343 lines
10 KiB
TeX

\documentclass{beamer}
\usetheme{Boadilla}
\usecolortheme{dolphin}
\usepackage{graphicx}
\begin{document}
\title[Participatory Budgeting]{Participatory Budgeting}
\subtitle{Algorithms and Complexity}
\author{Tobias Eidelpes}
\begin{frame}
\maketitle
\end{frame}
\begin{frame}
\frametitle{Table of Contents}
\tableofcontents
\end{frame}
\section{Introduction}
\begin{frame}
\frametitle{What is Participatory Budgeting?}
\begin{quote}
Participatory Budgeting (PB) is a democratic process in which
community members decide how to spend part of a public budget.
\end{quote}
\end{frame}
\begin{frame}
\frametitle{How does it work?}
\begin{itemize}
\item Designing the Process
\item Collecting Ideas
\item Developing Proposals
\item Voting
\item Funding Winners
\end{itemize}
\end{frame}
\begin{frame}
\frametitle{Benefits of Participatory Budgeting}
\begin{itemize}
\item More efficient spending
\item Diverse participants
\item Higher voter satisfaction
\item Democratic and citizenship learning
\item Institutional and political change
\end{itemize}
\end{frame}
\section{Computational Aspects}
\begin{frame}
\frametitle{Computational Aspects of PB}
\begin{itemize}
\item Discrete or continuous projects?
\item How do we model preferences mathematically?
\item How do we adequately capture voter's preferences?
\item How do we aggregate votes?
\end{itemize}
\end{frame}
\begin{frame}
\frametitle{Decision Space}
\begin{figure}
\centering
\includegraphics[width=\textwidth]{taxonomy.png}
\end{figure}
\end{frame}
\begin{frame}
\frametitle{Bounded Divisible PB}
\begin{itemize}
\item Projects are divisible
\item A cap for each project is defined
\item Fractional funding
\end{itemize}
\begin{exampleblock}{Example}
A project that seeks to donate a bounded amount of money to a
charity.
\end{exampleblock}
\end{frame}
\begin{frame}
\frametitle{Unbounded Divisible PB}
\begin{itemize}
\item Projects are divisible
\item No caps for projects
\item Generalizable to \emph{Portioning}
\end{itemize}
\begin{block}{Unbounded Divisible PB} $x_p = [0,1]$ denotes the
fraction of project $p\in P$ that is completed and $c_p(x_p) =
x_p$ is the cost function of project $p$. The set of feasible
budget allocations under a budget $B = 1$ is therefore defined
as \[ \{ \vec{x} : \sum_{p\in P}{x_p}\leq 1 \}. \]
\end{block}
\begin{exampleblock}{Example}
A project that seeks to donate an unbounded amount of money to a
charity. Every additional amount can be used effectively.
\end{exampleblock}
\end{frame}
\begin{frame}
\frametitle{Bounded Discrete PB}
\begin{itemize}
\item Projects are either fully implemented or not at all
\item Degree of completion has a cap
\item Budget is defined as subset of projects which can be
implemented subject to budget constraints
\end{itemize}
\begin{exampleblock}{Example}
A project for building a new school.
\end{exampleblock}
\end{frame}
\begin{frame}
\frametitle{Unbounded Discrete PB}
\begin{itemize}
\item Multiple degrees of completion
\item Substages of projects (milestones) can be defined
\item Still bounded by total available budget
\end{itemize}
\begin{exampleblock}{Example}
A project for building public toilets. The degree of completion
is the number of toilets that have already been built.
\end{exampleblock}
\end{frame}
\section{Preference Modeling}
\begin{frame}
\frametitle{Preference Modeling}
Model preferences as a cardinal utility function or an ordinal
preference relation:
\begin{block}{Cardinal utility function}
Each resident $i$ has a cardinal utility function $u_i :
A\rightarrow \mathbb{R}$, where $A$ is the set of feasible
allocations.
\end{block}
\begin{block}{Ordinal preference relation}
$\succ_i$ over $A$
\end{block}
\begin{alertblock}{Problem}
This does not adequately reflect any structural properties of
residents' preferences.
\end{alertblock}
\end{frame}
\begin{frame}
\frametitle{Preference Modeling}
\begin{itemize}
\item Impose a structural assumption on the utility function:
\[ u_i : 2^P\rightarrow\mathbb{R} \]
and $u_i$ satisfies subadditivity or superadditivity.
\item Use spatial models where preferences are situated in a
metric space and the distance between them models a
resident's utility for another allocation.
\item Take preferences over projects and use a rule to extend
them to allocations.
\end{itemize}
\end{frame}
\begin{frame}
\frametitle{Cardinal extensions}
\begin{block}{Scalar separable utility function}
A resident $i$ derives utility $u_{i,p}\cdot f_p(x_p)$ from each
project. A resident's utility for an allocation $\vec{x}$ is
additive across projects:
\[ u_i(\vec{x}) = \sum_{p\in P}{u_{i,p}\cdot f_p(x_p)} \]
\end{block}
\begin{block}{Dichotomous preferences}
Assuming $x_p\in\{0,1\}$ and $u_{i,p}\in\{0,1\}$, residents
either approve or disapprove a project and care only about the
number of projects implemented.
\end{block}
\begin{block}{Max set extension}
Utility of an allocation is defined as the utility for a
resident's most favorite project: $u_i(S) = \mathsf{max}_{p\in
S}u_{i,p}$ for each $S\subseteq P$.
\end{block}
\end{frame}
\begin{frame}
\frametitle{Ordinal extensions}
\begin{block}{Stochastic dominance extension}
For two allocations $\vec{x},\vec{y}\in A$ and
$E_i^1,\dots,E_i^{k_i}$ being the equivalence classes of the
relation $\succ_i$ in decreasing order of preferences: \[
\vec{x}\succ_{i}^{SD}\vec{y} \text{ iff }
\sum_{j=1}^{l}{\sum_{p\in E_i^j}{x_p}}\geq\sum_{j=1}^{l}{\sum_{p\in
E_i^j}{y_p}}\quad\text{for all } l\in\{1,\dots,k_i\}\]
\end{block}
\begin{block}{Lexicographic extension $\succ_i^{lex}$}
A resident $i$ cares significantly more about project $p$ than
about $p'$ whenever $p\succ_i p'$.
\end{block}
\begin{block}{Scoring rules}
Convert ordinal to cardinal preferences by taking a ranking
$\succ_i$ over projects and determining the utility as $u_{i,p}
= s_k$ where $k$ is the rank in a scoring vector $\vec{s} =
(s_1,\dots,s_m)$ and $s_1\geq\dots\geq s_m\geq 0$.
\end{block}
\end{frame}
\section{Preference Elicitation}
\begin{frame}
\frametitle{Preference elicitation}
\begin{itemize}
\item Also known as \emph{Ballot Design}
\item Communicating full preferences over sometimes
exponentially many allocations is difficult
\item Cognitive burden can lead to lower turnout rates
\end{itemize}
\end{frame}
\begin{frame}
\frametitle{Preference elicitation}
\begin{block}{Range voting}
Voters rate projects based on their utility for each project.
\end{block}
\begin{block}{$k$-Approval}
Voters approve the $k$ projects they like the most.
\end{block}
\begin{block}{Approval voting}
Voters approve all projects that they like.
\end{block}
\begin{block}{Threshold approval voting}
Voters approve projects where their utility is above a specified
threshold.
\end{block}
\begin{block}{Knapsack voting}
Voters provide ideal allocation based on their preferences.
\end{block}
\end{frame}
\section{Vote Aggregation}
\begin{frame}
\frametitle{Vote Aggregation}
\begin{itemize}
\item Voters' preferences are aggregated to determine which
projects to fund
\item Main interest for research
\item Three different approaches:
\begin{itemize}
\item Welfare Maximization
\item Use of Axioms
\item Notions of Fairness
\end{itemize}
\end{itemize}
\end{frame}
\begin{frame}
\frametitle{Welfare Maximization}
\begin{block}{Utilitarian Welfare}
The utilitarian welfare of an allocation is the sum of utilities it gives to
residents:
\[ UW(\vec{x}) =
\sum_{i\in N}{u_i(\vec{x})}\text{ for }\vec{x}\in A \]
\end{block}
\begin{block}{Egalitarian Welfare}
The egalitarian welfare of an allocation is the minimum utility
it gives to any resident:
\[ EW(\vec{x}) =
\mathsf{min}_{i\in N}{u_i(\vec{x})}\text{ for
}\vec{x}\in A \]
\end{block}
\begin{block}{Nash Welfare}
The Nash welfare of an allocation is the product of utilities it gives to
residents:
\[ NW(\vec{x}) =
\prod_{i\in N}{u_i(\vec{x})}\text{ for }\vec{x}\in A \]
\end{block}
\end{frame}
\begin{frame}
\frametitle{Use of Axioms}
\begin{block}{Exhaustiveness}
A feasible allocation $\vec{x}$ is called exhaustive if an
outcome $\vec{y}$ is not feasible whenever $y_p\geq x_p$ for all
projects $p$ and a strict inequality holds for at least one
project.
\end{block}
\begin{block}{Discount Monotonicity}
Project $p$'s cost function $c_p$ is revised to $c_p'(x_p)\leq
c_p(x_p)$ after a vote aggregation rule outputs allocation
$\vec{x}$. For the resulting allocation $\vec{y}$, $y_p\geq
x_p$ holds.
\end{block}
\begin{block}{Pareto Optimality}
An allocation $\vec{x}\in A$ Pareto dominates another allocation
$\vec{y}\in A$ if $u_i(\vec{x})\geq u_i(\vec{y})$ for all $i\in
N$ and $u_i(\vec{x}) > u_i(\vec{y})$ for some $i\in N$. An
allocation $\vec{z}\in A$ is optimal if no allocation dominates
it.
\end{block}
\end{frame}
\begin{frame}
\frametitle{Notion of Fairness}
\begin{block}{The Core of PB}
An allocation $\vec{x} \in A$ is a core solution if there is no
subset $S$ of voters who, given a budget of $(|S|/n)B$, could
compute an allocation $\vec{y}\in A$ such that every voter in
$S$ receives strictly more utility in $\vec{y}$ than in
$\vec{x}$.
\end{block}
\begin{block}{Proportionality}
An allocation $\vec{x}$ should be proportionally reflected by
the division of voters. A majority of voters should have a
majority of the budget under their control but a minority should
have a minority of the budget under their control.
\end{block}
\end{frame}
\section{Future Directions}
\begin{frame}
\frametitle{Future Areas of Interest}
\begin{itemize}
\item Multi-dimensional constraints
\item Hybrid models
\item Complex resident preferences
\item Market-based approaches
\item The role of information
\item Research spanning the entire PB process
\end{itemize}
\end{frame}
\begin{frame}
\centering
\Large
Thank you for your attention! \\
Questions \& Answers
\end{frame}
\end{document}