Title: The Reduction Approach to Machine Learning
Time: October 12-13, 2006
Location: Room B222, Exactum, Department of Computer Science, University of Helsinki, Gustaf Hällströmin katu 2b
Schedule: Lectures 12-15.
Grading: If you want to get credits (2-6 credit units), choose one of the following homework projects.
The reduction approach to machine learning solves learning problems by reducing them to known learning problems and applying known algorithms. I will discuss this approach in sufficient detail so that anyone may use (and improve) it in two 3 hour lectures (with breaks, naturally).
John Langford's page Machine Learning Reductions contains also the slides for this course.
On the first day, I'll cover the basics:
- What is a learning reduction?
- How do you reduce between various learning problems?
- What are known reductions results?
First day slides:
1 per page
4 per page
On the second day, I'll cover more advanced topics:
- How do you reduce complex problems into known primitives?
- Reductions work which is currently in progress.
- Bothersome open problems.
Second day slides:
1 per page
4 per page
---
Passing the course: If you want to get credits (2-6 credit units), choose one of the following homework projects.
- Apply the reductions approach to your own problem, and report what you learned.
- (Try to) make some progress on one of the (bothersome) open problems.
For more details on the project, credits, and other practical issues, please contact Matti Kääriäinen (matti.kaariainen@cs.helsinki.fi).