CS1420_Lecture_13.pdf - CS142 Machine Learning Spring 2017 Lecture 13 Instructor Pedro Felzenszwalb Scribes Dan Xiang Tyler Dae Devlin The

# CS1420_Lecture_13.pdf - CS142 Machine Learning Spring 2017...

• Notes
• 4

This preview shows page 1 - 2 out of 4 pages.

CS142: Machine Learning Spring 2017 Lecture 13 Instructor: Pedro Felzenszwalb Scribes: Dan Xiang, Tyler Dae Devlin The generalization problem In this lecture we address what is perhaps the most important theoretical question in all of machine learning: How do we know that low training error will translate to low test error? PAC learning “PAC” is an acronym that stands for probably approximately correct . Before we can understand where this phrase comes from, we first need to recall the setup for binary classification. The goal of binary classification is to approximate some unknown target func- tion f : X → {- 1 , +1 } . Suppose D is an unknown distribution over the input space X . We observe labeled examples ( x 1 , y 1 ) , . . . , ( x n , y n ), where the x i ’s are sampled i.i.d. according to the distribution D and the corresponding y i ’s are given by the hidden target function f , i.e. y i = f ( x i ). Given these labeled examples, our task is to select a function h from a hypothesis set H such that h f . For the linear classifier, the hypothesis space H is the set of all functions of the form h ( x ) = ( +1 w T x > 0 - 1 w T x 0 , and a particular hypothesis h is completely specified by its weight vector w . Two types of errors 1. A finite set of training data may not be sufficient to identify the target function f exactly. As a result, the hypothesis h that we end up choosing may be close — but not exactly equal — to f . We can quantify how different h

Subscribe to view the full document.

• Spring '14

### What students are saying

• As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

Kiran Temple University Fox School of Business ‘17, Course Hero Intern

• I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

Dana University of Pennsylvania ‘17, Course Hero Intern

• The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

Jill Tulane University ‘16, Course Hero Intern