LFD Book Forum

LFD Book Forum (http://book.caltech.edu/bookforum/index.php)
-   Chapter 1 - The Learning Problem (http://book.caltech.edu/bookforum/forumdisplay.php?f=108)
-   -   Hoeffding's Inequality vs. binomial distribution (http://book.caltech.edu/bookforum/showthread.php?t=4359)

BojanVujatovic 07-06-2013 09:11 AM

Hoeffding's Inequality vs. binomial distribution
 
Considering the bin case as described in the book and the lectures, we could write the probability of \mu being close to \nu:
P[|\mu - \nu| > \epsilon]
can be written excatly since \nu is binomaily distributed with parameters \mu and N. Now, the problem is that that expression has \mu in it which is unknown. But we can find the value of \mu for which P[|\mu - \nu| > \epsilon] has its maximum, and it turns out to be \mu = 0.5.
Now if we plug that in the expression we get the bound with same properties as Hoeffding's (valid for all \mu's, N's and \epsilon's).
Now, my view is that that bound is the tightest possible, tighter than the Hoeffding's. Am I correct? It could be (perhaps?) used in futher analysis, but the major donwside is that is is not so nice nor elegant to work with. Other opinions?

Thanks!

yaser 07-06-2013 10:30 AM

Re: Hoeffding's Inequality vs. binomial distribution
 
Quote:

Originally Posted by BojanVujatovic (Post 11176)
Considering the bin case as described in the book and the lectures, we could write the probability of \mu being close to \nu:
P[|\mu - \nu| > \epsilon]
can be written excatly since \nu is binomaily distributed with parameters \mu and N. Now, the problem is that that expression has \mu in it which is unknown. But we can find the value of \mu for which P[|\mu - \nu| > \epsilon] has its maximum, and it turns out to be \mu = 0.5.
Now if we plug that in the expression we get the bound with same properties as Hoeffding's (valid for all \mu's, N's and \epsilon's).
Now, my view is that that bound is the tightest possible, tighter than the Hoeffding's. Am I correct? It could be (perhaps?) used in futher analysis, but the major donwside is that is is not so nice nor elegant to work with.

You are correct. Different approaches to similifying the expression for this bound can lead to different bounds of varying tightness and elegance.


All times are GMT -7. The time now is 04:58 AM.

Powered by vBulletin® Version 3.8.3
Copyright ©2000 - 2019, 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.