Comments from 050120 meeting.
parent
0054717121
commit
166bf05a6a
11
sop.tex
11
sop.tex
|
@ -84,6 +84,8 @@ For notational convenience set
|
|||
\term_2\left(\wElem_1,\ldots, \wElem_{\prodsize}, \wElem'_1,\ldots, \wElem'_{\prodsize}\right) = &\ex{\prod_{i = 1}^{\prodsize}\sine(\wElem_i)\ind{\hfunc(\wElem_i) = j}} \cdot \ex{\prod_{i = 1}^{\prodsize}\conj{\sine(\wElem'_i)}\ind{\hfunc(\wElem'_i) = j'}} \label{eq:term-2}
|
||||
\end{align}
|
||||
Focusing on $\term_1$, observe that $\term_1 = 1$ if and only if all the $\wElem_i$'s are equal, all the $\wElem'_i$'s are equal, and the two groups of variables do not equal each other,
|
||||
\AH{Need to back up the above statement with lemma 1, indicator vars, 2kwise independence of s, h}
|
||||
\AH{Both of these terms need to be fixed, we forgot the indicator variables.}
|
||||
\begin{equation*}
|
||||
\term_1\left(\wElem_1,\ldots, \wElem_{\prodsize}\right) =
|
||||
\begin{cases}
|
||||
|
@ -388,7 +390,7 @@ When $\match{\surj}{\surj'}$, with $\dw_i = \dw'_i$ for all $i \in [\dist]$, \cr
|
|||
\AH{Should we mention this weird case for $\dist = \dist' = 1$, where for all $i \in [\prodsize]$ $\dw_i = \wElem$ and for all $i' \in [\prodsize]$ $\dw'_{i'} = \wElem'$ \textit{but} $\wElem \neq \wElem'$? This case makes $\term_1 - \term_2 = 0$, and doesn't change the results.}
|
||||
|
||||
By the fact that the expectations cancel when $\forall i, i', j, j'\in [\prodsize], \wElem_i = \wElem_j = \wElem, \wElem_{i'}' = \wElem_{j'}' = \wElem'$, for both $\wElem = \wElem'$ and $\wElem \neq \wElem'$, we can rid ourselves of $\term_2$, (\cref{eq:term-2}), the case when there exists only one distinct world value. This is precisely why we have not needed to account for the last two expectations in \cref{eq:sig-j-last}. We then need to sum up all the $\dist$ distinct world value possibilities for $\dist \in [2, \prodsize]$. Starting with \cref{eq:sig-j-distinct},
|
||||
|
||||
\AH{redo bringing in the $\term_2$. Advised to separate both (in the $\sigsq_j$ case) $\term_1, \term_2$, and do separate analysis for each.}
|
||||
\begin{align}
|
||||
\sigsq_j = &\sum_{\dist = 1}^{\prodsize}\sum_{\dist' = 1}^{\prodsize}\sum_{\surj, \surj'}\sum_{\substack{\dw_{_1} \prec \cdots \prec \dw_{_\dist},\\\dw'_{1},\ldots,\dw'_{\dist'}\\ \in W}}\prod_{i = 1}^{\prodsize}\vect_i(\dw_{\surj(i)})\vect_i(\dw'_{\surj'(i)})\cdot \left(\term_1\left(\dw_{\surj(1)},\ldots,\dw_{\surj(\prodsize)}, \dw'_{\surj'(1)},\ldots, \dw'_{\surj'(\prodsize)}\right) - \term_2\left(\dw_{\surj(1)},\ldots,\dw_{\surj(\prodsize)}, \dw'_{\surj'(1)},\ldots, \dw'_{\surj'(\prodsize)}\right)\right)\nonumber\\
|
||||
= &\sum_{\dist = 2}^{\prodsize}\sum_{\surj, \surj'}\sum_{\substack{\dw_{_1} \prec \cdots \prec \dw_{_\dist},\\\dw'_{1} \prec \cdots \prec\dw'_{\dist}\\ \in W}}\prod_{i = 1}^{\prodsize}\vect_i(\dw_{\surj(i)})\vect_i(\dw'_{\surj'(i)})\cdot \term_1\left(\dw_{\surj(1)},\ldots,\dw_{\surj(\prodsize)}, \dw'_{\surj'(1)},\ldots, \dw'_{\surj'(\prodsize)}\right)\label{eq:sig-j-bnd-1}\\
|
||||
|
@ -400,11 +402,18 @@ By the fact that the expectations cancel when $\forall i, i', j, j'\in [\prodsiz
|
|||
We obtain \cref{eq:sig-j-bnd-1} by the fact that $\dist = \dist'$ and the removal of $\term_2$. We conclude with \cref{eq:sig-j-bnd-2} by \cref{lem:sig-j-survive}.% as well as bringing out the indicator variables of $\term_1$. Equation \ref{eq:sig-j-bnd-3} is derived from the fact that $\forall i \in [\dist], \dw_i = \dw'_i$. We arrive at \cref{eq:sig-j-bnd-4}, since with $\dist$ distinct variables, the product of indicator variables will result in multiplying the uniform distribution probability distribution $\dist$ times.
|
||||
|
||||
Using \cref{eq:cvar-bound} and \cref{eq:sig-j-bnd-2}, we state the general bounds for $\sigsq$,
|
||||
\AH{Needs to be redone. Missing the outer $\sum_j$ and $\sum_{j \neq j'}$. Use another letter than h for the iteration, h is the hash function.}
|
||||
\[\sigsq = \sum_{\dist = 2}^{\prodsize}\frac{1}{\sketchCols^{\dist}}\sum_{\substack{\surj, \surj'\\\match{\surj}{\surj'}}}\sum_{\substack{\dw_{_1}, \ldots, \dw_{_\dist}\\ \in W}}\prod_{i = 1}^{\prodsize}\vect_i(\dw_{\surj(i)})\vect_i(\dw_{\surj'(i)}) -
|
||||
\frac{1}{B^2}\sum_{\wElem \in W}\prod_{i = 1}^{\prodsize}v_i^2(\wElem).\]
|
||||
\AH{Can start on SOP. Another thing I could work on would be revising lemma 1.}
|
||||
\AR{Remaining TODOs: (1) Give expression for general $\sigma^2$, i.e. deal with the general $\lambda(j,j')$ term. (2) Show how to use the analysis for general $k$-product to handle generic SoP expressions-- the expectation arguments would just follow from the above and linearity of expectation but the variance bounds might need a bit of extra work.}
|
||||
|
||||
\AH{make h a subscript. Avoid superscripts generally unless you already have a subscript; make another file with notation for a quick easy table. Make every product term to have the same k product terms.
|
||||
Also, the game plan is:
|
||||
use lemma 6 for when $\ell = \ell'$
|
||||
write lemma 6' for when $\ell \neq \ell'$.
|
||||
|
||||
}
|
||||
|
||||
\section{SOP}
|
||||
\subsection{Notation}
|
||||
|
|
Loading…
Reference in New Issue