Skip to content

Commit

Permalink
Split FGBZ into 2 pages, which was harder than I thought it would be.…
Browse files Browse the repository at this point in the history
… Polished these pages but they still need work, as outlined in the Issue #44.
  • Loading branch information
ndattani committed Sep 23, 2019
1 parent 6851db8 commit 3a4fc79
Show file tree
Hide file tree
Showing 2 changed files with 39 additions and 36 deletions.
Binary file modified Volume_1/Book_about_Quadratization.pdf
Binary file not shown.
75 changes: 39 additions & 36 deletions Volume_1/Book_about_Quadratization.tex
Original file line number Diff line number Diff line change
Expand Up @@ -3025,36 +3025,31 @@ \subsection{Reduction by Substitution (Rosenberg 1975)}
\item Used in: \cite{Perdomo2008, Bian2013}. %Perdomo2013 is protein folding problem, Bian2013 is Ramsey number problem.
\end{itemize}


\newpage


\subsection{FGBZ Reduction for Negative Terms (Fix-Gruber-Boros-Zabih, 2011)}

\summarysec

Here we consider a set $C$ of variables which occur in multiple monomials throughout the objective function.
Each application 'rips out' this common component from each term \cite{Fix2011}\cite{Boros2014}.

Let $\mathcal{H}$ be a set of monomials, where $C \subseteq H$ for each $H\in\mathcal{H}$ and each monomial $H$ has a weight $\alpha_{H}$.
The algorithm comes in 2 parts: when all $\alpha_{H}>0$ and when all $\alpha_{H}<0$. Combining the 2 gives the final method:
We consider a set $C$ of variables which can occur in multiple terms throughout the objective function.
Each application `rips out' this common component from each term \cite{Fix2011,Boros2014}.

\begin{enumerate}
\item $\alpha_{H}>0$
\begin{equation}
\sum_{H\in\mathcal{H}}\alpha_{H}\prod_{j\in H}b_{j}=\min_{b_a}\left(\sum_{H\in\mathcal{H}}\alpha_{H}\right)b_a\prod_{j\in C}b_{j}+\sum_{H\in\mathcal{H}}\alpha_{H}(1-b_a)\prod_{j\in H\setminus C}b_{j}
\end{equation}
\item $\alpha_{H}<0$
%\begin{enumerate}
%\item $\alpha_{H}>0$
%\begin{equation}
%\sum_{H\in\mathcal{H}}\alpha_{H}\prod_{j\in H}b_{j}=\min_{b_a}\left(\sum_{H\in\mathcal{H}}\alpha_{H}\right)b_a\prod_{j\in C}b_{j}+\sum_{H\in\mathcal{H}}\alpha_{H}(1-b_a)\prod_{j\in H\setminus C}b_{j}
%\end{equation}
%\item $\alpha_{H}<0$
\begin{equation}
\sum_{H\in\mathcal{H}}\alpha_{H}\prod_{j\in H}b_{j}=\min_{b_a}\sum_{H\in\mathcal{H}}\alpha_{H}\left(1-\prod_{j\in C}b_{j}-\prod_{j\in H\setminus C}b_{j}\right)b_a
\sum_{H}\alpha_{H}\prod_{j\in H}b_{j}\rightarrow\sum_{H}\alpha_{H}\left(1-\prod_{j\in C}b_{j}-\prod_{j\in H\setminus C}b_{j}\right)b_a
\end{equation}
\end{enumerate}
%\end{enumerate}

\costsec
\begin{itemize}

\item One auxiliary variable per application.
\item In combination with \ref{subsec:Negative-Monomial-Reduction}, it can be used to make an algorithm which can reduce $t$ positive monomials of degree $d$ in $n$ variables using $n+t(d-1)$ auxiliary variables in the worst case.
\item In combination with \ref{subsec:Negative-Monomial-Reduction}, it can reduce $t$ positive terms of degree $k$ in $n$ variables using $n+t(k-1)$ auxiliary variables in the worst case.
\end{itemize}

\prossec
Expand All @@ -3064,8 +3059,7 @@ \subsection{FGBZ Reduction for Negative Terms (Fix-Gruber-Boros-Zabih, 2011)}

\conssec
\begin{itemize}
\item $\alpha_{H}>0$ method converts positive terms into negative ones of same order rather than reducing them, though these can then be reduced more easily.
\item $\alpha_{H}<0$ method only works for $|C|>1$, and cannot quadratize cubic terms.
\item Cannot reduce the degree of the original function if $|C|\le 1$, and cannot quadratize anything for the other values of $|C|$ (but it can reduce their degree).
\end{itemize}

\examplesec
Expand All @@ -3087,53 +3081,60 @@ \subsection{FGBZ Reduction for Negative Terms (Fix-Gruber-Boros-Zabih, 2011)}
\item Original paper and application to image denoising: \citep{Fix2011}.
\end{itemize}

\newpage


\subsection{FGBZ Reduction for Positive Terms (Fix-Gruber-Boros-Zabih, 2011)}

\summarysec

Here we consider a set $C$ of variables which occur in multiple monomials throughout the objective function.
Each application 'rips out' this common component from each term \cite{Fix2011}\cite{Boros2014}.
We consider a set $C$ of variables which can occur in multiple terms throughout the objective function.
Each application `rips out' this common component from each term \cite{Fix2011,Boros2014}:

Let $\mathcal{H}$ be a set of monomials, where $C \subseteq H$ for each $H\in\mathcal{H}$ and each monomial $H$ has a weight $\alpha_{H}$.
The algorithm comes in 2 parts: when all $\alpha_{H}>0$ and when all $\alpha_{H}<0$. Combining the 2 gives the final method:
%Let $\mathcal{H}$ be a function, where $C \subseteq H$ for each $H\in\mathcal{H}$ and each term has a coefficient $\alpha_{H}$.
%The algorithm comes in 2 parts: when all $\alpha_{H}>0$ and when all $\alpha_{H}<0$. Combining the 2 gives the final method:

\begin{enumerate}
\item $\alpha_{H}>0$
\begin{equation}
\sum_{H\in\mathcal{H}}\alpha_{H}\prod_{j\in H}b_{j}=\min_{b_a}\left(\sum_{H\in\mathcal{H}}\alpha_{H}\right)b_a\prod_{j\in C}b_{j}+\sum_{H\in\mathcal{H}}\alpha_{H}(1-b_a)\prod_{j\in H\setminus C}b_{j}
\end{equation}
\item $\alpha_{H}<0$
%Each term $i$ of the objective function will involve some variables $b_{j_i}$

%\begin{enumerate}
%\item $\alpha_{H}>0$
\begin{equation}
\sum_{H\in\mathcal{H}}\alpha_{H}\prod_{j\in H}b_{j}=\min_{b_a}\sum_{H\in\mathcal{H}}\alpha_{H}\left(1-\prod_{j\in C}b_{j}-\prod_{j\in H\setminus C}b_{j}\right)b_a
\sum_{H}\alpha_H\prod_{j\in H} b_{j} \rightarrow \sum_{H}\alpha_{H}b_a\prod_{j\in C}b_{j}+\sum_{H}\alpha_{H}(1-b_a)\prod_{j\in H\setminus C}b_{j}.
\end{equation}
\end{enumerate}
%\item $\alpha_{H}<0$
%\begin{equation}
%\sum_{H\in\mathcal{H}}\alpha_{H}\prod_{j\in H}b_{j}\rightarrow - \sum_{H\in\mathcal{H}}\alpha_{H}\left(1-\prod_{j\in C}b_{j}-\prod_{j\in H\setminus C}b_{j}\right)b_a
%\end{equation}
%\end{enumerate}

\costsec
\begin{itemize}

\item One auxiliary variable per application.
\item In combination with \ref{subsec:Negative-Monomial-Reduction}, it can be used to make an algorithm which can reduce $t$ positive monomials of degree $d$ in $n$ variables using $n+t(d-1)$ auxiliary variables in the worst case.
\item In combination with \ref{subsec:Negative-Monomial-Reduction}, it can reduce $t$ positive terms of degree $k$ in $n$ variables using $n+t(k-1)$ auxiliary variables in the worst case.
\end{itemize}

\prossec
\begin{itemize}
\item It is a `perfect' transformation, meaning that after minimizing over $b_a$, the original degree-$k$ function is recovered.
\item Can reduce the connectivity of an objective function, as it breaks interactions between variables.
\end{itemize}

\conssec
\begin{itemize}
\item $\alpha_{H}>0$ method converts positive terms into negative ones of same order rather than reducing them, though these can then be reduced more easily.
\item $\alpha_{H}<0$ method only works for $|C|>1$, and cannot quadratize cubic terms.
\item If $|C|=1$ the first sum will result in a quadratic but the second sum will have degree $k$. If $|C|=k-1$ the second sum will be quadratic but the first term will have degree $k$. For any other value of $|C|$, both sums will be super-quadratic, but the part of the second sum involving $b_a$ will be negative and therefore can be quadratized easily.
%\item Converts positive terms into negative ones of same order rather than reducing them, though negative terms can be quadratized more easily.
%\item $\alpha_{H}<0$ method only works for $|C|>1$, and cannot quadratize any super-quadratic terms, but it can `sub-divide' a degree-$k$ term into a degree-$(|C|+1)$ term and a degree-$(k-|C|+1)$-term, so it can in fact reduce the overall degree.
\end{itemize}

\examplesec

First let $C=b_{1}$ and use the positive weight version:
With $C=b_{1}$ we can get:
\begin{eqnarray}
b_{1}b_{2}b_{3}+b_{1}b_{2}b_{4} & \mapsto & 2b_{a_1}b_{1}+(1-b_{a_1})b_{2}b_{3}+(1-b_{a_1})b_{2}b_{4}\\
b_{1}b_{2}b_{3}+b_{1}b_{2}b_{4} & \rightarrow & 2b_{a_1}b_{1}+(1-b_{a_1})b_{2}b_{3}+(1-b_{a_1})b_{2}b_{4}\\
& = & 2b_{a_1}b_{1}+b_{2}b_{3}+b_{2}b_{4}-b_{a_1}b_{2}b_{3}-b_{a_1}b_{2}b_{4}
\end{eqnarray}
now we can use \ref{subsec:Negative-Monomial-Reduction}:
now we can use \ref{subsec:Negative-Monomial-Reduction} to quadratize the two negative cubic terms:

\begin{eqnarray}
-b_{a_1}b_{2}b_{3}-b_{a_1}b_{2}b_{4} & \mapsto & 2b_{a_2}-b_{a_1}b_{a_2}-b_{a_2}b_{2}-b_{a_2}b_{3}+2b_{a_2}-b_{a_1}b_{a_2}-b_{a_2}b_{2}-b_{a_2}b_{4}\\
Expand All @@ -3146,6 +3147,8 @@ \subsection{FGBZ Reduction for Positive Terms (Fix-Gruber-Boros-Zabih, 2011)}
\end{itemize}


\newpage

\subsection{Pairwise Covers (Anthony-Boros-Crama-Gruber, 2017)}

\summarysec
Expand Down

0 comments on commit 3a4fc79

Please sign in to comment.