Networks for Learning: Regression and Classification
Course facts
- Payment: Free, no certificate
- Credits: Non Credit
- Learning style: Self-paced, start anytime
- Resources: Problem Sets; Written Assignments
MIT OpenCourseWare details
- As taught in: Spring 2001
- Department: Brain and Cognitive Sciences
| Course Details | ||||||
|---|---|---|---|---|---|---|
|
||||||
| Outcomes | ||||||
|
||||||
| Price & Terms | ||||||
|
||||||
| Technical | ||||||
|
||||||
| Specification | ||||||
|
||||||
Statistical learning theory for supervised learning: regularization in RKHS, VC theory, support vector machines, boosting and applications.
Supervised learning asks how a machine can generalize from a limited set of examples. This advanced course treats the problem within statistical learning theory: it reviews classical techniques such as regularization in reproducing kernel Hilbert spaces for approximating functions from sparse data, then develops VC theory and uses it to justify regularization networks and support vector machines. Research topics such as boosting and feature selection follow, with applications in computer vision, graphics, finance and neuroscience.
What You Will Achieve
- Frame supervised learning as a problem of function approximation
- Use regularization theory in reproducing kernel Hilbert spaces
- Apply VC theory and structural risk minimization to learning
- Explain support vector machines for classification and regression
Program Overview
Regularization and Approximation
Classical ways to learn functions from sparse data.
- The learning problem in perspective
- Reproducing kernel Hilbert spaces
- Nonparametric techniques and regularization networks
Statistical Learning Theory
The theory that justifies learning machines.
- Consistency of empirical risk minimization
- VC dimension and VC bounds
- Support vector machines for classification and regression
Current Topics of Research
Research directions and applications.
- Kernel engineering
- Feature selection with support vector machines
- Bagging and boosting
- Wavelets and frames
Course Completion
Job Outlook
This theory underpins research and advanced work in machine learning, computer vision and quantitative modeling.
Good for: Advanced students with strong mathematics who want the theoretical foundations of kernel methods and support vector machines.
Keep in mind: A spring 2001 course whose posted materials are two problem sets and a list of term paper projects; there are no lecture notes, and the research topics reflect the field as it stood then.
Bottom line: A rigorous outline of classical learning theory, useful mainly as a map of topics and projects.
Frequently Asked Questions
What could the term paper be about?
Fourteen suggested projects, such as kernels for strings, feature selection for SVMs, stability and generalization, and bagging and boosting.
Which applications are discussed?
Sessions apply learning techniques to finance, computer vision and computer graphics, and two sessions look at neuroscience.
Are there problem sets?
Yes. Two problem sets are posted, alongside the description of the term paper and its project list.
User Reviews
Write your review
You must be logged in to post a review.
Reviews can only be posted by registered learners. Log in or create a free account, then you will return here to post.
There are no reviews yet.