Posts and Telecommunication Institute of Technology

Faculty of Information Technology 1

Introduction to Artificial Intelligence

Informed search

Dao Thi Thuy Quynh

 Blind search & Informed search
} Blind search
o Expands nodes according to a pre-defined rule
o Moves in the state space without orientation; many states must be
considered
o Not suitable for problems with large state spaces

} Informed search
o Uses additional information from the problem to guide the search
o Uses a function 𝑓(𝑛) to evaluate the potential of node 𝑛, then
choose the best node to expand first
§ Best-first search
§ How to construct function 𝑓(𝑛)?

2 http://www.ptit.edu.vn
 Outline
} Greedy search
} A* search
} Heuristic functions
} Iterative deepening A* (IDA*)

3 http://www.ptit.edu.vn
 Example

4 http://www.ptit.edu.vn
 Greedy search
} Principle: expand the node with the cheapest path to a
goal first
o 𝑓 𝑛 = ℎ(𝑛): heuristic function, estimated cost of the path from 𝑛
to a goal node
o Example: ℎ(𝑛) = straight-line distance from 𝑛 to a goal node

} “Greedy”: expands the node that appears to be closest to

goal, regardless of the future

5 http://www.ptit.edu.vn
 Example of greedy search (1/4)
S
125

6 http://www.ptit.edu.vn
 Example of greedy search (2/4)

Expand S S

A B C E
123 82 118 72

7 http://www.ptit.edu.vn
 Example of greedy search (3/4)
Expand S S

Expand E A B C E
123 82 118

D G
115 0

8 http://www.ptit.edu.vn
 Example of greedy search (4/4)
Expand S S

Expand E A B C E
123 82 118

Expand G: Goal D G
115 0

9 http://www.ptit.edu.vn
 Properties of greedy search
} Completeness?
o No (can have a loop, or have a branch of infinite nodes with small
value of function ℎ but do not lead to a goal)
} Optimality?
o No
} Time?
o 𝑂(𝑏 ! )
o If the heuristic function is good, the algorithm can be much faster
} Space?
o 𝑂(𝑏 ! ): store all nodes in the memory
o If the heuristic function is good, the number of nodes to store can be
reduced significantly

10 http://www.ptit.edu.vn
 Exercise 1
} Use greedy search algorithm to find the path from 𝑆 to 𝐺?

(Phuong TM, 2016)

11 http://www.ptit.edu.vn
 Outline
} Greedy search
} A* search
} Heuristic functions
} Iterative deepening A* (IDA*)

12 http://www.ptit.edu.vn
 A* search: idea
} Overcome the disadvantage of greedy search
o Greedy: only care about the path to the goal
o A*: also care about the path from the initial node to the current
node
§ avoid expanding paths that are already expensive

} Method: 𝑓 𝑛 = 𝑔 𝑛 + ℎ(𝑛)
o 𝑔 𝑛 : cost so far to reach 𝑛
o ℎ(𝑛): heuristic function, estimated cost from 𝑛 to a goal node
o 𝑓 𝑛 : estimated total cost of the path from the initial node,
through 𝑛, to a goal node

13 http://www.ptit.edu.vn
 Example of A* search (1/5)
S
0+125=125

14 http://www.ptit.edu.vn
 Example of A* search (2/5)

Expand S S

A B C E
55+123=178 42+82=124 48+118=166 72+72=144

15 http://www.ptit.edu.vn
 Example of A* search (3/5)

Expand S S

Expand B A B C E
55+123=178 48+118=166 72+72=144

F
82+40=122

16 http://www.ptit.edu.vn
 Example of A* search (4/5)

Expand S S

Expand B A B C E
55+123=178 48+118=166 72+72=144

Expand F F

G
137+0=137

17 http://www.ptit.edu.vn
 Example of A* search (5/5)

Expand S S

Expand B A B C E
55+123=178 48+118=166 72+72=144

Expand F F

G
Expand G: Goal
137

18 http://www.ptit.edu.vn
 A* search algorithm
𝐴 ∗ (𝑄, 𝑆, 𝐺, 𝑃, 𝑐, ℎ)
(𝑄: state space, 𝑆: initial state, 𝐺: goals, 𝑃: successor function, c: cost,
h: heuristic)
Input: search problem, heuristic function ℎ
Output: goal state (path to the goal state)
Initialize: 𝑂 ← 𝑆 (𝑂: the open node list)

while (𝑂 ≠ Ø) do
1. take node 𝑛 whose 𝑓(𝑛) is the smallest from 𝑂
2. if 𝑛 ∈ 𝐺, return (path to 𝑛)
3. for each 𝑚 ∈ 𝑃(𝑛)
a) 𝑔(𝑚) = 𝑔(𝑛) + 𝑐(𝑛, 𝑚)
b) 𝑓(𝑚) = 𝑔(𝑚) + ℎ(𝑚)
c) add 𝑚 along with 𝑓(𝑚) to 𝑂
return no solution

19 http://www.ptit.edu.vn
 Properties of A* search
} Complete?
o Yes (unless there are infinite nodes 𝑛 with 𝑓(𝑛) ≤ 𝑓(𝐺))
} Optimality?
o Yes (if heuristic function ℎ is admissible)
} Time?
o 𝑂(𝑏 ! )
o If the heuristic function is good, the algorithm can be much faster
} Space?
o 𝑂(𝑏 ! ): store all nodes in the memory
o If the heuristic function is good, the number of nodes to store can be
reduced significantly

20 http://www.ptit.edu.vn
 The optimality of A* search
} Admissible heuristic function
o For every node 𝑛, ℎ 𝑛 ≤ ℎ∗(𝑛), where ℎ∗(𝑛) is the true cheapest
cost from 𝑛 to a goal node
o Example: straight-line distance is an admissible heuristic function

} Theorem: A* algorithm is guaranteed to find an optimal

solution if ℎ(𝑛) is admissible

21 http://www.ptit.edu.vn
 Exercise 2
} Use A* search to find the path from 𝑆 to 𝐺?

(Phuong TM, 2016)

22 http://www.ptit.edu.vn
 Outline
} Greedy search
} A* search
} Heuristic functions
} Iterative deepening A* (IDA*)

23 http://www.ptit.edu.vn
 Heuristic functions
} Heuristic functions are constructed depending on each
specific problem
o A problem may have some heuristic functions
o The quality of the heuristic function greatly affects the search
process

} Dominance
o If ℎ# (𝑛) and ℎ$ (𝑛) are 2 admissible heuristic functions satisfying
ℎ# (𝑛) ≤ ℎ$ (𝑛) for all nodes n, then ℎ$ dominates (is better than)
ℎ#

24 http://www.ptit.edu.vn
 Example of heuristic functions

} ℎ'(𝑛): number of wrong cells

o ℎ# 𝑆 = 8
} ℎ((𝑛): total of Manhattan distances
o ℎ$ 𝑆 = 3 + 1 + 2 + 2 + 2 + 3 + 3 + 2 = 18

25 http://www.ptit.edu.vn
 Outline
} Greedy search
} A* search
} Heuristic functions
} Iterative deepening A* (IDA*)

26 http://www.ptit.edu.vn
 Iterative deepening A* – IDA*
} Goal: deals with the memory problem in A* search
o a variant of IDS (uses the heuristic function from A* search to
limit nodes)

} Method: Iterate DFS over sub-trees where 𝑓(𝑛) is not

greater than a threshold
o The threshold is incremented after each loop so that we have new
nodes

27 http://www.ptit.edu.vn
 IDA* algorithm
𝐼𝐷𝐴∗(𝑄, 𝑆, 𝐺, 𝑃, 𝑐, ℎ)
Input: search problem, heuristic function ℎ
Output: goal state (path to the goal state)
Initialize: 𝑂 ← 𝑆 (𝑂: the open node list)
Threshold 𝑖 ← 0
while (1) do
1. while (𝑂 ≠ ∅) do
a) Take the first node 𝑛 from 𝑂
b) if 𝑛 ∈ 𝐺, return (path to 𝑛)
c) For each 𝑚 ∈ 𝑃(𝑛)
i) 𝑔(𝑚) = 𝑔(𝑛) + 𝑐(𝑚, 𝑛)
ii) 𝑓(𝑚) = 𝑔(𝑚) + ℎ(𝑚)
iii) if 𝑓(𝑚) ≤ 𝑖 then add 𝑚 to the head of 𝑂
2. 𝑖 ¬ 𝑖 + b, 𝑂 ¬ 𝑆
28 http://www.ptit.edu.vn
 Properties of IDA*

} Completeness?
o Yes
} Optimality?
o 𝛽-optimal (cost of the found solution does not exceed 𝛽 compared
to cost of the optimal solution)
} Time?
o Computational complexity is greater than that of A* search
} Space?
o Requires linear memory

29 http://www.ptit.edu.vn
 Exercise 3
} Use IDA* search to find the path from 𝑆 to 𝐺 with 𝛽 = 2?

(Phuong TM, 2016)

30 http://www.ptit.edu.vn
 When to add repeated nodes
to the open node list?
} Greedy
o No: adding repeated nodes does not change the algorithm (may
lead to loops)

} A*
o In cases the repeated node has better cost, it will be added to the
list (if it is already expanded) or updated to replace the old node
(if it is on the list)

} IDA*:
o Yes

31 http://www.ptit.edu.vn