\documentclass{article}
\usepackage{fleqn}
\usepackage{epsf}
\usepackage[dvips]{color}
\usepackage{aima2e-slides}
\begin{document}
\begin{huge}
\titleslide{Problem solving and search}{Chapter 3}
\sf
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Reminders}
\emph{Assignment 0 due 5pm today}
\emph{Assignment 1 posted}, due 2/9
\note{Section 105 will move to 9-10am} starting next week
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Outline}
\blob Problem-solving agents
\blob Problem types
\blob Problem formulation
\blob Example problems
\blob Basic search algorithms
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Problem-solving agents}
Restricted form of general agent:
\input{\file{algorithms}{simple-PS-agent-algorithm}}
Note: this is \note{offline} problem solving; solution executed ``eyes closed.''\\
\note{Online} problem solving involves acting without complete knowledge.
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: Romania}
On holiday in Romania; currently in Arad.\\
Flight leaves tomorrow from Bucharest
\note{Formulate goal}:\nl
be in Bucharest
\note{Formulate problem}:\nl
\note{states}: various cities\nl
\note{actions}: drive between cities
\note{Find solution}:\nl
sequence of cities, e.g., Arad, Sibiu, Fagaras, Bucharest
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: Romania}
\epsfxsize=1.08\maxfigwidth
\fig{\file{figures}{romania-distances.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Problem types}
\note{Deterministic, fully observable} $\Longrightarrow$ \defn{single-state problem}\nl
Agent knows exactly which state it will be in; solution is a sequence
\note{Non-observable} $\Longrightarrow$ \defn{conformant problem}\nl
Agent may have no idea where it is; solution (if any) is a sequence
\note{Nondeterministic} and/or \note{partially observable} $\Longrightarrow$ \defn{contingency problem}\nl
percepts provide \emph{new} information about current state\nl
solution is a \defn{contingent plan} or a \defn{policy} \nl
often \emph{interleave} search, execution
\note{Unknown state space} $\Longrightarrow$ \defn{exploration problem} (``online'')
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: vacuum world}
\begin{tabular}{lr}
\hbox{\begin{minipage}[t]{0.6\textwidth}
\note{Single-state}, start in \#5. \q{Solution}
\end{minipage}}
&
\hbox{\begin{minipage}[t]{0.4\maxfigwidth}
\epsfxsize=0.4\maxfigwidth
\raisebox{-0.35\maxfigwidth}[0pt][0pt]{\epsffile{\file{figures}{vacuum2-space.ps}}}
\end{minipage}}
\end{tabular}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: vacuum world}
\begin{tabular}{lr}
\hbox{\begin{minipage}[t]{0.6\textwidth}
\note{Single-state}, start in \#5. \q{Solution}
\\
\mat{$[Right,Suck]$}
\\[0.5\baselineskip]
\note{Conformant}, start in \mat{$\{1,2,3,4,5,6,7,8\}$}\\
e.g., \mat{$Right$} goes to \mat{$\{2,4,6,8\}$}. \q{Solution}
\end{minipage}}
&
\hbox{\begin{minipage}[t]{0.4\maxfigwidth}
\epsfxsize=0.4\maxfigwidth
\raisebox{-0.35\maxfigwidth}[0pt][0pt]{\epsffile{\file{figures}{vacuum2-space.ps}}}
\end{minipage}}
\end{tabular}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: vacuum world}
\begin{tabular}{lr}
\hbox{\begin{minipage}[t]{0.6\textwidth}
\note{Single-state}, start in \#5. \q{Solution}
\\
\mat{$[Right,Suck]$}
\\[0.5\baselineskip]
\note{Conformant}, start in \mat{$\{1,2,3,4,5,6,7,8\}$}\\
e.g., \mat{$Right$} goes to \mat{$\{2,4,6,8\}$}. \q{Solution}
\\
\mat{$[Right,Suck,Left,Suck]$}
\\[0.5\baselineskip]
\note{Contingency}, start in \#5\\
Murphy's Law: \mat{$Suck$} can dirty a clean carpet\\
Local sensing: dirt, location only.\\
\q{Solution}
\end{minipage}}
&
\hbox{\begin{minipage}[t]{0.4\maxfigwidth}
\epsfxsize=0.4\maxfigwidth
\raisebox{-0.35\maxfigwidth}[0pt][0pt]{\epsffile{\file{figures}{vacuum2-space.ps}}}
\end{minipage}}
\end{tabular}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: vacuum world}
\begin{tabular}{lr}
\hbox{\begin{minipage}[t]{0.6\textwidth}
\note{Single-state}, start in \#5. \q{Solution}
\\
\mat{$[Right,Suck]$}
\\[0.5\baselineskip]
\note{Conformant}, start in \mat{$\{1,2,3,4,5,6,7,8\}$}\\
e.g., \mat{$Right$} goes to \mat{$\{2,4,6,8\}$}. \q{Solution}
\\
\mat{$[Right,Suck,Left,Suck]$}
\\[0.5\baselineskip]
\note{Contingency}, start in \#5\\
Murphy's Law: \mat{$Suck$} can dirty a clean carpet\\
Local sensing: dirt, location only.\\
\q{Solution}
\\
\mat{$[Right,\k{if}\ dirt\ \k{then}\ Suck]$}
\end{minipage}}
&
\hbox{\begin{minipage}[t]{0.4\maxfigwidth}
\epsfxsize=0.4\maxfigwidth
\raisebox{-0.35\maxfigwidth}[0pt][0pt]{\epsffile{\file{figures}{vacuum2-space.ps}}}
\end{minipage}}
\end{tabular}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Single-state problem formulation}
A \defn{problem} is defined by four items:
\defn{initial state} \quad e.g., ``at Arad''
\defn{successor function} \mat{$S(x)$} = set of action--state pairs \nl
e.g., \mat{$S(Arad) = \{\, \ldots \}$}
\defn{goal test}, can be\nl
\note{explicit}, e.g., \mat{$x$} = ``at Bucharest''\nl
\note{implicit}, e.g., \mat{$NoDirt(x)$}
\defn{path cost} (additive)\nl
e.g., sum of distances, number of actions executed, etc.\nl
\mat{$c(x,a,y)$} is the \defn{step cost}, assumed to be \mat{$\geq 0$}
A \defn{solution} is a sequence of actions\\
leading from the initial state to a goal state
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Selecting a state space}
Real world is absurdly complex \nl
$\Rightarrow$ state space must be \emph{abstracted} for problem solving
(Abstract) state = set of real states
(Abstract) action = complex combination of real actions\nl
e.g., ``Arad $\rightarrow$ Zerind'' represents a complex set\nnl
of possible routes, detours, rest stops, etc. \\
For guaranteed realizability, \emph{any} real state ``in Arad''\al
must get to \note{some} real state ``in Zerind''
(Abstract) solution = \nl
set of real paths that are solutions in the real world
Each abstract action should be ``easier'' than the original problem!
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: vacuum world state space graph}
\epsfxsize=0.8\textwidth
\fig{\file{figures}{vacuum2-paths.ps}}
\q{states}\\
\q{actions}\\
\q{goal test}\\
\q{path cost}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: vacuum world state space graph}
\epsfxsize=0.8\textwidth
\fig{\file{figures}{vacuum2-paths.ps}}
\q{states}: integer dirt and robot locations (ignore dirt \note{amounts} etc.)\\
\q{actions}\\
\q{goal test}\\
\q{path cost}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: vacuum world state space graph}
\epsfxsize=0.8\textwidth
\fig{\file{figures}{vacuum2-paths.ps}}
\q{states}: integer dirt and robot locations (ignore dirt \note{amounts} etc.)\\
\q{actions}: \mat{$Left$}, \mat{$Right$}, \mat{$Suck$}, \mat{$NoOp$}\\
\q{goal test}\\
\q{path cost}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: vacuum world state space graph}
\epsfxsize=0.8\textwidth
\fig{\file{figures}{vacuum2-paths.ps}}
\q{states}: integer dirt and robot locations (ignore dirt \note{amounts} etc.)\\
\q{actions}: \mat{$Left$}, \mat{$Right$}, \mat{$Suck$}, \mat{$NoOp$}\\
\q{goal test}: no dirt\\
\q{path cost}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: vacuum world state space graph}
\epsfxsize=0.8\textwidth
\fig{\file{figures}{vacuum2-paths.ps}}
\q{states}: integer dirt and robot locations (ignore dirt \note{amounts} etc.)\\
\q{actions}: \mat{$Left$}, \mat{$Right$}, \mat{$Suck$}, \mat{$NoOp$}\\
\q{goal test}: no dirt\\
\q{path cost}: 1 per action (0 for \mat{$NoOp$})
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: The 8-puzzle}
\epsfxsize=0.6\textwidth
\fig{\file{figures}{8puzzle.ps}}
\q{states}\\
\q{actions}\\
\q{goal test}\\
\q{path cost}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: The 8-puzzle}
\epsfxsize=0.6\textwidth
\fig{\file{figures}{8puzzle.ps}}
\q{states}: integer locations of tiles (ignore intermediate positions)\\
\q{actions}\\
\q{goal test}\\
\q{path cost}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: The 8-puzzle}
\epsfxsize=0.6\textwidth
\fig{\file{figures}{8puzzle.ps}}
\q{states}: integer locations of tiles (ignore intermediate positions)\\
\q{actions}: move blank left, right, up, down (ignore unjamming etc.)\\
\q{goal test}\\
\q{path cost}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: The 8-puzzle}
\epsfxsize=0.6\textwidth
\fig{\file{figures}{8puzzle.ps}}
\q{states}: integer locations of tiles (ignore intermediate positions)\\
\q{actions}: move blank left, right, up, down (ignore unjamming etc.)\\
\q{goal test}: = goal state (given)\\
\q{path cost}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: The 8-puzzle}
\epsfxsize=0.6\textwidth
\fig{\file{figures}{8puzzle.ps}}
\q{states}: integer locations of tiles (ignore intermediate positions)\\
\q{actions}: move blank left, right, up, down (ignore unjamming etc.)\\
\q{goal test}: = goal state (given)\\
\q{path cost}: 1 per move
[Note: optimal solution of $n$-Puzzle family is NP-hard]
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Example: robotic assembly}
\epsfxsize=0.7\textwidth
\fig{\file{figures}{stanford-arm+blocks.ps}}
\q{states}: real-valued coordinates of robot joint angles\nl
parts of the object to be assembled
\q{actions}: continuous motions of robot joints
\q{goal test}: complete assembly \emph{with no robot included!}
\q{path cost}: time to execute
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Tree search algorithms}
Basic idea:\al
offline, simulated exploration of state space\al
by generating successors of already-explored states\nnl
(a.k.a.~\defn{expanding} states)
\input{\file{algorithms}{informal-ts-algorithm}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Tree search example}
\vspace*{0.3in}
\epsfxsize=1.0\textwidth
\fig{\sfile{figures}{search-map1.ps}}
%%%%%%%%%%%% Overlay %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\pheading{Tree search example}
\vspace*{0.3in}
\epsfxsize=1.0\textwidth
\fig{\sfile{figures}{search-map2.ps}}
%%%%%%%%%%%% Overlay %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\pheading{Tree search example}
\vspace*{0.3in}
\epsfxsize=1.0\textwidth
\fig{\sfile{figures}{search-map3.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Implementation: states vs.~nodes}
A \defn{state} is a (representation of) a physical configuration\\
A \defn{node} is a data structure constituting part of a search tree\nl
includes \note{parent}, \note{children}, \note{depth}, \note{path cost} \mat{$g(x)$}\\
States do not have parents, children, depth, or path cost!
\epsfxsize=0.65\textwidth
\fig{\file{figures}{state-vs-node.ps}}
The {\sc Expand} function creates new nodes, filling in the various
fields and using the {\sc SuccessorFn} of the
problem to create the corresponding states.
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Implementation: general tree search}
\input{\file{algorithms}{tree-search-algorithm}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Search strategies}
A strategy is defined by picking the \emph{order of node expansion}
Strategies are evaluated along the following dimensions:\nl
\defn{completeness}---does it always find a solution if one exists?\nl
\defn{time complexity}---number of nodes generated/expanded\nl
\defn{space complexity}---maximum number of nodes in memory\nl
\defn{optimality}---does it always find a least-cost solution?
Time and space complexity are measured in terms of\nl
\mat{$b$}---maximum branching factor of the search tree\nl
\mat{$d$}---depth of the least-cost solution\nl
\mat{$m$}---maximum depth of the state space (may be \mat{$\infty$})
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Uninformed search strategies}
\defn{Uninformed} strategies use only the information available\\
in the problem definition
Breadth-first search
Uniform-cost search
Depth-first search
Depth-limited search
Iterative deepening search
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Breadth-first search}
Expand shallowest unexpanded node
\emph{Implementation}:\nl
\v{fringe} is a FIFO queue, i.e., new successors go at end
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{bfs-progress1.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Breadth-first search}
Expand shallowest unexpanded node
\emph{Implementation}:\nl
\v{fringe} is a FIFO queue, i.e., new successors go at end
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{bfs-progress2.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Breadth-first search}
Expand shallowest unexpanded node
\emph{Implementation}:\nl
\v{fringe} is a FIFO queue, i.e., new successors go at end
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{bfs-progress3.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Breadth-first search}
Expand shallowest unexpanded node
\emph{Implementation}:\nl
\v{fringe} is a FIFO queue, i.e., new successors go at end
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{bfs-progress4.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of breadth-first search}
\q{Complete}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of breadth-first search}
\q{Complete} Yes (if \mat{$b$} is finite)
\q{Time}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of breadth-first search}
\q{Complete} Yes (if \mat{$b$} is finite)
\q{Time} \mat{$1+b+b^2+b^3+\ldots +b^d + b(b^d-1)= O(b^{d+1})$}, i.e., exp.~in \mat{$d$}
\q{Space}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of breadth-first search}
\q{Complete} Yes (if \mat{$b$} is finite)
\q{Time} \mat{$1+b+b^2+b^3+\ldots +b^d + b(b^d-1)= O(b^{d+1})$}, i.e., exp.~in \mat{$d$}
\q{Space} \mat{$O(b^{d+1})$} (keeps every node in memory)
\q{Optimal}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of breadth-first search}
\q{Complete} Yes (if \mat{$b$} is finite)
\q{Time} \mat{$1+b+b^2+b^3+\ldots +b^d + b(b^d-1)= O(b^{d+1})$}, i.e., exp.~in \mat{$d$}
\q{Space} \mat{$O(b^{d+1})$} (keeps every node in memory)
\q{Optimal} Yes (if cost = 1 per step); not optimal in general
\emph{Space} is the big problem; can easily generate nodes at 100MB/sec\nl
so 24hrs = 8640GB.
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Uniform-cost search}
Expand least-cost unexpanded node
\emph{Implementation}:\nl
\v{fringe} = queue ordered by path cost, lowest first
Equivalent to breadth-first if step costs all equal
\q{Complete} Yes, if step cost \mat{$\geq \epsilon$}
\q{Time} \# of nodes with \mat{$g \leq {}$} cost of optimal solution, \mat{$O(b^{\ceiling{C^*/\epsilon}})$}\al
where \mat{$C^*$} is the cost of the optimal solution
\q{Space} \# of nodes with \mat{$g \leq {}$} cost of optimal solution, \mat{$O(b^{\ceiling{C^*/\epsilon}})$}
\q{Optimal} Yes---nodes expanded in increasing order of \mat{$g(n)$}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Depth-first search}
Expand deepest unexpanded node
\emph{Implementation}:\nl
\v{fringe} = LIFO queue, i.e., put successors at front
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{dfs-progress01.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Depth-first search}
Expand deepest unexpanded node
\emph{Implementation}:\nl
\v{fringe} = LIFO queue, i.e., put successors at front
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{dfs-progress02.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Depth-first search}
Expand deepest unexpanded node
\emph{Implementation}:\nl
\v{fringe} = LIFO queue, i.e., put successors at front
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{dfs-progress03.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Depth-first search}
Expand deepest unexpanded node
\emph{Implementation}:\nl
\v{fringe} = LIFO queue, i.e., put successors at front
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{dfs-progress04.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Depth-first search}
Expand deepest unexpanded node
\emph{Implementation}:\nl
\v{fringe} = LIFO queue, i.e., put successors at front
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{dfs-progress05.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Depth-first search}
Expand deepest unexpanded node
\emph{Implementation}:\nl
\v{fringe} = LIFO queue, i.e., put successors at front
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{dfs-progress06.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Depth-first search}
Expand deepest unexpanded node
\emph{Implementation}:\nl
\v{fringe} = LIFO queue, i.e., put successors at front
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{dfs-progress07.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Depth-first search}
Expand deepest unexpanded node
\emph{Implementation}:\nl
\v{fringe} = LIFO queue, i.e., put successors at front
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{dfs-progress08.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Depth-first search}
Expand deepest unexpanded node
\emph{Implementation}:\nl
\v{fringe} = LIFO queue, i.e., put successors at front
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{dfs-progress09.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Depth-first search}
Expand deepest unexpanded node
\emph{Implementation}:\nl
\v{fringe} = LIFO queue, i.e., put successors at front
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{dfs-progress10.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Depth-first search}
Expand deepest unexpanded node
\emph{Implementation}:\nl
\v{fringe} = LIFO queue, i.e., put successors at front
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{dfs-progress11.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Depth-first search}
Expand deepest unexpanded node
\emph{Implementation}:\nl
\v{fringe} = LIFO queue, i.e., put successors at front
\epsfxsize=0.5\textwidth
\fig{\sfile{figures}{dfs-progress12.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of depth-first search}
\q{Complete}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of depth-first search}
\q{Complete} No: fails in infinite-depth spaces, spaces with loops\nl
Modify to avoid repeated states along path\nnl
$\Rightarrow$ complete in finite spaces
\q{Time}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of depth-first search}
\q{Complete} No: fails in infinite-depth spaces, spaces with loops\nl
Modify to avoid repeated states along path\nnl
$\Rightarrow$ complete in finite spaces
\q{Time} \mat{$O(b^m)$}: terrible if \mat{$m$} is much larger than \mat{$d$}\nl
but if solutions are dense, may be much faster than breadth-first
\q{Space}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of depth-first search}
\q{Complete} No: fails in infinite-depth spaces, spaces with loops\nl
Modify to avoid repeated states along path\nnl
$\Rightarrow$ complete in finite spaces
\q{Time} \mat{$O(b^m)$}: terrible if \mat{$m$} is much larger than \mat{$d$}\nl
but if solutions are dense, may be much faster than breadth-first
\q{Space} \mat{$O(bm)$}, i.e., linear space!
\q{Optimal}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of depth-first search}
\q{Complete} No: fails in infinite-depth spaces, spaces with loops\nl
Modify to avoid repeated states along path\nnl
$\Rightarrow$ complete in finite spaces
\q{Time} \mat{$O(b^m)$}: terrible if \mat{$m$} is much larger than \mat{$d$}\nl
but if solutions are dense, may be much faster than breadth-first
\q{Space} \mat{$O(bm)$}, i.e., linear space!
\q{Optimal} No
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Depth-limited search}
= depth-first search with depth limit \mat{$l$},\\
i.e., nodes at depth \mat{$l$} have no successors
\emph{Recursive implementation}:
\input{\file{algorithms}{recursive-dls-algorithm}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Iterative deepening search}
\input{\file{algorithms}{ids-algorithm}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Iterative deepening search \mat{$l=0$}}
\vspace*{0.3in}
\epsfxsize=1.1\textwidth
\fig{\sfile{figures}{ids-progress1.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Iterative deepening search \mat{$l=1$}}
\vspace*{0.3in}
\epsfxsize=1.1\textwidth
\fig{\sfile{figures}{ids-progress2.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Iterative deepening search \mat{$l=2$}}
\vspace*{0.3in}
\epsfxsize=1.1\textwidth
\fig{\sfile{figures}{ids-progress3.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Iterative deepening search \mat{$l=3$}}
\vspace*{0.3in}
\epsfxsize=1.1\textwidth
\fig{\sfile{figures}{ids-progress4.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of iterative deepening search}
\q{Complete}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of iterative deepening search}
\q{Complete} Yes
\q{Time}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of iterative deepening search}
\q{Complete} Yes
\q{Time} \mat{$(d+1)b^0 + d b^1 + (d-1)b^2 + \ldots + b^d = O(b^d)$}
\q{Space}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of iterative deepening search}
\q{Complete} Yes
\q{Time} \mat{$(d+1)b^0 + d b^1 + (d-1)b^2 + \ldots + b^d = O(b^d)$}
\q{Space} \mat{$O(bd)$}
\q{Optimal}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Properties of iterative deepening search}
\q{Complete} Yes
\q{Time} \mat{$(d+1)b^0 + d b^1 + (d-1)b^2 + \ldots + b^d = O(b^d)$}
\q{Space} \mat{$O(bd)$}
\q{Optimal} Yes, if step cost = 1\nl
Can be modified to explore uniform-cost tree
Numerical comparison for \mat{$b=10$} and \mat{$d=5$}, solution at far right leaf:
\mat{\begin{eqnarray*}
N(\mbox{IDS}) &=& 50 + 400 + 3,000 + 20,000 + 100,000 = 123,450 \\
N(\mbox{BFS}) &=& 10 + 100 + 1,000 + 10,000 + 100,000 + 999,990 = 1,111,100
\end{eqnarray*}}
IDS does better because other nodes at depth \mat{$d$} are not expanded
BFS can be modified to apply goal test when a node is \emph{generated}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Summary of algorithms}
\input{\file{tables}{search-summary-table}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Repeated states}
Failure to detect repeated states can turn a linear problem into an
exponential one!
\vspace*{0.3in}
\epsfxsize=0.7\maxfigwidth
\fig{\file{figures}{ribbon-space.ps}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Graph search}
\input{\file{algorithms}{graph-search-algorithm}}
%%%%%%%%%%%% Slide %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\heading{Summary}
Problem formulation usually requires abstracting away real-world details
to define a state space that can feasibly be explored
Variety of uninformed search strategies
Iterative deepening search uses only linear space\\
and not much more time than other uninformed algorithms
Graph search can be exponentially more efficient than tree search
\end{huge}
\end{document}