This is a light, fun, 75min/week undergraduate class designed for JHU's HEART.
- All administrative info (meeting times, calendar, grading policy, books/equipment needed) is contained in the syllabus, in the
course_infofolder (download). - All lessons are PDF files in the
course_contentfolder. You can click through the links on github to access individual files, or click "code" in the top right to download everything as a zipped folder.
Most classes on machine learning, mathematical modeling, and statistics focus on how to use a specified model. They give students the technical skills to minimize a cost function, solve a fluid flow equation, train a neural network on a task, or construct a 95% margin of error for a relative risk estimate. Compared to the voluminous answers for "How?", less attention is given to "When?" and "Why?" one would use each method.
This course will balance the scales, helping students guide their choice of methods by offering a menu of principles. Each meeting will be themed around one principle, such as "a method should tolerate wild outliers," "a method should tolerate certain mistaken assumptions," or "a method should make optimal use of limited data." Using computer simulations, students will evaluate multiple methods with respect to each principle, exploring the limits of alternative methods and rendering judgements about when they are usable. The course will introduce the basics of statistics and computer programming -- not comprehensively but with enough detail to participate fully.