SEARCH METHODS IN AI

STATE SPACE SEARCH

 COMPLEX PROBLEMS & SOLUTIONS

Path Finding Chess Playing Robot Assembly

VLSI Chip Design Time-Table Scheduling Symbolic Integration 2

 AUTOMATED PROBLEM SOLVING BY SEARCH
• Generalized Techniques for Solving Large Classes of Complex Problems
• Problem Statement is the Input and solution is the Output, sometimes even the
problem specific algorithm or method could be the Output
• Problem Formulation by AI Search Methods consists of the following key concepts
– Configuration or State
– Constraints or Definitions of Valid Configurations
– Rules for Change of State and their Outcomes
– Initial or Start Configurations
– Goal Satisfying Configurations
– An Implicit State or Configuration Space
– Valid Solutions from Start to Goal in the State Space
– General Algorithms which SEARCH for Solutions in this State Space
• ISSUES
– Size of the Implicit Space, Capturing Domain Knowledge, Intelligent Algorithms that
work in reasonable time and Memory, Handling Incompleteness and Uncertainty
3
 BASICS OF STATE SPACE MODELLING
• STATE or CONFIGURATION:
– A set of variables which define a state or configuration
– Domains for every variable and constraints among variables to define a valid configuration
• STATE TRANSFORMATION RULES or MOVES:
– A set of RULES which define which are the valid set of NEXT STATE of a given State
– It also indicates who can make these Moves (OR Nodes, AND nodes, etc)
• STATE SPACE or IMPLICIT GRAPH
– The Complete Graph produced out of the State Transformation Rules.
– Typically too large to store. Could be Infinite.
• INITIAL or START STATE(s), GOAL STATE(s)
• SOLUTION(s), COSTS
– Depending on the problem formulation, it can be a PATH from Start to Goal or a Sub-graph of And-ed
Nodes
• SEARCH ALGORITHMS
– Intelligently explore the Implicit Graph or State Space by examining only a small sub-set to find the
solution
– To use Domain Knowledge or HEURISTICS to try and reach Goals faster 4
 TWO JUG PROBLEM
• There is a large bucket B full of water and Two (02) jugs, J1 of volume 3 litre and J2 of
volume 5 litre. You are allowed to fill up any empty jug from the bucket, pour all water back
to the bucket from a jug or pour from one jug to another. The goal is to have jug J1 with
exactly one (01) litre of water
• State Definition: <J1, J2> <0,0>
• Rules: <0,5>
– Fill (J1): <J1, J2> to <3,J2> <3,0> <3,5>
– Fill (J2): <J1, J2> to <J1, 5>
– Empty (J1), Empty (J2): Similarly defined <3,2>
<0,3>
– Pour (J1, J2): <J1, J2> to <X,Y>, where <2,0>
• X = 0 and Y = J1 + J2 if J1+J2 ≤ 5,
• Y = 5 and X = (J1+ J2) – 5, if J1+J2 > 5 <3,3> <0,2>
<2,5>
– Pour (J2, J2): Similarly defined
<3,4> <0,4>
• Start: <0,0>, Goal: <1,0> <1,5>
• Part of State Space Shown on the right
(Not all Links shown here) <1,0> <0,1> <3,1>

5
3 DISK, 3 PEG TOWER of HANOI STATE SPACE

6
 STATES, SPACES, SOLUTIONS, SEARCH
• States
– Full / Perfect Information and Partial Information States
• State Transformation Rules
– Deterministic Outcomes
– Non-Deterministic / Probabilistic Outcomes
• State Spaces As Generalized Games
– Single Player: OR Graphs
– Multi-Player: And / Or, Adversarial, Probabilistic Graphs
• Solutions
– Paths
– Sub-graphs
– Expected Outcomes
• Costs
• Sizes
• Domain Knowledge
• Algorithms for Heuristic Search
7
OR-Graph: TRAVELLING SALESPERSON PROBLEM
A
100
120

200 190
B C
75
80
110
75 90

D E
50

8
OR-Graph: TRAVELLING SALESPERSON PROBLEM
A <A>,0
100
120
<AB>,100 <AC>,120 <AD>,200 <AE>190
200 190
B C
75
80 <ABC>,175 <ABD>,175 <ABE>,180
110
75 90

<ABCD>,285 <ABCE>,265 <ABDC>,285 <ABDE>,225

D E
50

<ABCDE>,335 <ABCED>,315 <ABDCE>,375 <ABDEC>,315

<ABCDEA>,525 <ABCEDA>,515 <ABDCEA>,565 <ABDECA>,435

9
 MODELLING AND/OR GRAPHS:
OR Nodes are ones for which one has a choice.

The AND nodes could be compositional (sum, product, min, max,

etc., depending on the way the sub-problems are composed),

Adversarial (game where the other parties have a choice)

or
Probabilistic (Environmental Actions)

10
AND/OR GRAPHS: ADVERSARIAL

11
AND/OR GRAPHS: COMPOSITIONAL / ADVERSARIAL / PROBABILISTIC

12
COMPOSITIONAL AND/OR GRAPHS – MATRIX CHAIN MULTIPLICATION
(M1 X (M2 X (M3 X M4))) = ((M1 x M2) X (M3 X M4)) = (((M1 x M2) X M3) X M4) =
(M1 X (M2 X M3) ) X M4)
BUT THE NUMBER OF MULTIPLICATIONS TO GET THE ANSWER DIFFER !!

Let A be a [p by q] Matrix and B be a [q by r] Matrix. The number of multiplications needed to

compute A X B = p*q*r

Thus if M1 be a [10 by 30] Matrix, M2 be a [30 by 5] Matrix and M3 be a [5 by 60] Matrix

Then the number of computations for

(M1 X M2) X M3 = 10*30*5 for P = (M1 X M2) and 10*5*60 for P X M3. Total = 4500

M1 X (M2 X M3) = 30*5*60 for Q = (M2 X M3) and 10*30*60 for M1 X Q. Total = 27000
13
 COMPOSITIONAL AND/OR GRAPHS: MATRIX CHAIN
OR NODE (Min) MULTIPLICATION
(M1 :M4)
AND NODE (Fn)

(M1 :M4) (M1 :M4)

(M1 :M4)

(M1 :M1) (M2 :M4) (M1 :M2) (M3 :M4) (M1 :M3) (M4 :M4)

(M2 :M4) (M2 :M4) (M1 :M3) (M1 :M3)

(M2 :M2) (M3 :M4) (M2 :M3) (M1 :M2) (M3 :M3)

(M1 X (M2 X (M3 X M4))) = ((M1 x M2) X (M3 X M4)) = (((M1 x M2) X M3) X M4) =
(M1 X (M2 X M3) ) X M4)
14
 SEARCHING IMPLICIT GRAPHS
Given the start state the <A>,0
SEARCH Algorithm will
create successors based on
the State Transformation <AB>,100 <AC>,120 <AD>,200 <AE>190
Rules and make part of the
Graph EXPLICIT.
It will EXPAND the Explicit
graph INTELLGENTLY to <ABC>,175 <ABD>,175 <ABE>,180
rapidly search for a solution
without exploring the entire
Implicit Graph or State Space
For OR Graphs, the solution <ABCD>,285 <ABCE>,265 <ABDC>,285 <ABDE>,225
is a PATH from start to Goal.
Cost is usually sum of the
edge costs on the path, <ABCDE>,335 <ABCED>,315 <ABDCE>,375 <ABDEC>,315
though it could be something
based on the problem
<ABCDEA>,525 <ABCEDA>,515 <ABDCEA>,565 <ABDECA>,435
15
 SEARCHING IMPLICIT GRAPHS
For And/OR Graphs, the Solution
is an AND Subgraph rooted at
the Start and each leaf is a Goal
Node.
The Cost of OR Node is usually a
Min or Max.
The Cost at the AND Node
depends on the type of Node
(Compositional, Adversarial,
Probabilistic).
For Adversarial two player
games, Max / Min is used at AND
Node (reverse of Or Node)

16
 SEARCHING IMPLICIT GRAPHS
The various Search Algorithms include
• BASIC Algorithms: Depth-First (DFS), Breadth-first
(BFS), Iterative Deepening (IDS)
• COST-based Algorithms: Depth-First Branch-and-
Bound, Best First Search, Best-First Iterative Deepening
• Widely Used Algorithms: A* and IDA* (Or Graphs), AO*
(And/Or Graphs), Alpha-beta Pruning (Game-Trees)

17
 BASIC ALGORITHMS in OR GRAPHS: DFS
1. [Initialize] Initially the OPEN List contains the
Start Node s. CLOSED List is Empty.
2. [Select] Select the first Node n on the OPEN List.
If OPEN is empty, Terminate
3. [Goal Test] If n is Goal, then decide on
Termination or Continuation / Cost Updation
4. [Expand]
a) Generate the successors n_1, n_2, …. n_k, of
node n, based on the State Transformation
Rules
b) Put n in LIST CLOSED
c) For each n_i, not already in OPEN or CLOSED
List, put n_i in the FRONT of OPEN List
d) For each n_i already in OPEN or CLOSED
decide based on cost of the paths
5. [Continue] Go to Step 2 18
 BASIC ALGORITHMS in OR GRAPHS: IDS
1. [Initialize] Initially the OPEN List contains the Start
Node s. CLOSED List is Empty.
2. [Select] Select the first Node n on the OPEN List. If
OPEN is empty, Terminate
3. [Goal Test] If n is Goal, then decide on Termination or
Continuation / Cost Updation
4. [Expand]
a) Generate the successors n_1, n_2, …. n_k, of node
n, based on the State Transformation Rules
b) Put n in LIST CLOSED
c) For each n_i, not already in OPEN or CLOSED List,
put n_i in the FRONT of OPEN List
d) For each n_i already in OPEN or CLOSED decide
based on cost of the paths
3. [Continue] Go to Step 2

Algorithm IDS Performs DFS Level by Level Iteratively

(DFS (1), DFS (2), ……. and so on) 19
 BASIC ALGORITHMS in OR GRAPHS: BFS
1. [Initialize] Initially the OPEN List contains the
Start Node s. CLOSED List is Empty.
2. [Select] Select the first Node n on the OPEN List.
If OPEN is empty, Terminate
3. [Goal Test] If n is Goal, then decide on
Termination or Continuation / Cost Updation
4. [Expand]
a) Generate the successors n_1, n_2, …. n_k, of
node n, based on the State Transformation
Rules
b) Put n in LIST CLOSED
c) For each n_i, not already in OPEN or CLOSED
List, put n_i in the END of OPEN List
d) For each n_i already in OPEN or CLOSED
decide based on cost of the paths
5. [Continue] Go to Step 2 20
EXAMPLE: SEARCHING A STATE SPACE GRAPH
A START

• DEPTH-FIRST SEARCH (DFS)

B C D
• BREADTH-FIRST SEARCH (BFS)

• ITERATIVE DEEPENDING SEARCH (IDS)

E F G H
• PROPERTIES
GOAL • SOLUTION GUARANTEES
• MEMORY REQUIREMENTS
I J
GOAL
GOAL

21
EXAMPLE: SEARCHING A STATE SPACE GRAPH
A START DEPTH-FIRST SEARCH:
1. OPEN ={A}, CLOSED = {}
2. OPEN = {B,C,D}, CLOSED = {A}
3. OPEN = {E,F,G,C,D}, CLOSED = (A,B}
B C D 4. OPEN = {I,J,F,G,C,D}, CLOSED = {A,B,E}
5. Goal Node I Found. Can Terminate with
Path from A to I or may Continue for
more Goal nodes if minimum length or
E F G H cost is a criteria

GOAL DFS MAY NOT TERMINATE IF THERE IS AN

INFINITE DEPTH PATH EVEN IF THERE IS A
I J GOAL NODE AT FINITE DEPTH
GOAL
GOAL DFS HAS LOW MEMORY REQUIREMENT

22
EXAMPLE: SEARCHING A STATE SPACE GRAPH
A START ITERATIVE DEEPENING SEARCH:
1. PERFORM DFS TILL LENGTH 1. NO
SOLUTION FOUND
2. PERFORM DFS TILL LEVEL 2. GOAL
B C D NODE H REACHED.
3. Can Terminate with Path from A to H.
This is guaranteed to be the minimum
length path.
E F G H
IDS GUARANTEES SHORTEST LENGTH
GOAL PATH TO GOAL

I J IDS MAY RE-EXPAND NODES MANY TIMES

GOAL
GOAL IDS HAS LOWER MEMORY REQUIREMENT
THAN BFS
23
EXAMPLE: SEARCHING A STATE SPACE GRAPH
A START BREADTH-FIRST SEARCH:
1. OPEN ={A}, CLOSED = {}
2. OPEN = {B,C,D}, CLOSED = {A}
3. OPEN = {C,D,E,F,G}, CLOSED = (A,B}
B C D 4. OPEN = {D,E,F,G}, CLOSED = {A,B,C}
5. OPEN = {E,F,G,H}. CLOSED = {A,B,C,D}
6. OPEN = {F,G,H,I,J}, CLOSED = {A,B,C,D,E}
7. OPEN = {G,H,I,J}, CLOSED = {A,B,C,D,E,F}
E F G H 8. OPEN = {H,I,J}, CLOSED = {A,B,C,D,E,F,G}
9. Goal Node H Found. Can Terminate with
GOAL Path from A to H. This is guaranteed to be
the minimum length path.
I J
BFS GUARANTEES SHORTEST LENGTH PATH
GOAL
GOAL TO GOAL BUT HAS HIGHER MEMORY
REQUIREMENT
24
SEARCHING STATE SPACE GRAPHS WITH EDGE COSTS
A START

10
12 5
B C D
2
6 7 6
8 7 5 4

9 1 13
E F G H
14 3
18 GOAL
18
I J
GOA GOAL
L

25
MODIFYING BASIC ALGORITHMS TO INCORPORATE COSTS
1. [Initialize] Initially the OPEN List contains the Start Node s. CLOSED List is Empty.
2. [Select] Select the first Node n on the OPEN List. If OPEN is empty, Terminate
3. [Goal Test] If n is Goal, then decide on Termination or Continuation / Cost Updation
4. [Expand]
a) Generate the successors n_1, n_2, …. n_k, of node n, based on the State
Transformation Rules
b) Put n in LIST CLOSED
c) For each n_i, not already in OPEN or CLOSED List, put n_i in the FRONT (for
DFS) / END (for BFS) of OPEN List
d) For each n_i already in OPEN or CLOSED decide based on cost of the paths
5. [Continue] Go to Step 2

Algorithm IDS Performs DFS Level by Level Iteratively (DFS (1), DFS (2), ……. and so
on)
26
NEXT: SEARCHING STATE SPACE GRAPHS WITH EDGE COSTS
A START
• COST ORDERED SEARCH:
10 • DFBB
12
5 • Best First Search,
B C D • Best First IDS
2 • Use of HEURISTIC Estimates:
6 7 6
8 7 5 4 Algorithm A* (Or Graphs), AO*
1 13 (And/Or Graphs)
9 F G H
E
14 • PROPERTIES
18 3 GOAL • SOLUTION GUARANTEES
18
• MEMORY REQUIREMENTS
I J
GOA GOAL
L

27
Thank you