EECS 598 Unsupervised Feature Learning

Instructor: Prof. Honglak Lee
Instructor webpage:
Office hours: Th 5pm-6pm, 3773 CSE
Classroom: 1690 CSE
Time: M W 10:30am-12pm

Course Description
Machine learning has proved a powerful tool for artificial intelligence and data mining problems. However, its success has usually relied on having a good feature representation of the data, and having a poor representation can severely limit the performance of learning algorithms. These feature representations are often hand-designed, require significant amounts of domain knowledge and human labor, and do not generalize well to new domains.

To address these issues, there has been much interest in algorithms that learn feature hierarchies from unlabeled data. For example, methods such as deep belief networks, sparse coding-based methods, convolutional networks, and deep Boltzmann machines, have shown promise and have been successfully applied to a variety of tasks in computer vision, audio processing, natural language processing, information retrieval, and robotics.

In this seminar course, we will focus on reviewing principles and recent progress in unsupervised feature learning algorithms, with a goal of developing useful features for machine learning applications. Topics include clustering, sparse coding, autoencoders, restricted Boltzmann machines, and deep belief networks. The course will require an open-ended research project.

Course Schedule

Text books
There is no required textbooks.

EECS 492 or 545 (desirable but not required);
Linear algebra and some knowledge of probability and statistics;
Programming experience in MATLAB, python, or C.

Attendance and participation: 10%
Homework: 10%
Paper presentations: 30%
Project: 50%

There will be a programming assignment to implement a basic unsupervised feature learning algorithm.

Paper presentation
Each student will present the assigned papers and lead discussions for one or two classes. All students are required to read the assigned papers before class.

This course is intended to be research-oriented. Students are encouraged to develop new unsupervised feature learning algorithms, find interesting applications, or apply to their own research problems. Please talk to the instructor before deciding about the project topic. There will be final presentations and report. Grading will be based on progress report (10%), final report(30%), and final presentation (10%).