LFD Book Forum

LFD Book Forum (http://book.caltech.edu/bookforum/index.php)
-   Homework 7 (http://book.caltech.edu/bookforum/forumdisplay.php?f=136)
-   -   Question about Generalization result of SV's (http://book.caltech.edu/bookforum/showthread.php?t=4014)

BojanVujatovic 02-21-2013 06:22 AM

Question about Generalization result of SV's
 
Hi, I want to compliment and thank Professor and others for the wonderful set of lectures and the textbook which explain the subject of Machine Learning with extraordinary ease and clarity.

I have a question about the Generalization result of SV's, in Lecture 14, slide 20. It says that:
E[E_{out}] = \frac{E[N_{SV}]}{N-1}
I don't understand how was this bound derived (from something like the VC bound or is it an observation)? Also I am interested in knowing does it hold for any classification problem that we apply SV's for?

yaser 02-21-2013 06:39 AM

Re: Question about Generalization result of SV's
 
Quote:

Originally Posted by BojanVujatovic (Post 9458)
Hi, I want to compliment and thank Professor and others for the wonderful set of lectures and the textbook which explain the subject of Machine Learning with extraordinary ease and clarity.

I have a question about the Generalization result of SV's, in Lecture 14, slide 20. It says that:
E[E_{out}] = \frac{E[N_{SV}]}{N-1}
I don't understand how was this bound derived (from something like the VC bound or is it an observation)? Also I am interested in knowing does it hold for any classification problem that we apply SV's for?

Thank you for your kind words. The proof of the inequality makes a number of assumptions. You can find a version of it in Vapnik's book "Statistical Learning Theory."

BojanVujatovic 02-21-2013 12:49 PM

Re: Question about Generalization result of SV's
 
OK, thank you!


All times are GMT -7. The time now is 03:07 AM.

Powered by vBulletin® Version 3.8.3
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.
The contents of this forum are to be used ONLY by readers of the Learning From Data book by Yaser S. Abu-Mostafa, Malik Magdon-Ismail, and Hsuan-Tien Lin, and participants in the Learning From Data MOOC by Yaser S. Abu-Mostafa. No part of these contents is to be communicated or made accessible to ANY other person or entity.