CX 4803 CML Computational Foundations of Machine Learning

Course Description

This course gives an introduction to the field of machine learning from a computational and mathematical point of view. The course will provide the mathematical background needed to develop intuition for deeply understanding machine learning algorithms. Statistics, numerical optimization, and linear algebra will be covered at a fundamental level necessary for quickly grasping and extending the main ideas behind machine learning.

In addition, we will implement machine learning algorithms and experiment with how they work on different types of data, to understand how to choose one algorithm over another. Students will implement their own algorithms from scratch, and we will not be using machine learning software frameworks.

This course is of interest to students who wish to build a foundation to pursue research in machine learning, as well as students wishing to pursue careers as algorithm designers and software framework developers for machine learning.

Prerequisites

Programming in any language, multivariable calculus (MATH 2551/2552), linear algebra (MATH 1554), and concepts in probability and statistics.

Some Topics

• Statistics background
• Regularization
• Kernel methods
• Bayesian regression and classification
• Gaussian processes
• MCMC and sampling algorithms
• Support vector machines
• Constrained optimization and duality
• Neural networks and stochastic optimization
• Variational inference

Grading

100% Assignments.
There will be an assignment about every 1.5 weeks. Many assignments will build on previous assignments. Note that the final assignment will be due during final exam week.

Students are expected to read announcements sent via Canvas. Canvas will be used for posting, submiting, and grading assignments. The late penalty is 1 percent per hour, automatically deducted by Canvas.Letter grade thresholds for A, B, C, D are 90, 80, 70, 60 percent. Pass/fail threshold is 60 percent. For audit credit, the threshold is 20 percent.

Demos

We will use these demos in the course: ML-MATLAB.