LFD Book Forum (http://book.caltech.edu/bookforum/index.php)
-   Homework 1 (http://book.caltech.edu/bookforum/forumdisplay.php?f=130)
-   -   Where to find explanation for Hoeffding's Inequality ? (http://book.caltech.edu/bookforum/showthread.php?t=3880)

 alinsoar 01-19-2013 01:07 AM

Where to find explanation for Hoeffding's Inequality ?

I am looking for a detailed explanation for this.

It seems that this inequality is the mother of ml.

I skim the videos of Andrew Ng and cannot find the place where it is detailed.

Does some of you help me to find a good resource, with proof and examples ?

Thanks.

 htlin 01-19-2013 08:12 AM

Re: Where to find explanation for Hoeffding's Inequality ?

If you want to try proving by yourself, please feel free to visit Homework e/2 of NTU ML Class 2012:

http://www.csie.ntu.edu.tw/~htlin/co...oc/hw0_5_e.pdf

Hope this helps.

 alinsoar 01-19-2013 10:16 PM

Re: Where to find explanation for Hoeffding's Inequality ?

Thank you !

This is exactly what I need.

However, I need some guidance to make a proof of each step. Otherwise, alone, I suppose I will spend many hours on each point ...

 alinsoar 01-21-2013 12:50 AM

Re: Where to find explanation for Hoeffding's Inequality ?

Somebody told me to look over this book, and after I consulted the book I definitively decided to buy it.

It is the best book fitted on the field I ever saw.

H's inequality is in the 5th chapter well proved and explained.

 Michael Reach 04-03-2013 08:21 PM

Re: Where to find explanation for Hoeffding's Inequality ?

Quote:
 Originally Posted by alinsoar (Post 8847) I am looking for a detailed explanation for this. It seems that this inequality is the mother of ml.
Interesting. I thought we were now studying the theoretical underpinnings of why ML works - but that the real mother of ML is the million dollar prize from Netflix. That is, the mother of ML is that there are effective techniques that can be seen to work for prediction: You know it works, because you leave some of the data over for testing, and when you're done, the g hypothesis does a very good job predicting on the new test data.
When you have that, you have Machine Learning, and the theoreticians can come later and explain why. Am I wrong?

As an example, I do know that Quantum Computing was no more than an interesting suggestion by various people (Feynman) until Shor came up with an actual working algorithm to factor very large numbers. [Working, if you'll just build a quantum computer.] That's when the field took off.

 All times are GMT -7. The time now is 04:35 PM.