Special Topics in Machine Learning (EECS 395/495)
Class Meets: 3:30PM-4:50PM TTh, Tech L158
Instructor: Doug Downey
Office Hours: Friday 1:00PM-2:00PM, Ford 3-345
Required Textbook: Koller and Friedman. Probabilistic Graphical Models: Principles and Techniques. The MIT Press.
Probabilistic graphical models are a powerful technique for handling uncertainty in machine learning. The course will cover how probability distributions can be represented in graphical models, how inference and learning are performed in the models, and how the models are utilized for machine learning in practice.
The course objective is for students to gain an understanding of how graphical models can be used to represent probability distributions, and how to perform inference and learning. Topics include directed and undirected graphical models, exact and approximate inference methods, and supervised and unsupervised parameter and structure learning.
The first seven weeks of the course will consist of lectures regarding probabilistic graphical models and their application to machine learning tasks. Four homeworks will be assigned during this period (50% of the course grade) and a midterm exam will be given at the end of week seven (25% of the course grade). The remainder of the course will consist of student-led discussions (15% of the grade, plus anther 10% for participating during discussions led by others) on recent papers focusing on PGMs, active learning, and crowdsourcing (the paper list is included in the course calendar).
- Homeworks (4): 50% of grade
- Midterm: 25% of grade
- Paper presentation: 15% of grade
- Class participation: 10% of grade
Homeworks, Lectures, etc.
Please see the course calendar.
Will be held Thursday, November 3 (the seventh week of class).
Links to Bayes Net Packages