Artificial Intelligence and Expert Systems

Lecture 2 Uninformed Search

October 5, 2024
Table of Contents

1. Classical Search Problems

2. Search Strategies
2.1 Breadth-First Search (BFS)
2.2 Depth First Search (DFS)
2.3 Uniform-Cost Search
2.4 Iterative Deepening Search

3. Conclusion

2/40
Classical Search Problems
Classical Search Problems

• Route Planning: Finding the shortest path between two points.

• Puzzle Solving: E.g., 8-puzzle, where tiles must be moved to reach a goal
configuration.
• Game Playing: Determining optimal moves in a game like chess or checkers.

3/40
Outline

• Problem-solving agents
• Problem types
• Problem formulation
• Example problems
• Basic search algorithms

4/40
Overview

• Reflex agents cannot operate well in complex environments

• Problem-solving agents use atomic representations
• Two general classes of search:
• Uninformed search algorithms
• Informed search algorithms
• Goals help organize behavior by limiting objectives
• We consider environments that are known, observable, discrete, and
deterministic
• Search: process of looking for a sequence of actions that reaches the goal

5/40
Well-defined Problems and Solutions

Key Components of a Problem:

1. Initial State: Start state, e.g., In(Arad).
2. Actions: Actions from a state, e.g., {Go(Sibiu)}.
3. Transition Model: Returns new state from action.
4. Goal Test: Checks if state is a goal, e.g., In(Bucharest).
5. Path Cost: Sum of action costs.

6/40
Problem-solving agents

• Goal formulation and problem setting.

• Execute action sequences toward solving the problem.
• Continue until the goal is achieved.
Problem Formulation:
• Abstraction: Ignores irrelevant details for simplicity.
• State Description: Focuses on crucial elements.
• Actions: Considers only key actions.

7/40
Example: Romania

• Goal: Be in Bucharest
• Problem formulation:
• States: Various cities
• Actions: Drive between cities
• Solution: Sequence of cities (e.g., Arad,
Sibiu, Fagaras, Bucharest)
Figure: Simplified map of Romania image
source Russell and Norvig (2010)

8/40
Problem types

• Deterministic, fully observable → single-state problem

• Non-observable → sensorless problem (conformant problem)
• Nondeterministic and/or partially observable → contingency problem
• Unknown state space → exploration problem

9/40
Selecting a state space

• The Real world is complex, state space must be abstracted

• Abstract state = set of real states
• Abstract action = complex combination of real actions
• Abstract solution = set of real paths that are solutions in the real world

10/40
8-Puzzle Problem

States: Each state represents a unique

configuration of the 8 tiles on a 3x3 board.
Initial State: Any configuration of tiles. Ex-
ample: the configuration shown above.
Actions: Slide a tile horizontally or verti-
cally into the empty space.
Transition Model: Given a state and ac-
tion, returns a new state with the tile
moved into the empty space.
Goal Test: A state where the tiles are ar- Figure: 8 Puzzle configuration Russell and
ranged in a numerical order with the empty Norvig (2010)
space at the bottom right corner.
Path Cost: Each move costs one step. The
path cost is the number of moves taken. 11/40
8 Queens Problem

States: Each state represents a unique ar-

rangement of up to 8 queens on an 8x8
chessboard.
Initial State: Any configuration of queens,
from 0 to 8, on the board.
Actions: Add a queen to any empty square
such that no two queens attack each
other.
Transition Model: Given a state and ac-
tion, returns a new state with the queen
added to the specified square.
Goal Test: A state with 8 queens on the
Figure: 8 Queens on a chessboard Russell and
board with none attacking each other.
Norvig (2010)
Path Cost: Typically 1 per move. However, 12/40
 Searching for Solutions

Understanding Search:
• Solutions are sequences of actions
considered by search algorithms.
• These sequences form a search tree,
starting from the initial state as the
root.
• Expanding a state generates new
states, adding branches to the tree.
• We add child nodes to parent nodes
through these expansions.

Figure: Illustration of a search tree

13/40
 Searching for Solutions - Cont’d

Expanding the Frontier:

• Nodes without children are leaves; all
leaves form the frontier.
• Expansion continues until a solution
is found or all states are explored.
• We focus next on the general
TREE-SEARCH algorithm.
• Search algorithms differ by their
state expansion strategy.
• TREE-SEARCH explores all paths;
GRAPH-SEARCH avoids redundant
paths. Figure: Illustration of a search tree
14/40
Searching for Solutions

• Redundant paths make problems intractable.

• Use explored set in TREE-SEARCH.
• Explored set prevents redundant exploration.

15/40
Infrastructure and Implementation

• Search requires structured data.

• Each node has four components.
• States differ from nodes.
• Nodes stored in specific queues.
• Queue types affect node retrieval.
Node structure:
• n.STATE: state in the state space
• n.PARENT: node in the search tree that generated this node
• n.ACTION: action applied to the parent
• n.PATH-COST: cost of the path from initial state to the node

16/40
States vs. Nodes

State:
• A representation of a physical configuration
• Part of the problem domain
• Independent of the search process

Node:
• A data structure used in the search algorithm
• Contains a state plus other information:
• Parent node
• Action taken to reach this node Figure: Node and State Image source Russell and
• Path cost (g(x)) Norvig (2010)
• Depth in the search tree
• Created by the Expand function during search

Key Difference: Multiple nodes can contain the same state but represent different paths to reach that state.
17/40
Search strategies

Evaluation criteria:
• Completeness
• Time complexity
• Space complexity
• Optimality
Measures:
• b: maximum branching factor
• d: depth of least-cost solution
• m: maximum depth of state space

18/40
Uninformed search strategies

• Breadth-first search
• Uniform-cost search
• Depth-first search
• Depth-limited search
• Iterative deepening search

19/40
Search Strategies
Breadth First Search

Breadth-First Search (BFS) is a layer-by-layer traversal method that explores nodes

in the order of their distances from the starting point. It uses a queue to keep track
of the next node to visit.

20/40
Breadth First Search BFS

Figure: Breadth First Search Image Source Russell and Norvig (2010)
21/40
Breadth-First Search (BFS)

S Queue Operations (FIFO):

Node Frontier
S A, B, D
A B, D, E, F
A B D B D, E, F, G, C
D E, F, G, C, K, M
E F, G, C, K, M, H, I
F G, C, K, M, H, I
E F G C K G M C, K, M, H, I, J
C K, M, H, I, J
K M, H, I, J
M H, I, J, N, L
H I J NH L I, J, N, L
I J, N, L
J N, L
N Goal Reached!
22/40
Breadth-First Search (BFS)

S Queue Operations (FIFO):

Node Frontier
S A, B, D
A B, D, E, F
A B D B D, E, F, G, C
D E, F, G, C, K, M
E F, G, C, K, M, H, I
F G, C, K, M, H, I
E F G C K G M C, K, M, H, I, J
C K, M, H, I, J
K M, H, I, J
M H, I, J, N, L
H I J NH L I, J, N, L
I Goal Reached!

23/40
Complexity and Properties

• Time Complexity: O(b{d} ), where b is the branching factor and d is the depth of
the shallowest solution.
• Space Complexity: O(b{d} ), as all nodes at each depth d are stored in the queue.
• Completeness: BFS is complete, meaning it will find a solution if one exists.
• Optimality: BFS is optimal for uniform cost paths.

24/40
Depth-First Search (DFS)

Algorithm:

1. Initialize an empty stack

2. Push the starting node onto the stack
3. While the stack is not empty:
a. Pop the top node
b. If it is the goal, return success
c. Otherwise, push all adjacent nodes onto the stack

Example: Generate all possible configurations of a puzzle game.

25/40
Depth-First Search (DFS) Visualization - Path to
Goal State N

Node Frontier
S
S D, B, A
A D, B, F, E
E D, B, F, I, H
H D, B, F, I
A B D
I D, B, F
F D, B
B D, C, G
G D, C, J
E F G C K M
J D, C
C D
D M, K
K M
H I J N L
M L, N
N “Reached Goal State” L26/40
How DFS Solves the Maze to Reach N

DFS explores nodes by expanding the deepest node in the current frontier until it
reaches the goal state N or can expand no further.
Key Steps:
• Start at S, push all children onto the stack starting from right to left.
• Continue deep-first exploration.
• M is the parent of N, discovered after popping K.
• N is found and identified as the goal state.
• Path to N: S → A → E → H → I → F → B → G → J → C → D → K → M → N (highlighted in yellow).
DFS is used when space is a concern, and it can be faster than BFS by finding a
solution without exploring all level nodes.

27/40
Uniform Cost Search

28/40
Figure: Uniform Cost Search Image Source Russell and Norvig (2010)
Uniform Cost Search (UCS) Visualization

UCS Path Finding Details:

S Frontier List Expand List Explored List
{(S, 0)}
3 2 1 {(D, 1), (B, 2), (A, 3)} S {S}
{(B, 2), (A, 3), (M, 3), (K, 5)} D {S, D}
{(A, 3), (M, 3), (G, 5), (K, 5), (C, 6)} B {S, D, B}
{(M, 3), (G, 5), (K, 5), (C, 6), (A, 7)} M {S, D, B, M}

A B D
{(G, 5), (K, 5), (C, 6), (A, 7), (N, 4), (L, 6)}
{(G, 5), (K, 5), (C, 6), (A, 7), (L, 6)}
N
G
{S, D, B, M, N}
{S, D, B, M, N, G}

4 2 3 4 4 2

E F G C K M

1 4 2 1 3

H I J N L

29/40
How UCS Solves the Map to Reach N

UCS explores paths in order of increasing cost:

• Start at node S.
• Continue by expanding the least costly node.
• Nodes are expanded until the goal node N is reached.
• Path to N: S → D → M → N.
• Total Cost: 4
UCS guarantees the lowest cost path in weighted graphs.

30/40
Uniform Cost Search (UCS) Visualization - Path to
Goal State I

UCS Path Finding Details:

S Frontier List Expand List Explored List
{(S, 0)}

3 2 1 {(D, 1), (B, 2), (A, 3)}

{(B, 2), (A, 3), (M, 3), (K, 5)}
S
D
{S}
{S, D}
{(A, 3), (M, 3), (G, 5), (K, 5), (C, 6)} B {S, D, B}
{(M, 3), (G, 5), (K, 5), (C, 6), (A, 7)} M {S, D, B, M}
{(G, 5), (K, 5), (C, 6), (A, 7), (N, 4), (L, 6)} N {S, D, B, M, N}

A B D {(G, 5), (K, 5), (C, 6), (A, 7), (L, 6)}
{(K, 5), (C, 6), (A, 7), (L, 6), (F, 8)}
G
G
{S, D, B, M, N, G}
{S, D, B, M, N, G, C}
{(K, 5), (L, 6), (A, 7), (F, 8), (E, 10)} K {S, D, B, M, N, G, C, K}
{(L, 6), (A, 7), (F, 8), (E, 10)} A {S, D, B, M, N, G, C, K, A}

4 2 3 4 4 2 {(L, 6), (F, 8), (E, 10), (H, 11), (I, 14)}
{(L, 6), (F, 8), (H, 11), (I, 14)}
E
I
{S, D, B, M, N, G, C, K, A, E}
{S, D, B, M, N, G, C, K, A, E, I}

E F G C K M

1 4 2 1 3

H I J N L
31/40
Iterative Deepening Search

IDS Process Explained:

• Starts at node S and iteratively deepens the search.
• Explores all nodes within the current depth limit before increasing the limit.
• Efficiently combines the space efficiency of DFS and the depth control of BFS.
• Ensures that no nodes are missed and finds the shortest path in terms of depth.
Benefits:
• IDS finds the shortest path with minimal memory usage compared to BFS.
• Effective in large search spaces where the depth of the solution is not known in
advance.

32/40
IDS Algorithm and Its Relation to DFS and BFS

Algorithm Details:
• Combines DFS depth-first search with BFS layer-by-layer expansion.
• DFS is applied with increasing depth limits, simulating BFS.
• Explores fully at each depth before proceeding deeper.
Execution Steps:
1. Begin with depth 0, perform DFS (only depth 0 is explored).
2. If the goal isn’t found, increase depth to 1.
3. Repeat DFS for depth 1 and higher until the goal is found.
4. This method simulates BFS by exploring all nodes at each depth before moving deeper.
Key Benefits:
• Uses minimal memory (like DFS), storing only the current path.
• Ensures completeness (like BFS) by exploring all nodes at each level. 33/40
Iterative Deepening Search

Figure: Iterative Deepening Search Image source Russell and Norvig (2010)

34/40
 Iterative Deepening Depth-First Search (IDDFS) -
Path to Goal State N

D0 S
IDDFS Process:
Depth Expanded Frontier
0 S S
D1 A B D 1 S A, B, D
2 A E, F, B, D
B E, F, G, C, D
D2 E F G C K M D E, F, G, C, K, M
3 E H, I, F, G, C, K, M
F H, I, G, C, K, M
D 3H I J N L G H, I, J, C, K, M
C H, I, J, K, M
K H, I, J, M
D4 M H, I, J, N, L
4 N H, I, J, L
Path to N: S → D → M → N
35/40
Iterative Deepening Depth-First Search (IDDFS) -
Path to Goal State I

D0 S
IDDFS Process:
Depth Expanded Frontier
0 S S
D1 A B D 1 S A, B, D
2 A E, F, B, D
B E, F, G, C, D
D2 E G
F C
K M D E, F, G, C, K, M
3 E H, I, F, G, C, K, M
F H, I, G, C, K, M
HD 3 I J N L G H, I, J, C, K, M
I H, I, J, C, K, M

D4
Path to I: S → A → E → I

36/40
Comparing uninformed Search Strategies

Figure: Comparing uninformed search methods Image Source Russell and Norvig (2010)

37/40
Conclusion
Conclusion

• Recap the significance and applications of classical search.

• Encourage further exploration and optimization of these algorithms.
• Chapter 3 In AI modern Approach for reference.

38/40
Reference

Some slides and ideas are adapted from the following sources:
• Artificial Intelligence: A Modern Approach Russell and Norvig (2010)

39/40
Thank you for your attention

October 5, 2024
Russell, S. and Norvig, P. (2010). Artificial Intelligence: A Modern Approach. Prentice
Hall, 3 edition.

40/40