- #1
EngWiPy
- 1,368
- 61
Hello,
I've read in a paper that the following binomial distribution
[tex]\sum_{k=floor(N/2)+1}^N{N\choose k}\varepsilon^k(1-\varepsilon)^{N-k}[/tex]
can be upper bounded using Chernoff bound by
[tex]e^{ floor(N/2)}\,\Phi(s_0)[/tex]
where
[tex]\Phi(s)=\left(1-\varepsilon(1-e^s)\right)^N[/tex]
and
[tex]floor(N/2)\,\Phi(s_0)=\frac{\partial}{\partial s}\left. \Phi(s)\right|_{s=s_0}[/tex]
Could anyone explain to me how?
Thanks
I've read in a paper that the following binomial distribution
[tex]\sum_{k=floor(N/2)+1}^N{N\choose k}\varepsilon^k(1-\varepsilon)^{N-k}[/tex]
can be upper bounded using Chernoff bound by
[tex]e^{ floor(N/2)}\,\Phi(s_0)[/tex]
where
[tex]\Phi(s)=\left(1-\varepsilon(1-e^s)\right)^N[/tex]
and
[tex]floor(N/2)\,\Phi(s_0)=\frac{\partial}{\partial s}\left. \Phi(s)\right|_{s=s_0}[/tex]
Could anyone explain to me how?
Thanks