INTELLIGENT SYSTEMS (CSE-303-F)

Section A
UNINFORMED AND INFORMED SEARCH

LECTURE 1
Outline
 Motivation
 Technical Solution
 Uninformed Search
 Depth-First Search
 Breadth-First Search
 Informed Search
 Best-First Search
 Hill Climbing
 A*
 Illustration by a Larger Example
 Extensions
 Summary
Motivation
 One of the major goals of AI is to help humans in solving complex tasks
 How can I fill my container with pallets?
 Which is the shortest way from Milan to Innsbruck?
 Which is the fastest way from Milan to Innsbruck?
 How can I optimize the load of my freight to maximize my revenue?
 How can I solve my Sudoku game?
 What is the sequence of actions I should apply to win a game?
 Sometimes finding a solution is not enough, you want the optimal solution
according to some “cost” criteria
 All the example presented above involve looking for a plan
 A plan that can be defined as the set of operations to be performed of an
initial state, to reach a final state that is considered the goal state
 Thus we need efficient techniques to search for paths, or sequences of
actions, that can enable us to reach the goal state, i.e. to find a plan
 Such techniques are commonly called Search Methods
Examples of Problems: Towers of
Hanoi
 3 pegs A, B, C
1
 3 discs represented as natural 2
numbers (1, 2, 3) which 3
correspond to the size of the A B C
discs Operators:
 The three discs can be Move disk to peg
arbitrarily distributed over the
three pegs, such that the
following constraint holds: Applying: Move 1 to C (1 → C)
to the initial state ((123)()())
di is on top of dj → di < dj
a new state is reached
 Initial status: ((123)()())
((23)()(1))
 Goal status: (()()(123))
Cycles may appear in the
solution!
Examples of Problems:
Blocksworld
E
D C B
E A B C A D
Initial State Goal State

 Objects: blocks • Initial state:

 Attributes (1-ary – ontable(E), cleartop(E)
– ontable(A), cleartop(A)
relations): cleartop(x), – ontable(B), cleartop(B)
ontable(x) – ontable(C)
 Relations: on(x,y) – on(D,C), cleartop (D)

 Operators: puttable(x) • Applying the move put(E,A):

where x must be – on(E,A), cleartop(E)
cleartop; put(x,y), where – ontable(A)
– ontable(B), cleartop(B)
x and y must be – ontable(C)
cleartop – on(D,C), cleartop (D)
TECHNICAL SOLUTION
Search Methods
Search Space Representation
 Representing the search Node
Loop
space is the first step to
enable the problem resolution Arc
 Search space is mostly
represented through graphs
 A graph is a finite set of
nodes that are connected by
arcs
 A loop may exist in a graph,
where an arc lead back to the
original node
 In general, such a graph is
not explicitly given
 Search space is constructed
during search
Search Space Representation
undirect direct
 A graph is undirected if
arcs do not imply a
direction, direct otherwise

connected disconnected
 A graph is connected if
every pair of nodes is
connected by a path
tree
 A connected graph with no
loop is called tree

 A weighted graph, is a
graph for which a value is weighted
1
associated to each arc 1
4 5
6
2 1
2 1
Example: Towers of Hanoi*
These nodes
are equals
1
2
3
A B C

2 2
3 1 3 1
A B C A B C

… 2
3 1 2 3 2 1 3 1
A B C A B C A B C
…
1 1 1 1
3 2 3 2 3 2 3 2
A B C A B C A B C A B C

… … … … …
1 …
2 3
A B C

… …
* A partial tree search space representation
Example: Towers of Hanoi*

* A complete direct graph representation

[http://en.wikipedia.org/wiki/Tower_of_Hanoi]
Search Methods
 A search method is defined by picking the order of node
expansion

 Strategies are evaluated along the following dimensions:

 completeness: does it always find a solution if one exists?
 time complexity: number of nodes generated
 space complexity: maximum number of nodes in memory
 optimality: does it always find the shortest path solution?

 Time and space complexity are measured in terms of

 b: maximum branching factor of the search tree
 d: depth of the shortest path solution
 m: maximum depth of the state space (may be ∞)
Search Methods
 Uninformed techniques
 Systematicallysearch complete graph, unguided
 Also known as brute force, naïve, or blind

 Informed methods
 Use problem specific information to guide search
in promising directions
UNINFORMED SEARCH
Brute force approach to explore search
space
Uninformed Search
 A class of general purpose algorithms that operates in a brute force way
 The search space is explored without leveraging on any information on the
problem
 Also called blind search, or naïve search
 Since the methods are generic they are intrinsically inefficient

 E.g. Random Search

 This method selects randomly a new state from the current one
 If the goal state is reached, the search terminates
 Otherwise the methods randomly select an other operator to move to the next
state

 Prominent methods:
 Depth-First Search
 Breadth-First Search
 Uniform-Cost Search
Depth-First Search
 Depth-First Search (DFS) begins at the root node and exhaustively search each
branch to it maximum depth till a solution is found
 The successor node is selected going in depth using from right to left (w.r.t. graph
representing the search space)
 If greatest depth is reach with not solution, we backtrack till we find an unexplored
branch to follow

 DFS is not complete

 If cycles are presented in the graph, DFS will follow these cycles indefinitively
 If there are no cycles, the algorithm is complete
 Cycles effects can be limited by imposing a maximal depth of search (still the algorithm is
incomplete)
 DFS is not optimal
 The first solution is found and not the shortest path to a solution

 The algorithm can be implemented with a Last In First Out (LIFO) stack or recursion
 Depth-First Search: Algorithm
List open, closed, successors={};
Node root_node, current_node;
insert-first(root_node,open)

while not-empty(open);
current_node= remove-first(open);
insert-first (current_node,closed);
if (goal(current_node)) return current_node;
else
successors=successorsOf(current_node);
for(x in successors)
if(not-in(x,closed)) insert-first(x,open);
endIf
endWhile

N.B.= this version is not saving the path for simplicity

 Depth-First Search: Example
1 S

A B
2
3

S B F F A C D
5
4
6
open={S} closed ={}

S A F D 0. Visit S: open={A,B}, closed={S}

1.Visit A: open={S,B,F,B}, closed={A,S}
2.Visit S: open={B,F,B}, closed={S,A,S}
3.Visit B: open={S,A,F,D,F,B}, closed={B,S,A,S}
4.Visit S: open={A,F,D,F,B}, closed={S,B,S,A,S}
5.Visit A: open={F,D,F,B}, closed={A,S,B,S,A,S}
6.Visit F: GOAL Reached!
Depth-First Search: Example
S

A B

S B F F A C D

S A F D

Result is: S->A->B->F

Depth-First Search: Complexity
 Time Complexity
 assume (worst case) that there d=0
is 1 goal leaf at the RHS
 so DFS will expand all nodes d=1

=1 + b + b2+ ......... + bm m=d=2

G
= O (bm)
 where m is the max depth of the
tree d=0

d=1
 Space Complexity
 how many nodes can be in the
queue (worst-case)? d=2
 at each depth l < d we have b-1 d=3
nodes
 at depth m we have b nodes m=d=4
 total = (d-1)*(b-1) + b = O(bm)
Breadth-First Search
 Breadth-First Search (BFS) begins at the root
node and explore level-wise al the branches

 BFS is complete
 If there is a solution, BFS will found it
 BFS is optimal
 The solution found is guaranteed to be the shortest
path possible

 The algorithm can be implemented with a First In

First Out (FIFO) stack
Breadth-First Search: Algorithm
List open, closed, successors={};
Node root_node, current_node;
insert-last(root_node,open)

while not-empty(open);
current_node=remove-first(open);
insert-last(current_node,closed);
if (goal(current_node)) return current_node;
else
successors=successorsOf(current_node);
for(x in successors)
if(not-in(x,closed)) insert-last(x,open);
endIf
endWhile

N.B.= this version is not saving the path for simplicity

Breadth-First Search: Example
1 S

2
A B
3

4 5
S B F F A C D

open = {S}, closed={}

0. Visit S: open = {A,B}, closed={S}
S A F D 1. Visit A: open={B,S,B,F}, closed={S,A}
2. Visit B: open={S,B,F,F,A,C,D}, closed={S,A,B}
3. Visit S: open={B,F,F,A,C,D}, closed={S,A,B,S}
4. Visit B: open={F,F,A,C,D,S,A,C,D},
closed={S,A,B,S,B}
5. Visit F: Goal Found!
Breadth-First Search: Example
S

A B

S B F F A C D

S A F D

Result is: S->A->F

Breadth-First Search: Complexity

 Time complexity is the same magnitude as DFS

 O (bm)
 where m is the depth of the solution

 Space Complexity
 how many nodes can be in the queue (worst-case)?
 assume (worst case) that there
1 is 1 goal leaf at the d=0
RHS
d=1
 so BFS will store
2
all nodes 3

d=2
=1 +4 b + b2+ .........
5
+ bm
6 7
= O (bm) d=3

8 9 10 11 12 13 14 G 15 d=4
Further Uninformed Search
Strategies
 Depth-limited search (DLS): Impose a cut-off (e.g. n
for searching a path of length n-1), expand nodes with
max. depth first until cut-off depth is reached (LIFO
strategy, since it is a variation of depth-first search).
 Bidirectional search (BIDI): forward search from initial
state & backward search from goal state, stop when
the two searches meet. Average effort O(bd/2) if
testing whether the search fronts intersect has
constant effort
 In AI, the problem graph is typically not known. If the
graph is known, to find all optimal paths in a graph
with labelled arcs, standard graph algorithms can be
used
INFORMED SEARCH
Using knowledge on the search space to
reduce search costs
Informed Search
 Blind search methods take O(bm) in the worst case

 May make blind search algorithms prohibitively slow

where d is large

 How can we reduce the running time?

 Use problem-specific knowledge to pick which states are better
candidates
Informed Search
 Also called heuristic search

 In a heuristic search each state is assigned a “heuristic

value” (h-value) that the search uses in selecting the
“best” next step

 A heuristic is an operationally-effective nugget of

information on how to direct search in a problem space

 Heuristics are only approximately correct

 Informed Search: Prominent
methods
 Best-First Search

 A*

 Hill Climbing
Cost and Cost Estimation

f(n)=g(n)+h(n)
 g(n) the cost (so far) to reach the node n
 h(n) estimated cost to get from the node to the
goal
 f(n) estimated total cost of path through n to
goal
Informed Search: Best-First Search
 Special case of breadth-first search
 Uses h(n) = heuristic function as its evaluation function
 Ignores cost so far to get to that node (g(n))
 Expand the node that appears closest to goal

 Best First Search is complete

 Best First Search is not optimal
 A solution can be found in a longer path (higher h(n) with a lower
g(n) value)

 Special cases:
 uniform cost search: f(n) = g(n) = path to n
 A* search

31
Best-First Search: Algorithm
List open, closed, successors={};
Node root_node, current_node;
insert-last(root_node,open)

while not-empty(open); returns the list of direct

current_node=remove-first (open); descendants of the
insert-last(current_node,closed); current node in shortest
if (goal(current_node)) return current_node; cost order
else
successors=estimationOrderedSuccessorsOf(current_node);
for(x in successors)
if(not-in(x,closed)) insert-last(x,open);
endIf
endWhile

N.B.= this version is not saving the path for simplicity

32
 Best-First Search: Example
1 S

h=1 h=1
2
A B
h=2
3 h=2
h=2 h=2 h=2 h=2 h=2
S B F F A C D

open = {S}, closed={}

0. Visit S: open = {A,B}, closed={S}
S A F D 1. Visit A: open={B,F,B,S}, closed={S,A}
2. Visit B: open={F,B,S,F,A,C,D}, closed={S,A,B}
3. Visit F: Goal Found!

In this case we estimate the cost as the distance from the root node (in term of nodes)

33
Best-First Search: Example
S

h=1, w=2 h=1, w=1

A B
h=2, w=7 h=2, w=4
h=2 h=2 h=2 h=2 h=2
S B F F A C D

S A F D

Result is: S->A->F!

If we consider real costs, optimal solution is:

S->B->F

34
A*
 Derived from Best-First Search
 Uses both g(n) and h(n)
 A* is optimal
 A* is complete
A* : Algorithm
List open, closed, successors={};
Node root_node, current_node, goal;
insert-back(root_node,open)
returns the list of direct
while not-empty(open); descendants of the
current_node=remove-front(open); current node in shortest
insert-back(current_node,closed); total estimation order
if (current_node==goal) return current_node;
else
successors=totalEstOrderedSuccessorsOf(current_node);
for(x in successors)
if(not-in(x,closed)) insert-back(x,open);
endIf
endWhile

N.B.= this version is not saving the path for simplicity

36
 A* : Example
S 1

h=1, w=2, g=2 h=1, w=1, g=1

2
A B
h=2, h=2, w=7, g=9 h=2, w=4, g=5
w=2, h=2, 3
h=2 h=2, w=4, g=5
g=4 w=1
w=3
g=3 h=2,
S B F F 5
A g=4
4
C w=1 D
g=2

open = {S}, closed={}

0. Visit S: open = {B,A}, closed={S}
S A F D 1. Visit B: open={A,C,A,F,D}, closed={S,B}
2. Visit A: open={C,A,F,D,B,S,F}, closed={S,B,A}
3. Visit C: open={A,F,D,B,S,F}, closed={S,B,A,C}
4. Visit A: open={F,D,B,S,F}, closed={S,B,A,C,A}
5. Visit F: Goal Found!

In this case we estimate the cost as the distance from the root node (in term of nodes)

37
 A* : Example
S
h=1, w=2, g=2 h=1, w=1, g=1

A B
h=2, h=2, w=7, g=9 h=2, w=4, g=5
w=2, h=2,
h=2 h=2, w=4, g=5
g=4 w=1
w=3
g=3 h=2,
S B F F A g=4 C w=1 D
g=2

S A F D

Result is: S->B->F!

38
Hill Climbing
 Special case of depth-first search
 Uses h(n) = heuristic function as its evaluation
function
 Ignores cost so far to get to that node (g(n))
 Expand the node that appears closest to goal

 Hill Climbing is not complete

 Unless we introduce backtracking
 Hill Climbing is not optimal
 Solution found is a local optimum
 Hill Climbing: Algorithm
List successors={}; Node root_node, current_node, nextNode;

current_node=root_node
while (current_node!=null)
if (goal(current_node)) return current_node;
else
successors=successorsOf(current_node);
nextEval = -∞; nextNode=null;
for(x in successors)
if(eval(x)> nextEval)
nexEval=eval(x);
nextNode=x;
current_node=nextNode,
endIf
endWhile
N.B.= this version is not using backtracking
40
 Hill Climbing: Example
1 S

h=1 h=1
A B
h=2
2
h=2
h=3 h=2 h=2 h=2 h=2
S B F F A C D

3 h=1 0. current_node=S, successors (A=1,B=1)

1. current_node=A, successors (B=2,F=2,S=2)
S A F D 2. current_node=B, successors (F=1,….)
3. current_node=F: Goal Found!

In this case we estimate the cost as the distance from the root node (in term of nodes)

41
Hill Climbing: Example
S

h=1 h=1
A B
h=2
h=2
h=3 h=2 h=2 h=2 h=2
S B F F A C D

h=1

S A F D

Result is: S->A->B->F!

Not optimal, more if at step 1 h(S)=2 we would

have completed without funding a solution

42
 Informed Search Algorithm
Comparison
Algorithm Time Space Optimal Complete Derivative
Best First O(bm) O(bm) No Yes BFS
Search
Hill Climbing O() O(b) No No
A* O(2N) O(bd) Yes Yes Best First
Search

b, branching factor
d, tree depth of the solution
m, maximum tree depth

43
ILLUSTRATION BY A LARGER
EXAMPLE
Route Search
 Start point:
Milan
 End point:
Innsbruck
 Search space:
Cities
 Nodes: Cities
 Arcs: Roads

 Let’s find a
possible route!
 Graph Representation
 We start from the
Feldkirck root node, and
Landeck Innsbruck (Goal) pick the leaves

 The same apply

Merano to each leaves
Bolzano
 But we do not
reconsider
Chiavenna
already used
Sondrio arcs
Lugano Trento
20
Lecco
Como
Bergamo
 The first node
picked is the first
Milan (Root) node on the right
Brescia Verona

Piacenza
 Depth-First Search
Milan

Piacenza Bergamo Lecco Como

2
Brescia Brescia

Verona Trento Verona Trento

3
Trento Bolzano Trento Bolzano
4
Bolzano Bolzano
Merano Merano
5 Innsbruck Innsbruck
Merano Merano
Innsbruck Landeck Innsbruck Landeck According to Google Maps:
464 km – 4 hours 37 mins
Landeck Landeck

Innsbruck Innsbruck

Innsbruck Innsbruck
N.B.: by building the tree, we are exploring the search space!
 Breadth-First search
Milan

1 4
2
3

Piacenza Bergamo Lecco Como

5 6 7 8 9 10
Brescia Brescia Chiavenna Sondrio Lugano Lecco
12
13
11 14 15 17 18 19 20 21
16
Verona Trento Verona Trento Landeck Lugano Merano Chi. Feld. Chi. Sondrio
22 23 24 25
Trento Bolzano Trento Bolzano 26
Bolzano Bolzano Innsbruck

Innsbruck Innsbruck

Merano Merano
Innsbruck Innsbruck According to Google Maps:
358 km – 5 hours 18 mins
Landeck Landeck

Innsbruck Innsbruck
N.B.: by building the tree, we are exploring the search space!
Depth-First Search vs Breadth-
First search
 Distance
 DFS: 464 km
 BFS: 358 km
 Q1: Can we use an algorithm to optimize according to distance?
 Time
 DFS: 4 hours 37 mins
 BFS: 5 hours 18 mins
 Q2: Can we use an algorithm to optimize according to time?
 Search space:
 DFS: 5 expansions
 BFS: 26 expansions
 Not very relevant… depends a lot on how you pick the order of node expansion, never the
less BFS is usually more expensive

 To solve Q1 and Q2 we can apply for example and Best-First Search

 Q1: the heuristic maybe the air distance between cities
 Q2: the heuristic maybe the air distance between cities x average speed (e.g. 90km/h)
 Graph Representation
with approximate distance
Feldkirck
70 Landeck Innsbruck (Goal)
63

100
90
180 140 Merano
25
Bolzano
135
Chiavenna
55
40 Sondrio
Lugano 60 Trento
42
20
25 Lecco
Como 90
50 Bergamo 80
45
45
55
Milan (Root) 60
Brescia Verona
60 70

Piacenza
 Best-First search
Milan

Piacenza 4 H=60 Bergamo 3 H=55 Lecco 2 H=50 Como 1 H=45

H=110 H=92
Brescia 11 H=130 Brescia 8 H=100 Chiavenna 10 Sondrio 7 Lugano 5 H=65 Lecco 6 H=70
H=190 H=220 H=160 H=190 H=150 H=227 H=105 H=245 H=130 H=142
H=250
Verona 16 Trento 18 Verona 15 Trento 17 Landeck 22 Lugano 14 Merano 19 Chi. 9 Feld. 21 Chi. 12 Sondrio 13
H=270 H=275 H=240 H=245
Trento Bolzano Trento 20 Bolzano 21 H=313
29
Bolzano Bolzano Innsbruck

Innsbruck Innsbruck

Merano Merano
Innsbruck Innsbruck According to Google Maps:
358 km – 5 hours 18 mins
Landeck
Landeck And this is really the shortest way!

Innsbruck Innsbruck
N.B.: by building the tree, we are exploring the search space!
EXTENSIONS
Variants to presented algorithms
 Combine Depth First Search and Breadth First Search, by
performing Depth Limited Search with increased depths until a goal
is found
 Enrich Hill Climbing with random restart to hinder the local
maximum and foothill problems
 Stochastic Beam Search: select w nodes randomly; nodes with
higher values have a higher probability of selection
 Genetic Algorithms: generate nodes like in stochastic beam search,
but from two parents rather than from one
SUMMARY
Summary
 Uninformed Search
 If the branching factor is small, BFS is the best
solution
 If the tree is depth IDS is a good choice

 Informed Search
 Heuristic function selection determines the efficiency
of the algorithm
 If actual cost is very expensive to be computed, then
Best First Search is a good solution
 Hill climbing tends to stack in local optimal solutions