CSCI 145: Data Mining
Instructor: R. Teal Witter. Please call me Teal.
Class Times: Mondays and Wednesdays from 2:45 to 4:00pm in Robert Day 126.
Office Hours: Tuesdays from 1:00 to 4:00pm in Adams Hall 213.
Problem Sets: Your primary opportunity to learn the material will be on problem sets. You may work with others to solve the problems, but you must write your solutions by yourself.
Quizzes: There will be short (5-minute) quizzes at the beginning of (randomly) selected classes. These quizzes will test your understanding of the problem sets and the concepts from the prior week.
Exams: The two midterm exams are the primary method of assessing your understanding of the material.
Project: The project offers a chance to explore an area that interests you, practice writing high quality code, and develop your ability to communicate technical ideas to an audience.
Resources: While we do not have a textbook, we do have readings for each lecture.
Big Picture: Modern deep learning is built from a small set of mathematical tools that return throughout the semester.
- Mathematical Foundations - Probability and linear algebra turn uncertainty and high-dimensional data into objects we can calculate with.
- Linear Models - A simple prediction rule becomes a complete learning algorithm once we choose a loss and optimize it.
- Neural Networks - Stacking linear maps with nonlinearities lets simple building blocks represent surprisingly complicated functions.
- Low-rank Structure - The singular value decomposition finds useful compressed structure in data and gives us a practical way to make large models smaller and cheaper to adapt.
- Architectures - Convolution and attention encode different ideas about which pieces of information should interact.
- Reinforcement Learning - Monte Carlo estimation lets an agent learn from delayed rewards even when no correct label is provided.