Overview of Last Topics

• Intelligent Agents
• Agents and Environments
• Rational Agent
• Characteristics of Rational Agent
• Structure of Agents

1/24/2024
 Artificial Intelligence
Chapter 3: Solving Problems
by Searching

Teshale G. (MSc)
Email: [email protected]
Department of Computer Science
Madda Walabu University

1/24/2024 Chapter 3: Solving Problems by Searching 2

 Overview
 Problem-Solving Agents
 Example Problems
 Searching for Solutions
 Uninformed Search Strategies
 Avoiding Repeated States

1/24/2024 Chapter 3: Solving Problems by Searching 3

Problem-Solving Agents
 Reach goals through sequences of actions
 Formulate the goal(s) and the problem
 Abstraction: Should be easier than original problem
 Search for a sequence of actions to reach a goal
state
 A solution is a sequence of actions from initial state to
a goal state
 Execute the sequence of actions

1/24/2024 Chapter 3: Solving Problems by Searching 4

Problem-Solving Agents

1/24/2024 Chapter 3: Solving Problems by Searching 5

Formulating the Problem
 Initial state = starting state for the agent
 Given a state and a set of actions, the
successor function gives the possible next
states
 Initial state + successor function yields the
state space
 State space = the set of all states reachable
from the initial state

1/24/2024 Chapter 3: Solving Problems by Searching 6

Formulating the Problem,
Continued
 The goal test determines if a state is a
goal state
 A path is a particular sequence of states,
connected by particular actions
 The path cost function assigns a cost to
each path
 Optimal solution: a solution with minimal path
cost

1/24/2024 Chapter 3: Solving Problems by Searching 7

Example Problems
 “Toy” Problems: Simple, but useful for
theory and analysis
 Vacuum world, Sliding-block puzzles, N-
Queens
 Real Problems: The solutions are more
useful, but not as helpful to theory
 Route-finding, touring, assembly sequencing,
Internet searching

1/24/2024 Chapter 3: Solving Problems by Searching 8

Example: The Vacuum World
 States: Location, dirt or not
 Initial State: Given by problem
 Goal: No dirt in any location
 Actions: Left, Right, Vacuum, NoOp
 Path Cost: 1 per action (0 for NoOp)

1/24/2024 Chapter 3: Solving Problems by Searching 9

State Space for the Vacuum World

1/24/2024 Chapter 3: Solving Problems by Searching 10

Example: Driving in Romania
 States: In a city
 Goal: Arrive in Bucharest
 Initial State: in Arad
 Actions: Drive from one city to another

1/24/2024 Chapter 3: Solving Problems by Searching 11

Example: Driving in Romania

1/24/2024 Chapter 3: Solving Problems by Searching 12

Searching for Solutions
 State space can be represented as a search tree
(or graph) of search nodes
 Each node represents a state
 Many nodes can represent the same state
 Each arc represents a valid action from one
state to another
 A solution is a path from initial to goal nodes

1/24/2024 Chapter 3: Solving Problems by Searching 13

Search Nodes
 A search node is a data structure that
contains
 The real-world state represented by the node
 The parent node that generated it
 The action that lead to it
 The path cost to get to it (from the initial state)
 Its depth (the number of steps to reach it from
the initial state)

1/24/2024 Chapter 3: Solving Problems by Searching 14

Searching for Solutions
 The search tree starts with just the root node,
which represents the initial state
 First, use the goal test to see if the root is a goal
 If not, expand this node, which generates new
nodes
 Use successor function to apply valid actions. The
results of these actions lead to new states,
represented by new nodes
 Nodes that are generated, but not yet expanded
are in the fringe, (physically a queue data
structure)
1/24/2024 Chapter 3: Solving Problems by Searching 15
Search Strategies
 Given several states, which to consider
(i.e., goal test, expand, etc.) first?
 This is determined by the particular search
strategy being used in the algorithm

1/24/2024 Chapter 3: Solving Problems by Searching 16

General Idea

Chapter 3: Solving Problems by

1/24/2024 Searching 17
Search Algorithm Performance
 Performance measured along four dimensions:
 Completeness: Will the algorithm always return a solution, if
one exists?
 Optimality: Will it always return the optimal solution, if one
exists?
 Time complexity: How long does it take?
 Space complexity: How much memory does it take?
 Branching factor (b), depth of shallowest goal node (d),
maximum length of any path in the state space (m)
 Search cost vs. total cost
 Tradeoffs

Chapter 3: Solving Problems by

1/24/2024 Searching 18
Uninformed Search
 Uninformed (or “blind”) search algorithms
can only generate successor nodes and
do the goal test
 No other “problem insight”, common sense,
etc. is used in finding a solution
 Search strategies differ in how they
choose which node to check/expand next

Chapter 3: Solving Problems by

1/24/2024 Searching 19
Some Uninformed Search
Algorithms
 Breadth-first search
 Uniform-cost search
 Depth-first search
 Depth-limited search
 Iterative deepening depth-first search
 Bidirectional search

Chapter 3: Solving Problems by

1/24/2024 Searching 20
Breadth-First Search (BFS)
 BFS considers all nodes at a given depth before
moving on to the next depth
 Always expands the shallowest node first
 Implemented in TREE-SEARCH using a first-in
first-out (FIFO) queue
 State to consider is pulled from front of queue
 Generated states are added at the back of the queue
 Check all successors of the root node. Then check all
of their successors, etc.
Chapter 3: Solving Problems by
1/24/2024 Searching 21
Breadth-First Search (BFS)

Chapter 3: Solving Problems by

1/24/2024 Searching 22
Performance of Breadth-First
Search
 Good news
 It is complete
 It is optimal, as long as the path cost function is non-
decreasing (e.g., if all actions have the same cost)
 Bad news
 Time and (especially) space complexity can be
prohibitively high
 Has to keep all nodes in memory: queue gets large
 If d is large, takes a long time to reach a goal

Chapter 3: Solving Problems by

1/24/2024 Searching 23
Uniform-Cost Search (UCS)
 UCS expands the closest (in terms of path cost)
node first
 Ifall step costs are equal, this is equivalent to BFS
 “Closest” means total path cost so far, which is not
necessarily the same as the number of steps
 Implemented in TREE-SEARCH using a first-in
first-out (FIFO) queue
 Stateto consider is pulled from front of queue
 Generated states are added to the queue, then
contents of queue are ordered according to path cost
Chapter 3: Solving Problems by
1/24/2024 Searching 24
Performance of Uniform-Cost
Search
 Good news
 It is complete and optimal, as long as step costs are
positive
 Bad news
 Time and (especially) space complexity can be
prohibitively high, just as with BFS
 Can get stuck in an infinite loop: zero-cost steps,
repeated states

Chapter 3: Solving Problems by

1/24/2024 Searching 25
Depth-First Search (DFS)
 DFS expands the deepest node first
 Implemented in TREE-SEARCH using a last-in
first-out (LIFO) queue
 State to consider is pulled from front of queue
 Generated states are added at the front of the queue
 Check first successor of the root node. Then check its
successor, etc.
 If a non-goal node is reached with no successors,
back up just until an unexpanded node is reached,
then continue.
Chapter 3: Solving Problems by
1/24/2024 Searching 26
Depth-First Search (DFS)

Chapter 3: Solving Problems by

1/24/2024 Searching 27
Depth-First Search (DFS)

Chapter 3: Solving Problems by

1/24/2024 Searching 28
Depth-First Search (DFS)

Chapter 3: Solving Problems by

1/24/2024 Searching 29
Depth-First Search (DFS)

Chapter 3: Solving Problems by

1/24/2024 Searching 30
Depth-First Search (DFS)

Chapter 3: Solving Problems by

1/24/2024 Searching 31
Depth-First Search (DFS)

Chapter 3: Solving Problems by

1/24/2024 Searching 32
Depth-First Search (DFS)

Chapter 3: Solving Problems by

1/24/2024 Searching 33
Depth-First Search (DFS)

Chapter 3: Solving Problems by

1/24/2024 Searching 34
Depth-First Search (DFS)

Chapter 3: Solving Problems by

1/24/2024 Searching 35
Depth-First Search (DFS)

Chapter 3: Solving Problems by

1/24/2024 Searching 36
Depth-First Search (DFS)

Chapter 3: Solving Problems by

1/24/2024 Searching 37
Performance of Depth-First Search

 Good news
 Space complexity is small
 Bad news
 Neithercomplete nor optimal
 Time complexity can be large
 Can get stuck in an infinite loop: repeated states

Chapter 3: Solving Problems by

1/24/2024 Searching 38
Depth-Limited Search (DLS)
 DLS is DFS with a depth limit, L
 Nodes at depth limit are treated as leaves
 Depth limits can come from domain knowledge
 E.g., L = diameter of state space
 Implemented same as DFS, but with depth limit
 Performance
 Incompleteif L < d
 Nonoptimal if L > d
 Time and space complexity <= DFS
Chapter 3: Solving Problems by
1/24/2024 Searching 39
Iterative Deepening Depth-First
Search (IDDFS)
 IDDFS performs DLS one or more times, each
time with a larger depth limit, l
 Start with l = 0, perform DLS
 If goal not found, repeat DLS with l = 1, and so on

 Performance
 Combines benefits of BFS (completeness, optimality)
with benefits of DFS (relatively low space complexity)
 Not as inefficient as it looks
 Repeating low-depth levels: these levels do not have
many nodes Chapter 3: Solving Problems by
1/24/2024 Searching 40
Bidirectional Search
 Perform two searches simultaneously
 One search from the initial state to a goal state
 Another search from a goal state to the initial state
 Stop searching when either search reaches a state
that is in the fringe of the other state (i.e., when they
“meet in the middle”)
 Performance
 Complete and optimal if both searches are BFS
 Not easy to generate predecessors for some goals
(e.g., checkmate)
Chapter 3: Solving Problems by
1/24/2024 Searching 41
Avoiding Repeated States
 Reaching a node with a repeated state
means that there are at least two paths to
the same state
 Can lead to infinite loops
 Keep a list of all nodes expanded so far
(called the closed list; the fringe is the
open list)
 If a generated node’s state matches a state in
the closed list, discard it rather than adding it
to the fringe Chapter 3: Solving Problems by
1/24/2024 Searching 42
 End of Chapter 3

Chapter 3: Solving Problems by

1/24/2024 Searching 43