covering VC bounds, boosting, convex optimization, online learning and bandits, with notes and solutions.
Machine learning finds patterns in data, and proving why its methods work takes serious mathematics. This course gives a rigorous account of those methods and their analysis, not a tour of software. It starts with statistical theory, moves through algorithms built on convexity, and ends with learning that happens online, one decision after another, with only partial information.
What You Will Achieve
- Derive generalization guarantees with concentration inequalities and VC theory
- Analyze boosting and support vector machines mathematically
- Prove convergence results for gradient and mirror descent
- Reason about regret in bandit and online learning problems
Program Overview
Statistical Theory of Classification
The first part asks when a classifier learned from a sample will work on new data.
- Binary classification
- Concentration inequalities
- The VC inequality
- Covering numbers and chaining
Convexity and Gradient Descent
Next come algorithms whose analysis depends on convex relaxations and optimization.
- Convexification and boosting
- Support vector machines
- Projected gradient and mirror descent
- Stochastic gradient descent
Online Learning and Bandits
The last part treats sequential prediction, including settings where only partial feedback arrives.
- Prediction with expert advice
- Stochastic and adversarial bandits
- Linear bandits
- Blackwell’s approachability
Course Completion
Job Outlook
This theory supports machine learning research, algorithm development and quantitative roles where method guarantees matter as much as results.
Good for: Mathematically minded students and researchers who want proofs behind machine learning methods, and practitioners moving toward theoretical work.
Keep in mind: This is proof-driven theory with no coding exercises listed, and there are no videos. It dates from 2015, so it leaves out later developments in the field.
Bottom line: A clear, rigorous route into learning theory for readers who enjoy proofs more than programming.
Frequently Asked Questions
What study materials does it provide?
It provides lecture notes and three problem sets with solutions. A final project presentation closes the original schedule, but the published materials do not include a project description.
Does this course teach machine learning programming?
No coding exercises appear in the materials. The focus is mathematical analysis of methods such as boosting, support vector machines and gradient descent, so expect proofs instead of software.
Which topics are covered in online learning?
Sessions cover prediction with expert advice, follow the perturbed leader, stochastic, adversarial and linear bandits, prediction of individual sequences and Blackwell's approachability.



There are no reviews yet.