Course : Artificial Intelligence

Effective Period : July 2020

Local Search

Session 03

1
 Learning Outcomes
At the end of this session, students will be able to:
 LO 2: Describe what is AI and identify concept of intelligent
agent
 LO 3: Explain various intelligent search algorithms to solve
the problems

2
 Outline
1. Local Search Algorithms and Optimization Problems
2. Hill Climbing Search (Steepest-Ascent)
3. Genetic Algorithms
4. Local Search in Continuous Spaces
5. Constraint Satisfaction Problems Definition
6. Local Search for Constraint Satisfaction Problems

3
 Local Search
• In the problem we studied so far, the solution is the path.
For example:
– The solution of the traveling in Romania problem is a
sequence of cities to get to Bucharest
• In many optimization problem, the path is irrelevant. The
goal itself is the solution. For example:
– The 8 queens problem
• What matters is the final configuration of queens,
not the order in which they are added
4
 Local Search
• Local search algorithms operate using a single current
node (rather than multiple paths) and generally move only
to neighbors of that node.
– Typically, the paths followed by the search are not
retained.
• Local search has two advantages
– They use very little memory
– They can often find reasonable solutions in large or
infinite (continuous) state spaces
5
 Local Search
• Local search algorithms are useful for solving pure
optimization problems
– To find a best state according to an objecting function

6
 Local Search
• The aim of the optimization algorithm is to find the highest
peak (global maximum)

7
 Local Search
• A complete algorithm always finds a goal if it exists
• An optimal algorithm always finds the global
maximum/minimum

8
 Hill Climbing
• It is simply that continually moves in the direction of
increasing value (uphill)
• It terminates when it reaches a “peak” where no neighbor
has a higher value
• It starts with a random (potentially poor) solution, and
iteratively makes small changes to the solution, each time
improving it a little.

9
 Hill Climbing
• All successors of a node are evaluated and the one that
give the most improvement is selected

• Choose the successor with the best solution

– Highest-valued = best solution

• Hill-climbing is like “climbing a mountain without

compass”

10
 Hill Climbing
• Example 1 2 3
Start state 4 6
h(x) = 2
7 5 8

1 3 1 2 3 1 2 3 1 2 3
4 2 6 4 6 4 5 6 4 6
7 5 8 7 5 8 7 8 7 5 8
h(x) = 3 h(x) = 3 h(x) = 1 h(x) = 3

1 2 3
Goal state 4 5 6 h(x) = 0
7 8
11
 Problems in Hill Climbing
• Hill Climbing search often get stuck due to the following
conditions:
– Local maxima, ridges, and plateaux

12
 Problems in Hill Climbing
• Local maxima/minima
– A peak that is higher than each of its neighboring
states but lower than the global maximum.

13
 Problems in Hill Climbing
• Ridges (Sequence of local maxima)
– It is very difficult for greedy algorithms to navigate

14
 Problems in Hill Climbing
• Plateaux
– A flat area of the state-space landscape
– It can be a flat local maximum or a shoulder

Not Difference ?

15
 Problems in Hill Climbing
• Ways Out
– Backtrack to some earlier node and try going in a
different direction.

– Make a big jump to try to get in a new section.

– Moving in several directions at once.

16
 Genetic Algorithm
• Is a variant of stochastic beam search
– Successor states are generated by combining two
parent states
– Similar with our DNA which is a mixed of our parents’
DNA

17
 Genetic Algorithm
• Starts with k randomly generated states, called population
– Each state is an individual, represented as a string
over a finite alphabet (ex. 01010100)
• The objective function is called fitness function: better
states have high values of fitness function

18
 Genetic Algorithm
• In the 8-queen problem, an individual can be represented
by a string digits 1 to 8, that represents the position of the 8
queens in the 8 columns.

19
 Genetic Algorithm
• Possible fitness function is the number of non-attacking
pairs of queens
• Fitness function of the solution
– 7 + 6 + 5 + 4 + 3 + 2 + 1 = 28
• How much? 

20
 Genetic Algorithm
• Pairs of individuals are selected at random for
reproduction w.r.t. some probabilities

21
 Genetic Algorithm
• A crossover point is chosen randomly in the string
• Offspring are created by crossing the parents at the
crossover point.

22
 Genetic Algorithm
• Each element in the string is also subject to some
mutation with a small probability

23
Genetic Algorithm

24
 Local Search in Continuous Space
• Suppose we want to site three airports in Jakarta:

– 6-D state space defined by (x1,y1), (x2,y2), (x3,y3)

– Objective function f(x1,y1, x2,y2, x3,y3) = sum of squared

distances from each city to nearest airports
• How to choose select the next successor?
– Use gradient to increase/reduce f, i.e. by

25
 Local Search in Continuous Space
• Important: we need to understand what function that we
want to optimize!
– Highest value is better  Gradient ascent optimization
– Lowest value is better  Gradient descent optimization

• Local search in continuous space plays important role in

the current state-of-the-art deep learning algorithm

26
Defining Constraint Satisfaction Problem
• Standard search problem

– state is a "black box“ – any data structure that supports

successor function, heuristic function, and goal test
• Constraint satisfaction problem (CSP)

– state is defined by variables Xi with values from

domain Di

– goal test is a set of constraints specifying allowable

combinations of values for subsets of variables

27
Defining Constraint Satisfaction Problem
Map coloring

• Variables WA, NT, Q, NSW, V, SA, T

• Domains Di = {red, green, blue}
• Constraints: adjacent regions must have different colors
• e.g., WA ≠ NT, NT ≠ SA, (WA,NT) є {(red, green),(red, blue),(green,
red), (green, blue),(blue, red),(blue, green), …} 28
Defining Constraint Satisfaction Problem

• Solution is assigning value for each variable fulfill the constraints

• e.g., {WA = red, NT = green, Q = red, NSW = green, V = red, SA =
blue, T = green}

29
 Local Search for CSP
• Hill-climbing typically works with "complete" states, i.e., all
variables assigned
• To apply to CSP

– allow states with unsatisfied constraints

– operators reassign variable values

• Variable selection: randomly select any conflicted variable

30
 Local Search for CSP
• Value selection by min-conflicts heuristic:
– choose value that violates the fewest constraints
– i.e., hill-climb with h(n) = total number of violated
constraints
• Given random initial state, can solve n-queens in almost
constant time for arbitrary n with high probability (e.g., n =
10,000,000)

31
 Local Search for CSP: Example
• States: 4 queens in 4 columns (44 = 256 states)

• Actions: move queen in column

• Goal test: no attacks

• Evaluation: h(n) = number of attacks

32
 Local Search for CSP: Example
• States: 8 queens in 8 columns (88 = 65,536 states)

• Actions: move queen in column

• Goal test: no attacks

• Evaluation: h(n) = number of attacks

33
 References
• Stuart Russell, Peter Norvig. 2010. Artificial Intelligence :
A Modern Approach. Pearson Education. New Jersey.
ISBN:9780132071482
• Local Search:
[Link]
T2017_2+type@asset+block@AI_edx_SearchAgents_Local.p
df

34