Tr tu nhn to (Articial Intelligence)$

Cc chin lc tm kim c i th

Phm Bo Sn

Gii thiu$
Nghin cu my tnh chi c xut hin t rt sm) Khng lu sau khi my tnh lp trnh ra i, nm 1950, Claude Shannon vit chng trnh chi c u tin) Vic my tnh chi c l mt bng chng v kh nng my tnh c th lm nhng cng vic i hi tr thng minh ca con ngi)
Phm Bo Sn 2

Types of Games$
Discrete Games)
Fully observable, deterministic (check, checkers, go, othello)) Fully observable, stochastic (backgammon, monopoly)) Partially observable (bridge, poker, scrabble))

Continuous, embodied games:)

Robocup soccer, pool (snooker))
Phm Bo Sn 3

Go $

Phm Bo Sn

Checkers$

Phm Bo Sn

Robocup Soccer $

Phm Bo Sn

Deep Green chi bi-a

Phm Bo Sn

Why Games?$
Unpredictable opponent: solution is a strategy:)
Must respond to every possible opponent reply)

Time limits: must rely on approximation)

Tradeoff between speed and accuracy)

Games have been a key driver of new techniques in CS and AI.)

Phm Bo Sn 8

Tm kim c i th$
S nghin cu cc tr chi c 2 ngi tham gia) Xem xt cc vn :)
Chi c c th xem nh vn tm kim trong khng gian trng thi) Chin lc tm kim nc i Minimax) Phng php ct ct -: - pruning )
Phm Bo Sn 9

Bi ton$
Gi s mt ngi chi cm qun Trng, mt ngi chi cm qun en nghin cu my tnh cm qun Trng) Nghin cu cc tr chi: )
Hai ngi thay phin nhau a ra cc nc i tun theo lut i no . ) Cc lut ny l nh nhau cho c hai ngi) Bit c thng tin y v cc tnh th trong tr chi (khng ging nh chi bi)) V d: C vua, c tng)

Bi ton: Tm kim nc i trong cc nc i c th thng)

Phm Bo Sn 10

Khng gian trng thi cho tr chi ()$

Vn tm kim khi c i th s phc tp hn v ngi chi khng bit c i th ca mnh s i nc no trong tng lai) Vn chi c c th xem nh vn tm kim trong khng gian trng thi:)
Mi trng thi l mt tnh th) Trng thi ban u l tnh th lc bt u cuc chi) Cc ton t l cc nc i hp l) Cc trng thi kt thc l cc tnh th m cuc chi dng) Mt hm kt cuc ng vi mi trng thi kt thc vi mt gi tr no , v d thng l 1, thua l -1, ha l 0)
Phm Bo Sn 11

Khng gian trng thi cho tr chi ()$

Vn ca Trng l tm mt dy nc i sao cho xen k vi cc nc i ca en to thnh mt ng i t trng thi ban u ti gi tr kt cuc l cao nht) Khng bit trc nc i ca i th: nghim l 1 chin thut chi.)
lit k ht cc nc i ca i th.)

Phm Bo Sn

12

Cy tr chi$
thun li, ta biu din khng gian trng thi di dng cy tr chi) Cy tr chi c xy dng nh sau:)
Gc ca cy ng vi trng thi ban u) Gi nh ng vi trng thi m Trng (en) s a ra nc i l nh Trng (en))

Phm Bo Sn

13

Game Tree 2-player, deterministic$

Phm Bo Sn

14

Chin lc Minimax ()$

Perfect Play for deterministic, perfect-information games.) Idea: choose move to position with highest minimax value = best achievable payoff against best play)

Phm Bo Sn

15

 Chn nc i vi gi tr minimax ln nht:) Trng = Max, en = Min.) i ngc t cc trng thi kt thc) Gn gi tr cho cc trng thi kt thc l gi tr ca hm kt cuc) i ngc t di ln: ) nu l nh trng th gn gi tr l max ca gi tr nhng nt con ca n; ) nu l nh en th gn gi tr l min ca gi tr nhng nt con ca n) Trng: chn nc i l nt con c gi tr ln nht)
Phm Bo Sn 16

Chin lc Minimax$

Chin lc Minimax ()$

Phm Bo Sn

17

Chin lc Minimax ()$

L thut ton tm kim theo su) Cho php ta chn c nc i ti u) Tuy nhin phc tp qu ln) C th hn ch su ca cy tr chi v s dng hm nh gi gim bt cy tm kim)
Phm Bo Sn 18

Chin lc Minimax ()$

Phm Bo Sn

19

nh gi Minimax$
Complete?) Optimal? ) Time complexity? ) Space complexity? )

Phm Bo Sn

20

nh gi Minimax$
Complete? Yes (if tree is nite)) Optimal? Yes (against an optimal opponent)) Time complexity? O(bm)) Space complexity? O(bm) (depth-rst exploration))

Chi c vua: b ~ 35, m ~100: tm nghim ti u l khng kh thi. )

bm = 106 b= 35 -> m = 4) 4-ply = ngi mi chi) 8-ply = my tnh thng thng, ngi chi gii) 12-ply= Deep Blue, Kasparov )

Phm Bo Sn

21

Hm nh gi$
Thng l gi tr nh gi li th ca trng thi u eval(u)) Trng thi u cng thun li cho Trng th eval(u) l s dng cng ln: MAX ) Trng thi u cng thun li cho en th eval(u) l s m cng nh: MIN) eval(u) 0 i vi trng thi u khng li th cho ai c) eval(u) = rt ln Trng thng) eval(u) = rt nh - en thng)

Phm Bo Sn

22

Hm nh gi ()$
Cht lng ca chng trnh chi c rt ph thuc vo hm nh gi) Tuy nhin tt ca mt hm nh gi thng mu thun vi thi gian tnh n)

Phm Bo Sn

23

Hm nh gi ()$
V d: C vua, c tng) Mi loi qun gn cho mt gi tr ph hp vi sc mnh, v d Tt (1), M (5) ri ly tng) C th xt n mi tng quan v tr gia cc qun c)

Phm Bo Sn

24

Pruning - Motivation$

Q1. Why would Queen to G5 be a bad move for Black?) Q2. How many White replies did you need to consider in answering?) Once we have seen one reply scary enough to convince us the move is really bad, we can abandon this move and continue searching elsewhere. )
Phm Bo Sn 25

Phng php ct ct alphabeta$

Cho php ct b cc nhnh khng cn thit cho s nh gi nh u)
Nu a l nh trng, nu u, v c gn v gi tr ca u > gi tr ca v th khng cn gn gi tr ca a na Nu a l nh en, nu u, v c gn v gi tr ca u < gi tr ca v th khng cn gn gi tr ca a na
Phm Bo Sn

c b u v
26

Phng php ct ct alphabeta$

Phm Bo Sn

27

Phng php ct ct alphabeta$

Phm Bo Sn

28

Phng php ct ct alphabeta$

Phm Bo Sn

29

Phng php ct ct alphabeta$

Phm Bo Sn

30

Phng php ct ct alphabeta$

Phm Bo Sn

31

Phng php ct ct alphabeta$

Phng php ct ct alpha-beta khng nh hng n kt qa cui cng, ch nh hng n thi gian tm kim.) Th t sp xp cc bc i trong cy tm kim nh c nh hng ln n cht lng ca phng php ct ct alpha-beta.) Vi mt sp xp hon ho, time complexity = ) ) O(bm/2). )

Phm Bo Sn

32

Ti sao li gi l -?$
l gi tr tt nht (i.e., gi tr cao nht) tnh n thi im hin ti cho max) Nu v km hn , max s loi b v)
ct nhnh)

nh ngha tng t cho min)

Phm Bo Sn

33

Phng php ct ct alphabeta$

Phm Bo Sn

34

Deterministic games in practice$

Checkers: Chinook ended 40-year-reign of human world champion Marion Tinsley in 1994. Used a pre-computed endgame database dening perfect play for all positions involving 8 or fewer pieces on the board, a total of 444 billion positions.) Chess: Deep Blue defeated human world champion Garry Kasparov in a six-game match in 1997. Deep Blue searches 200 million positions per second, uses very sophisticated evaluation, and undisclosed methods for extending some lines of search up to 40 ply.) Othello: human champions refuse to compete against computers, who are too good.) Go: human champions refuse to compete against computers, who are too bad. In go, b > 300, so most programs use pattern knowledge bases to suggest plausible moves.)

Phm Bo Sn

35

Tm tt$
Cc bi ton tr chi c i th rt th v. ) Chi c c th xem nh vn tm kim trong khng gian trng thi) Chin lc tm kim nc i Minimax) Phng php ct ct -)
Phm Bo Sn 36

References$
Articial Intelligence: A modern approach. Chapter 6.) Articial Intelligence Illuminated. Chapter 6.)

Phm Bo Sn

37