Presentation is loading. Please wait.

Presentation is loading. Please wait.

Chapter 3 Heuristic Search Techniques 323-670 Artificial Intelligence ดร. วิภาดา เวทย์ประสิทธิ์ ภาควิชาวิทยาการคอมพิวเตอร์ คณะ วิทยาศาสตร์ มหาวิทยาลัยสงขลานครินทร์

Similar presentations


Presentation on theme: "Chapter 3 Heuristic Search Techniques 323-670 Artificial Intelligence ดร. วิภาดา เวทย์ประสิทธิ์ ภาควิชาวิทยาการคอมพิวเตอร์ คณะ วิทยาศาสตร์ มหาวิทยาลัยสงขลานครินทร์"— Presentation transcript:

1 Chapter 3 Heuristic Search Techniques Artificial Intelligence ดร. วิภาดา เวทย์ประสิทธิ์ ภาควิชาวิทยาการคอมพิวเตอร์ คณะ วิทยาศาสตร์ มหาวิทยาลัยสงขลานครินทร์

2 Artificial Intelligence Lecture 7-12Page 2 Production System  Working memory  Production set = Rules Figure 5.3  Trace Figure 5.4 Data driven Figure 5.9 Goal driven Figure 5.10  Iteration #  Working memory  Conflict sets  Rule fired

3 Artificial Intelligence Lecture 7-12Page 3 And-Or Graph a Data driven Goal driven b cd efg

4 Artificial Intelligence Lecture 7-12Page 4 Generate-and-test  Generate all possible solutions  DFS + backtracking  Generate randomly  Test function  yes/no  Algorithm page 64

5 Artificial Intelligence Lecture 7-12Page 5 Hill Climbing  Similar to generate-and-test  Test function + heuristic function  Stop  Goal state meet  No alternative state to move

6 Artificial Intelligence Lecture 7-12Page 6 Simple Hill Climbing  Task specific knowledge into the control process  Is one state better than another  The first state is better than the current state  Algorithm page 66

7 Artificial Intelligence Lecture 7-12Page 7 Steepest-Ascent Hill Climbing  Consider all moves from the current state  Select the best one as the next state  Algorithm page 67  Searching time?

8 Artificial Intelligence Lecture 7-12Page 8 Hill Climbing Problem  No solution found : Problem  Local maximum : a state that is better than all its neighbors but it is not better than some other states farther away.  backtracking  Plateau : a flat area of the search space in which a whole set of neighboring states have the same value. It is not possible to determine the best direction by using local comparison.  Make big jump

9 Artificial Intelligence Lecture 7-12Page 9 Hill Climbing Problem  Ridge : an area of the search space that is higher than surrounding areas and itself has a slope. We can not do with a single move.  Fired more rules for several direction

10 Artificial Intelligence Lecture 7-12Page 10 Hill Climbing Characteristic  Local method  It decides what to do next by looking only at the immediate consequences of its choice (rather than by exhaustively exploring all of the consequence)  Look only one more ahead

11 Artificial Intelligence Lecture 7-12Page 11 Local heuristic function  Block world figure 3.1 p. 69  Local heuristic function 1. Add one point for every block that is resting on the thing it is supposed to be resting on. 2. Subtract one point for every block that is sitting on the wrong thing.  Initial state score = 4 (6-2)  C,D,E,F,G,H correct = 6  A,B wrong = -2  Goal state score = 8  A,B,C,D,E,F,F,H all correct

12 Artificial Intelligence Lecture 7-12Page 12 Local heuristic function  Current state : จากรูป 3.1  หยิบ A วางบนโต๊ะ B C D E F G H วางเรียง เหมือนเดิม  Score = 6 (B C D E F G H correct)  Block world figure 3.2 p. 69  Next state score = 4  All 3 cases  Stop : no better score than the current state = 6  Local minimum problem  ติดอยู่ในกลุ่มระดับ local มองไปไม่พ้นอ่าง

13 Artificial Intelligence Lecture 7-12Page 13 Global heuristic function  Block world figure 3.1 p. 69  Global heuristic function 1.For each block that has the correct support structure add one point for every block in the support structure. ( นับ หมด )  For each block that has an incorrect support structure, subtract one point for every block in the existing support structure.

14 Artificial Intelligence Lecture 7-12Page 14 Global heuristic function  initial state score = -28  C = -1, D = -2, E = -3, F = -4, G = -5, H = -6, A = -7  Goal state score = 28  B = 1, C = 2, D = 3, E = 4, F = 5, G = 6, H = 7

15 Artificial Intelligence Lecture 7-12Page 15 Global heuristic function  Current state : จากรูป 3. 1  หยิบ A วางบนโต๊ะ B C D E F G H วางเรียง เหมือนเดิม  Score = -21  (C = -1, D = -2, E= -3, F = -4, G= -5, H = -6)  Block world figure 3.2 p. 69  Next state : move to case(c)  Case(a) = -28 same as initial state  Case(b) = -16  (C = -1, D = -2, E= -3, F = -4, G= -5, H = -1)  Case(c) = -15  (C = -1, D = -2, E= -3, F = -4, G= -5)  No Local minimum problem  It’s work

16 Artificial Intelligence Lecture 7-12Page 16 New heuristic function 1. Incorrect structure are bad and should be taken apart. More subtract score 2. Correct structure are good and should built up.  Add more score for the correct structure.  สิ่งที่เราต้องพิจารณา  How to find a perfect heuristic function?  เข้าไปในเมองที่ไม่เคยไปจะหลีกเลี่ยงทางตัน dead end ได้อย่างไร

17 Artificial Intelligence Lecture 7-12Page 17 Simulated Annealing  Hill climbing variation  At the beginning of the process some down hill moves may be made.  Do enough exploration of the whole space early on so that the final solution is relatively insensitive to the starting state.  ป้องกันปัญหา local maximum, plateau,ridge  Use objective function (not heuristic function)  Use minimize value of objective function

18 Artificial Intelligence Lecture 7-12Page 18 Simulated Annealing  Annealing schedule  ถ้าเราทำให้เย็นเร็วมาก จะได้ผลลัพธ์ high energy อาจเกิด local minimum ได้  ถ้าเราทำให้เย็นช้ามาก จะได้ผลลัพธ์ดี แต่ เสียเวลามาก at low temperatures a lot of time may be wasted after the final structure has already been formed.  ควรทำแบบพอดี empirical structure

19 Artificial Intelligence Lecture 7-12Page 19 Simulated Annealing  Annealing : metals are melted  Cool down to get the solid structure  Objective function : energy level  Try to use less energy  P : probability  T : temperature : annealing schedule  K : Boltzmann’s constant : describe the correspondence between the units of temperature and the unit of energy   E = ( value of current) – (value of new state)  positive change in the energy -  e/KT p = e

20 Artificial Intelligence Lecture 7-12Page 20 Simulated Annealing  Probability of a large uphill move is lower than probability of a small uphill move  Probability uphill move decrease when temperature decrease.  In the beginning of the annealing large upward moves may occur early on  Downhill moves are allowed anytime  Only relative small upward moves are allowed until finally the process converges to a local minimum configuration

21 Artificial Intelligence Lecture 7-12Page 21 Simulated Annealing -  e/KT p = e

22 Artificial Intelligence Lecture 7-12Page 22  เหมาะสำหรับปัญหาที่มีจำนวน move มากๆ หลักการ 1. What is initial Temperature 2. Criteria for decreasing T 3. Level to decrease T value 4. When to quit ข้อสังเกต 1. When T approach 0 simulated annealing identical with simple hill climbing Algorithm Simulated Annealing

23 Artificial Intelligence Lecture 7-12Page 23 ข้อแตกต่าง Algorithm Simulated Annealing p.71 และ Hill Climbing 1. The annealing schedule must be maintained. 2. Move to worse states may be accepted. 3. Maintain the best state found so far. If the final state is worse than that earlier state, then earlier state is still available. Algorithm Simulated Annealing

24 Artificial Intelligence Lecture 7-12Page 24 Best first search  OR GRAPH : Search in the graph Heuristic function : min value page 74

25 Artificial Intelligence Lecture 7-12Page 25 Best First Search  OR GRAPH : each of its branches represents an alternative problem-solving pattern  we assumed that we could evaluate multiple paths to the same node independently of each other  we want to find a single path to the goal  use DFS : select most promising path  use BSF : when no promising path/ switch part to receive the better value  old branch is not forgotten  solution can be find without all completing branches having to expanded

26 Artificial Intelligence Lecture 7-12Page 26 Best First Search f’ = g + h’ g: cost from initial state to current state h’: estimate cost current state to goal state f’: estimate cost initial state to goal state Open node : most promising node Close node : keep in memory, already discover node.

27 Artificial Intelligence Lecture 7-12Page 27 Best First Search Algorithm page 75-76

28 Artificial Intelligence Lecture 7-12Page 28 A* algorithm  h’ : count the nodes that we step down the path, 1 level down = 1 point, except the root node.  Underestimate : we generate up until f’(F)= 6 > f’(C) =5 then we have to go back to C. 1 level 2 level 3 level f ’ (E) = f ’ (C) = 5 f’ = g + h’

29 Artificial Intelligence Lecture 7-12Page 29 A* algorithm f’ = g + h’ 1 level 2 level 3 level Overestimate : Suppose the solution is under D : we will not generate D because F ’ (D) = 6 > f ’ (G) = 4.

30 Artificial Intelligence Lecture 7-12Page 30 A* Algorithm page 76

31 Artificial Intelligence Lecture 7-12Page 31 A* Algorithm

32 Artificial Intelligence Lecture 7-12Page 32 Agenda  Agenda : a list of tasks a system could perform.  a list of reasons  a rating representing overall weight of evidence suggesting that the task would be useful  When a new task is created, insert into the agenda in its proper place, we need to re- compute its rating and move it to the correct place in the list  find the better location  put at the end of agenda  need a lot more time to compute a new rating

33 Artificial Intelligence Lecture 7-12Page 33 Not acceptable dialog  agenda is not good for when interacting with people page  person China computer  person Italy  computer  person  computer China  something reasonable now may not be continue to be so after the conversation has processed for a while.

34 Artificial Intelligence Lecture 7-12Page 34 Agenda

35 Artificial Intelligence Lecture 7-12Page 35 And-Or Graph / Tree  can be solved by decomposing them into a set of smaller problems  And arcs are indicated with a line connecting all the components

36 Artificial Intelligence Lecture 7-12Page 36 And-Or Graph / Tree  each arc with the successor has a cost of choose lowest value = f ’ (B) = 5 ABEF =

37 Artificial Intelligence Lecture 7-12Page 37 And-Or Graph / Tree  Futility : some value use to compare the result/ threshold value  If... the estimate cost of a solution > Futility then.....abandon the search

38 Artificial Intelligence Lecture 7-12Page 38 Problem Reduction

39 Artificial Intelligence Lecture 7-12Page 39 Problem Reduction E come from J not C

40 Artificial Intelligence Lecture 7-12Page 40 Problem Reduction Can not find a solution from this algorithm because of C

41 Artificial Intelligence Lecture 7-12Page 41 Problem Reduction : AO* Algorithm

42 Artificial Intelligence Lecture 7-12Page 42 Problem Reduction : AO* Algorithm  use a single structure GRAPH  we will not store g  algorithm will insert all ancestor nodes into a set path C will always be better than path B

43 Artificial Intelligence Lecture 7-12Page 43 Problem Reduction : AO* Algorithm change G from 5 to 10 no backward propagationneed backward propagation


Download ppt "Chapter 3 Heuristic Search Techniques 323-670 Artificial Intelligence ดร. วิภาดา เวทย์ประสิทธิ์ ภาควิชาวิทยาการคอมพิวเตอร์ คณะ วิทยาศาสตร์ มหาวิทยาลัยสงขลานครินทร์"

Similar presentations


Ads by Google