Compare commits
9 Commits
810784419a
...
master
Author | SHA1 | Date | |
---|---|---|---|
19c544e04c | |||
642260d568 | |||
98f0985195 | |||
848a0b5c6b | |||
114b2896eb | |||
f464f47e17 | |||
421bbca37d | |||
0ddd930119 | |||
1361b5fac1 |
@@ -7,19 +7,19 @@ jobs:
|
||||
|
||||
steps:
|
||||
- name: Check out the repository
|
||||
uses: actions/checkout@v4
|
||||
uses: http://localhost:3000/sxlxc/checkout@v4
|
||||
|
||||
- name: Compile LaTeX using local TeX Live
|
||||
# These commands run directly in your machine's shell
|
||||
run: |
|
||||
echo "Compiling document..."
|
||||
latexmk -pdf poster.tex
|
||||
latexmk -pdf slides.tex
|
||||
latexmk poster.tex -pdf
|
||||
latexmk slides.tex -pdf
|
||||
|
||||
- name: List files in the workspace
|
||||
run: ls -l
|
||||
|
||||
- uses: akkuman/gitea-release-action@v1
|
||||
- uses: http://localhost:3000/sxlxc/gitea-release-action@v1
|
||||
with:
|
||||
body: ''
|
||||
prerelease: true
|
||||
|
@@ -7,8 +7,8 @@
|
||||
\ProvidesPackage{beamerthemePoster}
|
||||
|
||||
\RequirePackage[dvipsnames]{xcolor}
|
||||
% \RequirePackage{tikz}
|
||||
% \usetikzlibrary{arrows,decorations.pathmorphing,backgrounds,calc}
|
||||
\RequirePackage{tikz}
|
||||
\usetikzlibrary{arrows,decorations.pathmorphing,backgrounds,calc}
|
||||
% \RequirePackage{lmodern}
|
||||
% \RequirePackage{textcomp}
|
||||
\RequirePackage{amsmath,amssymb,amsthm}
|
||||
@@ -185,7 +185,7 @@
|
||||
\end{center}
|
||||
\end{minipage}
|
||||
\end{center}
|
||||
\vskip1cm
|
||||
\vskip0.8cm
|
||||
}
|
||||
|
||||
%
|
||||
@@ -195,11 +195,11 @@
|
||||
\renewcommand{\section}[1]{
|
||||
\par\vskip\medskipamount%
|
||||
%
|
||||
% \begin{flushleft}
|
||||
% \begin{tikzpicture}[remember picture,overlay]
|
||||
% \shade [inner color=color2,outer color=color3] (0,0) rectangle (\columnwidth,0.3cm);
|
||||
% \end{tikzpicture}
|
||||
% \end{flushleft}
|
||||
\begin{flushleft}
|
||||
\begin{tikzpicture}[remember picture,overlay]
|
||||
\shade [inner color=color2,outer color=color3] (0,0) rectangle (\columnwidth,0.3cm);
|
||||
\end{tikzpicture}
|
||||
\end{flushleft}
|
||||
% \vspace{5pt}
|
||||
%
|
||||
\begin{center}
|
||||
@@ -208,12 +208,12 @@
|
||||
{\parskip0pt\par}
|
||||
\end{center}
|
||||
%
|
||||
% \begin{flushleft}
|
||||
% \vskip-1cm
|
||||
% \begin{tikzpicture}[remember picture,overlay]
|
||||
% \shade [inner color=color2,outer color=color3] (0,0) rectangle (\columnwidth,0.3cm);
|
||||
% \end{tikzpicture}
|
||||
% \end{flushleft}
|
||||
\begin{flushleft}
|
||||
\vskip-1cm
|
||||
\begin{tikzpicture}[remember picture,overlay]
|
||||
\shade [inner color=color2,outer color=color3] (0,0) rectangle (\columnwidth,0.3cm);
|
||||
\end{tikzpicture}
|
||||
\end{flushleft}
|
||||
% \vspace{2.5pt}
|
||||
%
|
||||
{\parskip0pt\par}
|
||||
|
BIN
image/chatgpt.png
Normal file
BIN
image/chatgpt.png
Normal file
Binary file not shown.
After Width: | Height: | Size: 4.5 MiB |
BIN
image/poster.png
Normal file
BIN
image/poster.png
Normal file
Binary file not shown.
After Width: | Height: | Size: 428 KiB |
16
poster.tex
16
poster.tex
@@ -39,31 +39,31 @@
|
||||
|
||||
|
||||
\begin{document}
|
||||
|
||||
% larger font
|
||||
\large
|
||||
|
||||
\begin{frame}[t]
|
||||
\vspace{-3cm}
|
||||
\begin{multicols}{2}
|
||||
|
||||
\section{Problem}
|
||||
We consider the incentive allocation problem with additional constraints.
|
||||
|
||||
\textbf{Input}: A set of coupons $E=\bigcupdot_i E_i$, where each coupon $e\in E$ has value and cost $v_e,c_e\in \mathbb{Z}_+$. Budget $B\in \mathbb{Z}_+$. Constraints $\mathcal F_i$ on each subset $E_i$.
|
||||
\textbf{Input}: A set of coupons $E=\bigcupdot_i E_i$, where each coupon $e\in E$ has a value and a cost $v_e,c_e\in \mathbb{Z}_+$. Budget $B\in \mathbb{Z}_+$. Constraints $\mathcal F_i$ on each subset $E_i$.
|
||||
|
||||
\textcolor{Gray}{
|
||||
\textbf{Output}: A subset $X\subset E$ of coupons that maximizes the total value $\sum_{e\in X}v_e$ while satisfying $\sum_{e\in X}c_e\leq B$ and additional constraints $X\cap E_i\in \mathcal F_i$.
|
||||
\textbf{Output}: A subset $X\subset E$ of coupons that maximizes the total value $\sum_{e\in X}v_e$ while satisfying the budget constraint $\sum_{e\in X}c_e\leq B$ and additional constraints $X\cap E_i\in \mathcal F_i$.
|
||||
}
|
||||
|
||||
This problem is NP-hard. Consider its LP relaxation.
|
||||
\begin{equation}\label{LP}
|
||||
\begin{equation*}\label{LP}
|
||||
\begin{aligned}
|
||||
\tau(B)=\max_x&\; & v\cdot x& & & \\
|
||||
s.t.&\; & c \cdot x &\leq B & &\\
|
||||
& & x_{E_i}&\in \conv(\mathcal{F}_i) & &\;\forall i\in [n]\\
|
||||
& & x&\in [0,1]^m & &
|
||||
\end{aligned}
|
||||
\end{equation}
|
||||
\end{equation*}
|
||||
\textbf{Output}: The entire curve $\tau(B)$ for $B\in [0,\infty)$.
|
||||
|
||||
We consider 3 cases of additional constraints $x_{E_i}\in \mathcal{F}_i$ :
|
||||
@@ -101,17 +101,17 @@ We consider 3 cases of additional constraints $x_{E_i}\in \mathcal{F}_i$ :
|
||||
\end{table}
|
||||
|
||||
\section{Methods}
|
||||
The idea is to take advantage of the independence among the constraints $\mathcal{F}_i$ and to reduce the optimization problem to one in computational geometry.
|
||||
The idea is to take advantage of the independence among the constraints $\mathcal{F}_i$ and reduce the optimization problem to one in computational geometry.
|
||||
|
||||
\textcolor{DarkOrchid}{\textit{Signature Function.}} Let $f_i(\lambda) = \max\{(v_{E_i}-\lambda c_{E_i}) x | x\in \conv(\mathcal F_i) \}$ be the signature function of agent $i$. The signature function is piecewise-linar and convex.
|
||||
|
||||
\textcolor{DarkOrchid}{\textit{Lagrangian Dual.}} The Lagrangian dual of LP\autoref{LP} is therefore
|
||||
\begin{equation}
|
||||
\begin{equation*}
|
||||
\label{eq:Lagrangiandual}
|
||||
\begin{aligned}
|
||||
\min_{\lambda} \left( B\lambda+\sum_i f_i(\lambda)\right).
|
||||
\end{aligned}
|
||||
\end{equation}
|
||||
\end{equation*}
|
||||
|
||||
\begin{theorem}[4]\large
|
||||
$\tau(B)$ is piecewise-linear and concave.
|
||||
|
36
slides.tex
36
slides.tex
@@ -4,6 +4,15 @@
|
||||
\usepackage{algo}
|
||||
\usepackage{soul}
|
||||
% \usepackage{cancel}
|
||||
\newcommand{\munepsfig}[3][scale=1.0]{% <===============================
|
||||
\begin{figure}[!htbp]
|
||||
\centering
|
||||
\vspace{2mm}
|
||||
\setlength{\fboxrule}{#3} % <===================================
|
||||
\framebox{\includegraphics[#1]{#2.png}} % <=====================
|
||||
\label{fig:#2}
|
||||
\end{figure}
|
||||
}
|
||||
|
||||
\title[Incentive allocation]{Large-Scale Trade-Off Curve Computation for Incentive Allocation with Cardinality and Matroid Constraints}
|
||||
\date{August 30, 2025}
|
||||
@@ -28,8 +37,9 @@ A ride sharing company wants to send riders promotional coupons in the hope of m
|
||||
|
||||
% each agent gets at most 1 coupon.
|
||||
\begin{figure}
|
||||
placeholder\\
|
||||
\includegraphics[width=.5\textwidth]{image/landscape.png}
|
||||
\includegraphics[width=.7\textwidth]{image/chatgpt.png}
|
||||
|
||||
\scriptsize Image courtesy: ChatGPT-5
|
||||
\end{figure}
|
||||
|
||||
\end{frame}
|
||||
@@ -37,24 +47,24 @@ placeholder\\
|
||||
\begin{frame}{Multiple-choice knapsack}
|
||||
\textbf{Input}: $n$ sets of coupons $K_1,\dots,K_n$. Each coupon $e\in K_i$ has a non-negative cost $c_e\in \Z_+$ and value $v_e\in \Z_+$. A positive budget $b\in \Z_+$.
|
||||
|
||||
\textbf{Output}: A (multi)set of coupons $K$ that maximizes the total value $\sum_{e\in K} c_e$ while satisfying \textcolor{Red}{$|K\cap K_i|\leq 1$} and $\sum_{e\in K} c_e\leq b$.
|
||||
\textbf{Output}: A subset of coupons $K$ that maximizes the total value $\sum_{e\in K} v_e$ while satisfying \textcolor{Red}{$|K\cap K_i|\leq 1$} and $\sum_{e\in K} c_e\leq b$.
|
||||
|
||||
\vspace{1em}
|
||||
\pause
|
||||
|
||||
Three problems with this modeling:
|
||||
Three problems with this formulation:
|
||||
\begin{enumerate}
|
||||
\item Finding the exact optimum is NP-hard. So we consider solving it approximately.
|
||||
\item Companies may run multiple campaigns at the same time. So a trade-off curve between budget and profit will be useful.
|
||||
\item The multiple-choice constraint \textcolor{Red}{$|K\cap K_i|\leq 1$} is too weak for real applications.
|
||||
\item The multiple-choice constraint \textcolor{Red}{$|K\cap K_i|\leq 1$} is too strong for real-world applications.
|
||||
\end{enumerate}
|
||||
|
||||
\end{frame}
|
||||
|
||||
\begin{frame}{Linear programming formulation}
|
||||
\textcolor{gray}{
|
||||
\begin{frame}{Linear programming relaxation}
|
||||
% \textcolor{gray}{
|
||||
\textbf{Input}: $n$ sets of coupons $K_1,\dots,K_n$. Each coupon $e\in K_i$ has a non-negative cost $c_e\in \Z_+$ and value $v_e\in \Z_+$. \st{A positive budget $b\in \Z_+$.}
|
||||
}
|
||||
% }
|
||||
\pause
|
||||
\begin{equation*}
|
||||
\begin{aligned}
|
||||
@@ -64,7 +74,7 @@ s.t.& & c\cdot x&\leq b & &\\
|
||||
\end{aligned}
|
||||
\end{equation*}
|
||||
|
||||
\textbf{Output}: A compact representation of $\tau(b)$.
|
||||
\textbf{Output}: function $\tau(b)$ for $b\in (0,+\infty)$.
|
||||
\pause
|
||||
|
||||
We focus on 2 kinds of constraints of \textcolor{Plum}{$x_{K_i}\in P_{K_i}$}.
|
||||
@@ -88,6 +98,12 @@ $k=O(mp^{1/3})$ and $\tau$ can be computed in $O((k+m)\log m)$ time.
|
||||
\end{frame}
|
||||
|
||||
\begin{frame}[plain]
|
||||
poster
|
||||
\centering
|
||||
\Large \textit{Let's discuss this in detail at my poster!}
|
||||
|
||||
\munepsfig[width=.8\linewidth]{image/poster}{1pt}
|
||||
% \begin{figure}
|
||||
% \includegraphics[width=0.8\linewidth]{image/poster.png}
|
||||
% \end{figure}
|
||||
\end{frame}
|
||||
\end{document}
|
Reference in New Issue
Block a user