TR TU NHN T O Artificial Intelligence

TS. Nguy n nh Thun Khoa Cng ngh Thng tin i h c Nha Trang Email: thuanvinh122@gmail.com
Nha Trang 8-2007

Noi dung mon hoc

Chng 1: Gi i thi u

M u Lnh v c nghin c u c a AI ng d ng c a AI Cc v n t ra

Slide 2

Noi dung mon hoc (ti p)

Chng 2:Tm ki m trn khng gian tr ng thi

Bi ton tm ki m Gi i thu t t ng qut Depth first search (DFS) Breath first search (BFS)

Chng 3:Tm ki m theo Heuristic

Gi i thi u v Heuristic Tm ki m theo heuristic Gi i thu t Best first search (BFS), Gi i thu t AT, AKT, A* Chi n l c Minimax, Alpha Beta
Slide 3

Noi dung mon hoc (ti p)

Chng 4:Bi u di n tri th c

B ba i t ng Thu c tnh Gi tr Cc lu t d n M ng ng ngha Frame Logic m nh , Logic v t Thu t gi i Vng H o, Thu t gi i Robinson Cc hnh th c h c Thu t gi i Quinland H c theo b t nh
Slide 4

Chng 5: My h c

Th c hnh &Ti li u tham kh o

Th c hnh Prolog / C++ / Pascal

Cc gi i thu t tm ki m Bi u di n tri th c Bi t p l n Bi gi ng Tr tu nhn t o TS Nguy n nh Thun Gio trnh Tr tu nhn t o - GS Hong Ki m HQGTPHCM Tr tu nh n t oPGS Nguy n Thanh Th y H Bch Khoa HN i Artificial Inteligent George F. Luget & Cilliam A. Stubblefied

Ti li u tham kh o

Slide 5

Chng 1: GII THIEU

TS. Nguy n nh Thun Khoa Cng ngh Thng tin i h c Nha Trang Email: thuanvinh122@gmail.com

1.1 M

Tr tu l g: Theo t i n Bch khoa ton th Webster: Tr tu l kh nng:

Ph n ng m t cch thch h p l i nh ng tnh hu ng m i thng qua i u ch nh hnh vi m t cch thch h p. Hi u r m i lin h gi a cc s ki n c a th gi i bn ngoi nh m a ra nh ng hnh vi ph h p t c m c ch.
Slide 7

Thng Minh

Khi ni m v tnh thng minh c a m t i t ng th ng bi u hi n qua cc ho t ng:

S hi u bi t v nh n th c c tri th c S l lu n t o ra tri th c m i d a trn tri th c c Hnh ng theo k t qu c a cc l lu n K nng (Skill)
TRI TH C ???
Slide 8

Tri th c (Knowledge)
Tri th c l nh ng thng tin ch a

ng 2 thnh ph n

Cc khi ni m:
Cc khi ni m c b n: l cc khi ni m mang tnh quy c Cc khi ni m pht tri n: c hnh thnh t cc khc ni m c b n thnh cc khi ni m ph c h p ph c t p hn.

Cc phng php nh n th c:
Cc qui lu t, cc th t c Phng php suy di n, l lu n,..

Tri th c l i u ki n tin quy t c a cc hnh x thng minh hay S thng minh Tri th c c c qua s thu th p tri th c v s n sinh tri th c Qu trnh thu th p v s n sinh tri th c l hai qu trnh song song v n i ti p v i nhau khng bao gi ch m d t trong m t th c th Thng Minh

Slide 9

Tri th c Thu th p v s n sinh

Thu th p tri th c:

Tri th c c thu th p t thng tin, l k t qu c a m t qu trnh thu nh n d li u, x l v lu tr . Thng th ng qu trnh thu th p tri th c g m cc b c sau:
Xc nh lnh v c/ph m vi tri th c c n quan tm Thu th p d li u lin quan d i d ng cc tr ng h p c th . H th ng ha, rt ra nh ng thng tin t ng qut, i di n cho cc tr ng h p bi t T ng qut ha. Xem xt v gi l i nh ng thng tin lin quan n v n c n quan tm , ta c cc tri th c v v n .

S n sinh tri th c:

Tri th c sau khi c thu th p s c a vo m ng tri th c c. Trn c s th c hi n cc lin k t, suy di n, ki m ch ng s n sinh ra cc tri th c m i.
Slide 10

Tri th c Tri th c siu c p

Tr th c siu c p (meta knowledge) hay Tri th c v Tri th c

L cc tri th c dng : nh gi tri th c khc nh gi k t qu c a qu trnh suy di n Ki m ch ng cc tri th c m i

Phng ti n truy n tri th c: ngn ng t nhin

Slide 11

Hanh x thong minh Ket luan

Hnh x thng minh khng n thu n l cc hnh ng nh l k t qu c a qu trnh thu th p tri th c v suy lu n trn tri th c. Hnh x thng minh cn bao hm

S tng tc v i mi tr ng nh n cc ph n h i S ti p nh n cc ph n h i i u ch nh hnh ng - Skill S ti p nh n cc ph n h i hi u ch nh v c p nh t tri th c

Tnh ch t thng minh c a m t i t ng l s t ng h p c a c 3 y u t : thu th p tri th c, suy lu n v hnh x c a i t ng trn tri th c thu th p c. Chng ha quy n vo nhau thnh m t th th ng nh t S Thng Minh Khng th nh gi ring l b t k m t kha c nh no ni v tnh thng minh.

THNG MINH C N TRI TH C

Slide 12

1.2

i t ng nghien cu cua AI

AI l lnh v c c a Cng ngh thng tin, c ch c nng nghin

c u v t o ra cc chng trnh m ph ng ho t ng t duy c a con ng i. Tr tu nhn t o nh m t o ra My ng i? M c tiu Xy d ng l thuy t v thng minh gi i thch cc ho t ng thng minh Tm hi u c ch s thng minh c a con ng i

C ch lu tr tri th c C ch khai thc tri th c

Xy d ng c ch hi n th c s thng minh p d ng cc hi u bi t ny vo cc my mc ph c v con ng i.

Slide 13

1.2

i t ng nghien cu cua AI(ti p)

AI l ngnh nghin c u v cch hnh x thng minh (intellgent behaviour) bao g m: thu th p, lu tr tri th c, suy lu n, ho t ng v k nng. i t ng nghin c u l cc hnh x thng minh ch khng ph i l s thng minh. Gi i quy t bi ton b ng AI l tm cch bi u di n tri th c, tm cch v n d ng tri th c gi i quy t v n v tm cch b sung tri th c b ng cch pht hi n tri th c t nh ng thng tin s n c (my h c)

Slide 14

1.3 L ch s pht tri n c a AI : Giai o n c i n

Giai o n c

i n (1950 1965)

C 2 k thu t tm ki m c b n: K thu t generate and test : ch tm c 1 p n/ cha ch c t i u. K thu t Exhaustive search (vt c n): Tm t t c cc nghi m, ch n l a phng n t t nh t.

(Bng n t h p mn v i m>=10)

Slide 15

L ch s pht tri n c a AI : Giai o n vi n vng Giai o n vi n vng (1965 1975)

y l giai o n pht tri n v i tham v ng lm cho my hi u c con ng i qua ngn ng t nhin. Cc cng trnh nghin c u t p trung vo vi c bi u di n tri th c v phng th c giao ti p gi a ng i v my b ng ngn ng t nhin. K t qu khng m y kh quan nhng cng tm ra c cc phng th c bi u di n tri th c v n cn c dng n ngy nay tuy cha th t t t nh:
Semantic Network (m ng ng ngha) Conceptial graph ( th khi ni m) Frame (khung) V p ph i tr ng i v nng l c Script (k ch b n) c a my tnh
Slide 16

L ch s pht tri n c a AI : Giai o n hi n i Giai o n hi n

i (t 1975)

Xc

nh l i m c tiu mang tnh th c ti n hn c a AI:

Tm ra l i gi i t t nh t trong kho ng th i gian ch p nh n c. Khng c u ton tm ra l i gi i t i u

Tinh th n HEURISTIC ra i v c p d ng m nh m kh c ph c bng n t h p. Kh ng nh vai tr c a tri th c ng th i xc nh 2 tr ng i l n l bi u di n tri th c v bng n t h p. Nu cao vai tr c a Heuristic nhng cng kh ng nh tnh kh khn trong nh gi heuristic.
Pht tri n ng d ng m nh m : H chuyn gia, H chu n on,..
Slide 17

Better than nothing

1.4 Cc lnh v c ng d ng
Game Playing: Tm ki m / Heuristic Automatic reasoning & Theorem proving: Tm ki m / Heuristic Expert System: l h ng pht tri n m nh m nh t v c gi tr ng d ng cao nh t. Planning & Robotic: cc h th ng d bo, t ng ha Machine learning: Trang b kh nng h c t p gi i quy t v n kho tri th c:

Supervised : Ki m sot c tri th c h c c. Khng tm ra ci m i. UnSupervised:T h c, khng ki m sot. C th t o ra tri th c m i nhng cng nguy hi m v c th h c nh ng i u khng mong mu n.

Slide 18

1.4 Cc lnh v c ng d ng(ti p)

Natural Language Understanding & Semantic modelling: Khng c pht tri n m nh do m c ph c t p c a bi ton c v tri th c & kh nng suy lu n. Modeling Human perfromance: Nghin c u c ch t ch c tr tu c a con ng i p d ng cho my. Language and Environment for AI:Pht tri n cng c v mi tr ng xy d ng cc ng d ng AI. Neural network / Parallel Distributed processing: gi i quy t v n nng l c tnh ton v t c tnh ton b ng k thu t song song v m ph ng m ng th n kinh c a con ng i.
Slide 19

ng dung AI
M hnh ng d ng AI hi n t i: AI = Presentation & Search
M c d m c tiu t i th ng c a ngnh TTNT l xy d ng m t chi c my c nng l c t duy tng t nh con ng i nhng kh nng hi n t i c a t t c cc s n ph m TTNT v n cn r t khim t n so v i m c tiu ra. Tuy v y, ngnh khoa h c m i m ny v n ang ti n b m i ngy v ang t ra ngy cng h u d ng trong m t s cng vi c i h i tr thng minh c a con ng i. Hnh nh sau s gip b n hnh dung c tnh hnh c a ngnh tr tu nhn t o.

Slide 20

Cc bi ton
Xt cc bi ton sau: 1. i ti n (Vt c n v Heuristic) 2. Tm ki m chi u r ng v su 3. Tic tac toe. 4. ong d u. 5. Bi ton TSP 6. 8 puzzle. 7. C vua 8. C t ng 9. Ng i nng dn qua sng. 10. Con th v con co 11. Con kh v n i chu i
Slide 21

Chng 2: TM KI M TRN KHNG GIAN TR NG THI (State Space Search)

TS. Nguy n nh Thun Khoa Cng ngh Thng tin i h c Nha Trang Email: thuanvinh122@gmail.com

Bi ton tm ki m
Tm ki m ci g? Bi u di n v tm ki m l k thu t ph bi n gi i cc bi ton trong lnh v c AI Cc v n kh khn trong tm ki m v i cc bi ton AI

c t v n ph c t p Khng gian tm ki m l n c tnh i t ng tm ki m thay i p ng th i gian th c Meta knowledge v k t qu t i u

Slide 23

Kh khn v k thu t

C u trc chung c a bi ton tm ki m M t cch chung nh t, nhi u v n -bi ton ph c t p u c d ng "tm ng i trong th " hay ni m t cch hnh th c hn l "xu t pht t m t nh c a m t th , tm ng i hi u qu nh t n m t nh no ". M t pht bi u khc th ng g p c a d ng bi ton ny l:
Cho tr c hai tr ng thi T0 v TG hy xy d ng chu i tr ng thi T0, T1, T2, ..., Tn-1, Tn = TG sao cho :

th a mn m t i u ki n cho tr c (th ng l nh nh t).

Slide 24

2.2 Gi i thu t t ng qut K hi u:

s nh xu t pht g: nh ch n: nh ang xt (n): t p cc nh c th i tr c ti p t nh n Open: t p cc nh c th xt b c k ti p Close: t p cc nh xt

Slide 25

2.2 Gi i thu t t ng qut (ti p)

Begin Open := {s}; Close := ; While (Open <> ) do begin n:=Retrieve(Open); if (n=g) then Return True; Open := Open (n); // ((n) Close) Close := Close {n}; end; Return False; End;
Slide 26

V d : Xt graph sau:
A B E F I J C G D

s = A l nh b t u g= G l nh ch

Slide 27

2.3 Breath First Search V d

Xt graph sau: A B E F I J C G
L nl p n

(n) {B, C, D} {E, F} {F, G} {H, I} {J} True

Open

Close

0 1 2 3 D4 5 6 7

A B C D E F G

{A} {B, C, D} {C, D, E, F} {D, E, F, G} {E, F, G} {F, G, H, I} {G, H, I, J}

{A} {A, B} {A, B, C} {A, B, C, D} {A, B, C, D, E} {A, B, C, D, E, F}

Slide 28

2.3 Breath First Search V d 1

Xt graph sau:A->U A B E F I J C G
L nl p n

(n)

Open

Close

0 1 2 3 D4 5 6 7 8 9 10

{A} A {B, C, D} {B,C,D} B {E, F} {C,D, E,F} C {F, G} {D,E, F,G} D {E, F, G} E {H, I} {F, G, H, I} F {J} {G, H, I, J} G {H, I, J} H {I, J} I {J} J FALSE

{A} {A, B} {A, B, C} {A, B, C, D} {A, B, C, D, E} {A, B, C, D, E, F} {A, B, C, D, E, F,G} {A,B,C, D, E, F,G,H} {A,B,C, D, E, F,G,H,I} {A,B,C, D, E, F,G,H,I,J}

Slide 29

V d : Xt graph sau:
A B E F I J C G D

Slide 30

2.4 Depth First Search V d

Xt graph sau: A B E F I J C G
L nl p n

(n)

Open

Close

0 1 2 3 4 5 6 7 8 9

{A} A {B, C, D} {B, C, D} B {E, F} {E, F, C, D} {H, I, F, C, D} DE {H, I} H {I, F, C, D} I {F, C, D} F {J} {J, C, D} J {C, D} C {F, G} {G, D} G True

{A} {A, B} {A, B, E} {A, B, E, H} {A, B, E, H, I} {A, B, E, H, I, F} {A, B, E, H, I, F,J} {A,B,E,H,I, F,J,C}

Slide 31

Breath First vs Depth First

Breath First: open c t ch c d ng FIFO Depth First: open c t ch c d ng LIFO Hi u qu

Breath First lun tm ra nghi m c s cung nh nh t Depth First th ng cho k t qu nhanh hn. BFS, DFS ch c ch n tm ra k t qu n u c.

K t qu

Bng n t h p l kh khn l n nh t cho cc gi i thu t ny. Gi i Php cho bng n t h p??

Slide 32

Tm Ki m R ng
1. 2. 2. 3. 4. 5. 6.

Open = [A]; closed = [] Open = [B,C,D]; closed = [A] Open = [C,D,E,F]; closed = [B,A] Open = [D,E,F,G,H]; closed = [C,B,A] Open = [E,F,G,H,I,J]; closed = [D,C,B,A] Open = [F,G,H,I,J,K,L];closed = [E,D,C,B,A] Open = [G,H,I,J,K,L,M];(v L c trong open); closed = [F,E,D,C,B,A]

Slide 33

Tm ki m Su
1. 2. 3. 4. 5. 6. 7. 8. 9.

Open = [A]; closed = [] Open = [B,C,D]; closed = [A] Open = [E,F,C,D];closed = [B,A] Open = [K,L,F,C,D]; closed = [E,B,A] Open = [S,L,F,C,D]; closed = [K,E,B,A] Open = [L,F,C,D]; closed = [S,K,E,B,A] Open = [T,F,C,D]; closed = [L,S,K,E,B,A] Open = [F,C,D]; closed = [T,L,S,K,E,B,A]

Slide 34

Depth first search c gi i h n

Depth first search c kh nng l p v t n do cc tr ng thi con sinh ra lin t c. su tng v t n. Kh c ph c b ng cch gi i h n su c a gi i thu t. Su bao nhiu th v a? Chi n l c gi i h n:

C nh m t su MAX, nh cc danh th chi c tnh tr c c s n c nh t nh Theo c u hnh resource c a my tnh Meta knowledge trong vi c nh gi i h n su.

Gi i h n su => co h p khng gian tr ng thi => c th m t nghi m.

Slide 35

Chng 3: HEURISTIC SEARCH

TS. Nguy n nh Thun Khoa Cng ngh Thng tin i h c Nha Trang Email: thuanvinh122@gmail.com

3.1 Gi i thi u v Heuristic

Heuristic l g?

Heuristic l nh ng tri th c c rt t a t nh ng kinh nghi m, tr c gic c a con ng i. Heuristic c th l nh ng tri th c ng hay sai. Heuristic l nh ng meta knowledge v th ng ng.

Heuristic dng

lm g?

Trong nh ng bi ton tm ki m trn khng gian tr ng thi, c 2 tr ng h p c n n heuristic:

V n c th khng c nghi m chnh xc do cc m nh khng pht bi u ch t ch hay thi u d li u kh ng nh k t qu . V n c nghi m chnh xc nhng ph t n tnh ton tm ra nghi m l qu l n (h qu c a bng n t h p)

Heuristic gip tm ki m

t k t qu v i chi ph th p hn
Slide 37

Heuristic (ti p)
Thu t gi i Heuristic l m t s m r ng khi ni m thu t ton. N th hi n cch gi i bi ton v i cc c tnh sau:

Th ng tm c l i gi i t t (nhng khng ch c l l i gi i t t nh t) Gi i bi ton theo thu t gi i Heuristic th ng d dng v nhanh chng a ra k t qu hn so v i gi i thu t t i u, v v y chi ph th p hn. Thu t gi i Heuristic th ng th hi n kh t nhin, g n gi v i cch suy ngh v hnh ng c a con ng i.
Slide 38

Heuristic (ti p)
C nhi u phng php xy d ng m t thu t gi i Heuristic, trong ng i ta th ng d a vo m t s nguyn l c b n nh sau:
Nguyn l vt c n thng minh: Trong m t bi ton tm ki m no , khi khng gian tm ki m l n, ta th ng tm cch gi i h n l i khng gian tm ki m ho c th c hi n m t ki u d tm c bi t d a vo c th c a bi ton nhanh chng tm ra m c tiu. Nguyn l tham lam (Greedy): L y tiu chu n t i u (trn ph m vi ton c c) c a bi ton lm tiu chu n ch n l a hnh ng cho ph m vi c c b c a t ng b c (hay t ng giai o n) trong qu trnh tm ki m l i gi i. Nguyn l th t : Th c hi n hnh ng d a trn m t c u trc th t h p l c a khng gian kh o st nh m nhanh chng t c m t l i gi i t t. Hm Heuristic: Trong vi c xy d ng cc thu t gi i Heuristic, ng i ta th ng dng cc hm Heuristic. l cc hm nh gi th, gi tr c a hm ph thu c vo tr ng thi hi n t i c a bi ton t i m i b c gi i. Nh gi tr ny, ta c th ch n c cch hnh ng tng i h p l trong t ng b c c a thu t gi i.
Slide 39

Heuristic Greedy
Bi ton i ti n: i s ti n n thnh cc lo i ti n cho tr c sao cho s t l t nh t Bi ton hnh trnh ng n nh t (TSP): Hy tm m t hnh trnh cho m t ng i giao hng i qua n i m khc nhau, m i i m i qua m t l n v tr v i m xu t pht sao cho t ng chi u di o n ng c n i l ng n nh t. Gi s r ng c con ng n i tr c ti p t gi a hai i m b t k.

Vt c n: (n-1)! (V i n l n ???) Greedy 1: M i b c ch n i j sao cho j g n i nh t trong nh ng nh n i v i i cn l i Greedy 2: M i b c ch n i j sao cho i g n j nh t trong nh ng nh n i v i j cn l i
Slide 40

V d : TSP v i n=8
1 1 2 3 4 5 6 7 8 0 730 640 840 800 430 380 1010 2 730 0 710 1040 500 300 540 470 3 640 710 0 1420 1050 600 920 1160 4 840 1040 1420 0 740 950 570 900 5 800 500 1050 740 0 520 460 200 6 430 300 600 950 520 0 390 690 7 380 540 920 570 460 390 0 660 8 1010 470 1160 900 200 690 660 0
Slide 41

V d : TSP v i n=8
*V i Greedy 1: 1 7 6 2 8 5 4 3 1 T ng chi ph: 4540 *V i Greedy 2: 1 7 4 5 8 2 6 3 1 T ng chi ph: 3900

Bi ton 3: Bi ton t mu b n

Slide 42

Heuristic (tt)
Heuristic dng nh th no trong tm ki m?

Tm ki m trn khng gian tr ng thi theo chi u no? Su hay r ng? Tm theo Heuristic : Heuristic nh h ng qu trnh tm ki m theo h ng m n cho r ng kh nng t t i nghi m l cao nh t. Khng su cng khng r ng Vi c tm ki m theo nh h ng c a heuristic c k t qu t t hay x u ty theo heuristic ng hay sai. Heuristic c kh nng b st nghi m Heuristic cng t t cng d n n k t qu nhanh v t t. Lm sao tm c Heuristic t t???
Slide 43

K t qu c a tm ki m v i Heuristic

3.2 Tm ki m t i u (Best First Search)

OPEN : t p ch a cc tr ng thi c sinh ra nhng cha c xt n (v ta ch n m t tr ng thi khc). Th c ra, OPEN l m t lo i hng i u tin (priority queue) m trong , ph n t c u tin cao nh t l ph n t t t nh t. CLOSE : t p ch a cc tr ng thi c xt n. Chng ta c n lu tr nh ng tr ng thi ny trong b nh phng tr ng h p khi m t tr ng thi m i c t o ra l i trng v i m t tr ng thi m ta xt n tr c .
Thu t gi i BEST-FIRST SEARCH

1. t OPEN ch a tr ng thi kh i u. 2. Cho n khi tm c tr ng thi ch ho c khng cn nt no trong OPEN, th c hi n : 2.a. Ch n tr ng thi t t nh t (Tmax) trong OPEN (v xa Tmax kh i OPEN) 2.b. N u Tmax l tr ng thi k t thc th thot. 2.c. Ng c l i, t o ra cc tr ng thi k ti p Tk c th c t tr ng thi Tmax. i v i m i tr ng thi k ti p Tk th c hi n : Tnh f(Tk); Thm Tk vo OPEN

Slide 44

3.2 Tm ki m t i u (ti p)
Thu t gi i BEST-FIRST SEARCH
Begin Open := {s}; Close := ; While (Open <> ) do begin n:=Retrieve(Open); if (n=g) then Return True; Open := Open (n); // ((n) Close) Close := Close {n}; end; Return False;

Begin open:={s}; While (open<> ) do begin n:= Retrieve(Open) //Ch n tr ng thi t t nh t t Open. if (n=g) then return True else begin T o (n) for m i nt con m c a (n) do Gn gi tr chi ph cho m Open:=Open{m}; end; End; Return False; End;

Slide 45

3.2 Tm ki m t i u (ti p)
- BFS kh n gi n. Tuy v y, trn th c t , cng nh tm ki m chi u su v chi u r ng, hi m khi ta dng BFS m t cch tr c ti p. Thng th ng, ng i ta th ng dng cc phin b n c a BFS l AT, AKT v A*

Thng tin v qu kh v tng lai

- Thng th ng, trong cc phng n tm ki m theo ki u BFS, t t f c a m t tr ng thi c tnh d a theo 2 hai gi tr m ta g i l l g v h. h chng ta bi t, l m t c l ng v chi ph t tr ng thi hi n hnh cho n tr ng thi ch (thng tin tng lai). Cn g l "chi u di qung ng" i t tr ng thi ban u cho n tr ng thi hi n t i (thng tin qu kh ). Lu r ng g l chi ph th c s (khng ph i chi ph c l ng).

Slide 46

3.3 Thu t gi i AT

Phn bi t khi ni m g v h

Slide 47

3.3 Thu t gi i AT
Thu t gi i AT l m t phng php tm ki m theo ki u BFS v i t t c a nt l gi tr hm g t ng chi u di con ng i t tr ng thi b t u n tr ng thi hi n t i.

Begin open:={s}; While (open<> ) do begin n:= Retrieve(Open) //Ch n n sao cho g(n) nh nh t t Open. if (n=g) then return True else begin T o (n) for m i nt con m c a (n) do if (mOpen) then Begin g(m):=g(n)+Cost(n,m) Open:=Open{m}; end else So snh g(m) va gNew (m) v c p nh t end; Return False; End;

Slide 48

3.3 Thu t gi i CMS (Cost Minimazation Search)

Thu t gi i CMS l m t phng php tm ki m theo ki u BFS v i nt l gi tr hm g v b sung t p Close: t p nh xt). t tc a

Begin open:={s}; close := While (open<> ) do begin n:= Retrieve(Open) //Ch n n sao cho g(n) nh nh t t Open. if (n=g) then return True else begin T o (n) for m i nt con m c a (n) do if (mOpen) and (mClose) then Begin g(m):=g(n)+Cost(n,m) Open:=Open{m}; end else So snh g(m) va gNew (m) v c p nh t close = close {n} end; Return False; End;

Slide 49

V d :
Xt graph sau: A

20
B

30 35 15
C D

40
E

45
F

10
G

30
H

25
I

10
J

20 u
Slide 50

s = A l nh b t g= J l nh ch

V d :
Xt graph sau: Tr c Sau * A A B A C A D B E C F C G F J
L nl p 0 1 A B D C G {B,C,D} {E,F} {F,G} {J} (J} n (n) Open {(A,0)} {(B,20), (C,35), (D,30)} (C,35), (D,30),(E,60),(F,65) (C,35),(E,60),(F,65) (E,60),(F,50),(G,45) (E,60),(F,50),(J,65) (E,60),(J,60)

s = A l nh b t g= J l nh ch

F J

Slide 51

3.4 Thu t gi i AKT

(Algorithm for Knowlegeable Tree Search) Thu t gi i AKT m r ng AT b ng cch s d ng thm thng tin c l ng h. c a m t tr ng thi f l t ng c a hai hm g v h. t t

Begin open:={s}; While (open<> ) do begin n:= Retrieve(Open) //Ch n n sao cho f(n) nh nh t t Open. if (n=g) then return True else begin T o (n) for m i nt con m c a (n) do Begin g(m):=g(n)+Cost(n,m) f(m):= g(m)+h(m); Open:=Open{m}; end; end; Return False; End;

Slide 52

3.5 Thu t gi i A*

Thu t gi i A*
A* l m t phin b n c bi t c a AKT p d ng cho tr ng h p th . Thu t gi i A* c s d ng thm t p h p CLOSE lu tr nh ng tr ng h p c xt n. A* m r ng AKT b ng cch b sung cch gi i quy t tr ng h p khi "m " m t nt m nt ny c s n trong OPEN ho c CLOSE.

Slide 53

3.5 Thu t gi i A* (ti p)

Begin open:={s}; close:=; While (open<> ) do begin n:= Retrieve(Open) //sao cho f(n) min. if (n=g) then return path t s n g else begin T o (n) for m i nt con m c a (n) do case m of m Open v m Close: begin Gn gi tr heuristic cho m Open:=Open{m}; end; m Open:
if n c m b ng m t path ng n hn then C p nh t l i m trong Open. m Close if n c m b ng m t path ng n hn then begin Close:=Close-{m}

Open:=Open{m}
end; end; /*end case*/ Close:=Close{n} end; / while/ return false; End;

Slide 54

Hm l ng gi Heuristic
Hm l ng gi Heuristic l hm c l ng ph t n i t tr ng thi hi n t i n tr ng thi goal. C s xc nh hm l ng gi l d a vo tri th c/kinh nghi m thu th p c. Hm l ng gi cho k t qu ng (g n th c th ) hay sai (xa gi tr th c) s d n n k t qu tm c t t hay x u. Khng c chu n m c cho vi c nh gi m t hm l ng gi Heuristic. L do:

Khng c c u trc chung cho hm l ng gi Tnh ng/sai thay i lin t c theo t ng v n c th Tnh ng/sai thay i theo t ng tnh hu ng c th trong m t v n

C th dng nhi u hm l ng gi khc nhau theo tnh hu ng c n hm l ng gi v cc hm l ng gi.

Slide 55

Tr
Tr 15

8 hay 15
u Tr ng thi ch
1 12 11 10 9 2 13 3 14 15 8 4 5 6 7

Tr ng thi ban
11 10 1 9 14 6 2 12 13 8 4 7 5 15 3

Tr 8

2 3 6 5 2

8 7 1

1 8 7

3 4

C n bi u di n KGTT cho bi ton ny nh th no?

C 3 Tm ki m khng gian tr ng thi

Slide 56

Thu t gi i A* V d
l

Xt bi ton 8 pussle v i goal l:

1 2 3 8 4 7 6 5

2 8 3 1 6 4 7 5 2 8 3 1 4

Heuristic 1: T ng s mi ng sai v tr Heuristic 2: T ng kho ng cch sai v tr c a t ng mi ng.

7 6 5 2 8 3 1 6 4 7 5

Vi c ch n l a hm Heuristic l kh khn v c ngha quy t nh i v i t c c a gi i thu t

Slide 57

Hm l ng gi Heuristic C u trc
Xt l i ho t

ng c a gi i thu t Best First Search:

Khi c 2 nt cng c gi tr k v ng t n m c tiu b ng nhau th nt c path t nt b t u n nt ng n hn s c ch n tr c nh v y nt ny c gi tr Heuristic t t hn. Hay ni cch khc hm l ng gi Heuristic cho nt g n start hn l t t hn n u k v ng n goal l b ng nhau. V y ch n nt no n u k v ng c a 2 nt khc nhau? Nt k v ng t t hn nhng xa start hay nt k v ng x u hn nhng g n root

Hm l ng gi bao g m c 2 v c c u trc: F(n) := G(n) + H(n) G(n): ph t n th c t root n n H(n): ph t n c lu ng heuristic t n

n goal.

Slide 58

Thu t gi i A* V d
l

Xt v d l bi ton 8 puzzle v i:
2 8 3 1 6 4 7 5 1 2 3 8 4 7 6 5

B t

ch

Hm l ng gi: F(n) = G(n) + H(n) V i G(n): s l n chuy n v tr th c hi n H(n): S mi ng n m sai v tr Nt X c gi tr heuristic t t hn nt Y n u F(x) < F(y). Ta c ho t ng c a gi i thu t Best First search trn nh hnh sau:
Slide 59

3.5 Thu t gi i A* (ti p)

Begin open:={s}; close:=; While (open<> ) do begin n:= Retrieve(Open) //sao cho f(n) min. if (n=g) then return path t s n g else begin T o (n) for m i nt con m c a (n) do case m of m Open v m Close: begin Gn gi tr heuristic cho m Open:=Open{m}; end; m Open:
if n c m b ng m t path ng n hn then C p nh t l i m trong Open. m Close if n c m b ng m t path ng n hn then begin Close:=Close-{m}

Open:=Open{m}
end; end; /*end case*/ Close:=Close{n} end; / while/ return false; End;

Slide 60

V d
1
2 1 7 8 6 3 4 5

State A F(a) =0+4=4

3 4 5

2 1

8 6 7

3 4 5

State B F(b) =1+5=6

2 1 7

8 6

State C F(c) =1+3=4

2 1 7

8 6 5

3 4

State D F(D) =1+5=6

2 7

8 1 6

3 4 5

State E F(e) =2+3=5

2 1 7 8 6

3 4 5

State F

2 1 7

8 4 6

3 5

State G F(g) =2+4=6

2 1 7

8 6

3 4 5

F(f) =2+3=5

x
2 7

8 1 6

3 4 5

State H F(h) =3+3=6

2 7

8 1 6

3 4 5

State I F(i) =3+4=7

2 1 7

8 6

3 4 5

Slide 61

V d
4
2 1 7 8 6 3 4 5

State F F(f) =2+3=5

3 8 6 4 5

5
1 7

2 8 6

3 4 5

State J F(j) =3+2=5

2 1 7

State K F(k) =3+4=7

2 1 7

8 6

3 4 5

State Close

1 7

2 8 6

3 4 5

State L F(l) =4+1=5

2 1 7 8 6

3 4 5

Close

y
1 7

2 8 6

3 4 5

State Close

1 8 7

2 6

3 4 5

State M F(m) =5+0=5

1 7

2 8 6

3 4 5

State N F(n) =5+1=7

Slide 62

C 3 Tm ki m khng gian tr ng thi Slide 63

The 8-puzzle searched by a production system with loop detection and depth bound 5

Tr chi 8-puzzle

Ho t
L n n

ng theo gi i thu t A*
Open Close

0 1 2 3 4 5 6 7

A4 C4 E5 F5 J5 l5 m5

{A4} {C4,B6,D6} {E5,F5,G6,B6,D6} {F5,H6,G6,B6,D6,I7} {J5,H6,G6,B6,D6,K7,I7} {L5,H6,G6,B6,D6,K7,I7} {M5,H6,G6,B6,D6,K7,I7,N7}

{} {A4} {A4,C4} {A4,C4,E5} {A4,C4,E5,F5} {A4,C4,E5,F5,J5} {A4,C4,E5,F5,J5,L5}

Slide 64

nh gi gi i thu t Heuristic
Admissibility Tnh ch p nh n

M t gi i thu t Best first search v i hm nh gi F(n) = G(n) + H(n) v i

N : Tr ng thi b t k G(n) : Ph t n i t nt b t u n nt n H(n) : Ph t n c l ng heuristic i t nt n

n goal

c g i l gi i thu t A

M t gi i thu t tm ki m c xem l admissible n u i v im t th b t k n lun d ng path nghi m t t nh t (n u c). Gi i thu t A*: L gi i thu t A v i hm heuristic H(n)lun lun gi tr th c i t n n goal. Gi i thu t A* l admissible Slide 65

nh gi gi i thu t Heuristic
Monotonicity n i u

M t hm heuristic H(n) c g i l monotone ( n i u) n u: ni, nj : nj l nt con chu c a ni ta c H(ni)-H(nj) ph t n th t i t ni n nj nh gi heuristic c a ch l 0 : H(goal) = 0.

Gi i thu t A c hm H(n) monotone l gi i thu t A* v Admissible Informedness Xt 2 hm heuristic H1(n) v H2(n) n u ta c H1(n) H2(n) v i m i tr ng thi n th H2(n) c cho l informed hn H1(n).
Slide 66

Heuristic trong tr chi

Gi i thu t minimax:

i khng

Hai u th trong tr chi c g i l MIN v MAX. M i nt l c gi tr :

1 n u l MAX th ng, 0 n u l MIN th ng.

Minimax s truy n cc gi tr ny ln cao d n trn nt cha m k ti p theo cc lu t sau:

th , qua cc

N u tr ng thi cha m l MAX, gn cho n gi tr l n nh t c trong cc tr ng thi con. N u tr ng thi b , m l MIN, gn cho n gi tr nh nh t c trong cc tr ng thi con.

C 4 Tm ki m Heuristic

Slide 67

Hy p d ng GT Minimax vo Tr Chi NIM

C 4 Tm ki m Heuristic

Slide 68

Minimax v i
Minimax

su l p c
nh.

nh

i v i m t KGTT gi

Cc nt l c gn cc gi tr heuristic Cn gi tr t i cc nt trong l cc gi tr nh n c d a trn gi i thu t Minimax

C 4 Tm ki m Heuristic Slide 69

Heuristic trong tr chi tic-tac-toe

Hm Heuristic: E(n) = M(n) O(n) Trong : M(n) l t ng s ng th ng c th c a ti O(n) l t ng s ng th ng c th c a i th E(n) l tr s nh gi t ng c ng cho tr ng thi n
C 4 Tm ki m Heuristic Slide 70

Minimax 2 l p trong tic-tac-toe

C 4 Tm ki m Heuristic

Trch t Nilsson (1971).

Slide 71

Gi i thu t c t t a -
Tm ki m theo ki u depth-first. Nt MAX c 1 gi tr (lun tng) Nt MIN c 1 gi tr (lun gi m) TK c th k t thc d i b t k:

Nt MIN no c c a b t k nt cha MAX no. Nt MAX no c c a b t k nt cha MIN no.

Gi i thu t c t t a - th hi n m i quan h gi a cc nt l p n v n+2, m t i ton b cy c g c t i l p n+1 c th c t b .

C 4 Tm ki m Heuristic

Slide 72

C tta
MAX S= A=

MIN

Z - cut =z z

C 4 Tm ki m Heuristic

Slide 73

C tta
MIN S = A= Z - cut =z z

MAX

C 4 Tm ki m Heuristic

Slide 74

GT C t T a - p d ng cho KGTT gi

nh

Cc nt khng c gi tr l cc nt khng c duy t qua

C 4 Tm ki m Heuristic

Slide 75

Chng 4: Bi u di n v suy lu n tri th c

TS. Nguy n nh Thun Khoa Cng ngh Thng tin i h c Nha Trang Email: thuanvinh122@gmail.com

4.1. M

tri th c, lnh v c v bi u di n tri th c.

4.2. Cc lo i tri th c: c chia thnh 5 lo i

1.

2.

Tri th c th t c: m t cch th c gi i quy t m t v n . Lo i tri th c ny a ra gi i php th c hi n m t cng vi c no . Cc d ng tri th c th t c tiu bi u th ng l cc lu t, chi n l c, l ch trnh v th t c. Tri th c khai bo: cho bi t m t v n c th y nh th no. Lo i tri th c ny bao g m cc pht bi u n gi n, d i d ng cc kh ng nh logic ng ho c sai. Tri th c khai bo cng c y hn th l m t danh sch cc kh ng nh nh m m t v i t ng hay m t khi ni m no .
Slide 77

4.2. Cc
3.

lo i tri th c (ti p)

Siu tri th c:

m t tri th c v tri th c. Lo i tri th c ny gip l a ch n tri th c thch h p nh t trong s cc tri th c khi gi i quy t m t v n . Cc chuyn gia s d ng tri th c ny i u ch nh hi u qu gi i quy t v n b ng cch h ng cc l p lu n v mi n tri th c c kh nng hn c .

4.

Tri th c heuristic: m t

cc "m o" d n d t ti n trnh l p lu n. Tri th c heuristic l tri th c khng b m m hon ton 100% chnh xc v k t qu gi i quy t v n . Cc chuyn gia th ng dng cc tri th c khoa h c nh s ki n, lu t, sau chuy n chng thnh cc tri th c heuristic thu n ti n hn trong vi c gi i quy t m t s bi ton.

5.

tri th c theo c u trc. Lo i tri th c ny m t m hnh t ng quan h th ng theo quan i m c a chuyn gia, bao g m khi ni m, khi ni m con, v cc i t ng; di n t ch c nng v m i lin h gi a cc tri th c d a theo c u trc xc nh.

Tri th c c c u trc: m t

Slide 78

V d : Hy phn lo i cc tri th c sau

1. 2. 3. 4. 5. 6. 7. 8. 9. Nha Trang l thnh ph p. B n Lan thch c sch. Thu t ton tm ki m BFS, DFS Thu t gi i Greedy M t s cch chi u t ng trong vi c chi c t ng. H th ng cc khi ni m trong hnh h c. Cch t p vi t ch p. Tm t t quy n sch v Tr tu nhn t o. Ch n lo i c phi u mua c phi u.

Slide 79

4.3. CC K THU T BI U DI N TRI TH C

4.3.1 B ba i t ng-Thu c tnh-Gi tr 4.3.2 Cc lu t d n 4.3.3 M ng ng ngha 4.3.4 Frames 4.3.5 Logic

Slide 80

4.3.1 B ba

i t ng-Thu c tnh-Gi tr

M t s ki n c th c dng xc nh n gi tr c a m t thu c tnh xc nh c a m t vi i t ng. V d , m nh "qu bng mu " xc nh n " " l gi tr thu c tnh "mu" c a i t ng "qu bng". Ki u s ki n ny c g i l b ba i t ng-Thu c tnh-Gi tr (O-A-V Object-AttributeValue).

Hnh 2.1. Bi u di n tri th c theo b ba O-A-V

Slide 81

4.3.1 B ba

i t ng-Thu c tnh-Gi tr (ti p)

Trong cc s ki n O-A-V, m t i t ng c th c nhi u thu c tnh v i cc ki u gi tr khc nhau. Hn n a m t thu c tnh cng c th c m t hay nhi u gi tr . Chng c g i l cc s ki n n tr (single-valued) ho c a tr (multi-valued). i u ny cho php cc h tri th c linh ng trong vi c bi u di n cc tri th c c n thi t. Cc s ki n khng ph i lc no cng b o m l ng hay sai v i ch c ch n hon ton. V th , khi xem xt cc s ki n, ng i ta cn s d ng thm m t khi ni m l tin c y. Phng php truy n th ng qu n l thng tin khng ch c ch n l s d ng nhn t ch c ch n CF (certainly factor). Khi ni m ny b t u t h th ng MYCIN (kho ng nm 1975), dng tr l i cho cc thng tin suy lu n. Khi , trong s ki n O-A-V s c thm m t gi tr xc nh tin c y c a n l CF.
Slide 82

4.3.2 Cc lu t d n
Lu t l c u trc tri th c dng lin k t thng tin bi t v i cc thng tin khc gip a ra cc suy lu n, k t lu n t nh ng thng tin bi t. Trong h th ng d a trn cc lu t, ng i ta thu th p cc tri th c lnh v c trong m t t p v lu chng trong c s tri th c c a h th ng. H th ng dng cc lu t ny cng v i cc thng tin trong b nh gi i bi ton. Vi c x l cc lu t trong h th ng d a trn cc lu t c qu n l b ng m t module g i l b suy di n.

Slide 83

4.3.2 Cc lu t d n(ti p) Cc d ng lu t c b n: 7 d ng
1. Quan h :
IF Bnh i n h ng THEN Xe s khng kh i ng c

2. L i khuyn:
IF Xe khng kh i THEN i b ng c

3. H ng d n
IF Xe khng kh i ng c AND H th ng nhin li u t t THEN Ki m tra h th ng i n

Slide 84

4.3.2 Cc lu t d n(ti p)
4. Chi n l c
IF Xe khng kh i ng c THEN u tin hy ki m tra h th ng nhin li u, sau ki m tra h th ng i n

5. Di n gi i
IF Xe n AND ti ng gin THEN ng c ho t ng bnh th ng

6. Ch n on
IF S t cao AND hay ho AND H ng THEN Vim h ng

7. Thi t k
IF L n AND Da sng THEN Nn ch n Xe Spacy AND Ch n mu sng

Slide 85

4.3.3 M ng ng

ngha

M ng ng ngha l m t phng php bi u di n tri th c dng th trong nt bi u di n i t ng v cung bi u di n quan h gi a cc i t ng.

Hnh 2.3. "S l Chim" th hi n trn m ng ng ngha

Slide 86

4.3.3 M ng ng

ngha(ti p)

Hnh 4.4. Pht tri n m ng ng ngha

Slide 87

V d : Gi i bi ton tam gic t ng qut

C 22 y u t c a tam gic. Nh v y c C322 -1 cch xy d ng hay xc nh m t tam gic. Theo th ng k, c kho ng 200 cng th c lin quan n c nh v gc 1 tam gic. gi i bi ton ny b ng cng c m ng ng ngha, s d ng kho ng 200 nh ch a cng th c v kho ng 22 nh ch a cc y u t c a tam gic. M ng ng ngha cho bi ton ny c c u trc nh sau : nh c a th bao g m hai lo i : nh ch a cng th c (k hi u b ng hnh ch nh t) nh ch a y u t c a tam gic (k hi u b ng hnh trn) Cung : ch n i t nh hnh trn n nh hnh ch nh t cho bi t y u t tam gic xu t hi n trong cng th c no * Lu : trong m t cng th c lin h gi a n y u t c a tam gic, ta gi nh r ng n u bi t gi tr c a n-1 y u t th s tnh c gi tr c a y u t cn l i. Ch ng h n nh trong cng th c t ng 3 gc c a tam gic b ng 1800 th khi bi t c hai gc, ta s tnh c gc cn l i.
Slide 88

V d : Gi i bt tam gic t ng qut (tt) B1 : Kch ho t nh ng nh hnh trn cho ban u (nh ng y u t c gi tr ) B2 : L p l i b c sau cho n khi kch ho t c t t c nh ng nh ng v i nh ng y u t c n tnh ho c khng th kch ho t c b t k nh no n a. N u m t nh hnh ch nh t c cung n i v i n nh hnh trn m n-1 nh hnh trn c kch ho t th kch ho t nh hnh trn cn l i (v tnh gi tr nh cn l i ny thng qua cng th c nh hnh ch nh t).
Slide 89

V d : Gi i bt tam gic t ng qut (tt)

V d : "Cho hai gc , v chi u di c nh a c a tam gic. Tnh chi u di ng cao hC".

p=(a+b+c)/2

Slide 90

4.3.4 Frame

Hnh 2.6. C u trc frame

Hnh 2.7. Nhi u m c c a frame m t quan h ph c t p hn

Slide 91

4.3.5 Logic
1. Logic m nh
IF Xe khng kh i ng c (A) AND Kho ng cch t nh n ch lm l xa (B) THEN S tr gi lm (C) Lu t trn c th bi u di n l i nh sau:AB C

2. Logic v t
Logic v t , cng gi ng nh logic m nh , dng cc k hi u th hi n tri th c. Nh ng k hi u ny g m h ng s , v t , bi n v hm.
Slide 92

4.4 SUY DI N D
1. Modus ponens
1. E1 2. E1 E2 3. E2 N u c tin 1. E2 2. E1 E2 3. E1

LI U

khc, c d ng E2 E3 th E3 c a vo danh sch.

2. Modus tollens

Slide 93

4.5 Ch ng minh m nh
M t trong nh ng v n kh quan tr ng c a logic m nh l ch ng minh tnh ng n c a php suy di n (a b). y cng chnh l bi ton ch ng minh th ng g p trong ton h c. V i hai php suy lu n c b n c a logic m nh (Modus Ponens, Modus Tollens) c ng v i cc php bi n i hnh th c, ta cng c th ch ng minh c php suy di n. Tuy nhin, thao tc bi n i hnh th c l r t kh ci t c trn my tnh. Th m ch i u ny cn kh khn v i c con ng i! V i cng c my tnh, c th cho r ng ta s d dng ch ng minh c m i bi ton b ng m t phng php bi t l l p b ng chn tr . Tuy v l thuy t, phng php l p b ng chn tr lun cho c k t qu cu i cng nhng ph c t p c a phng php ny l qu l n, O(2n) v i n l s bi n m nh . Sau y chng ta s nghin c u hai phng php ch ng minh m nh v i ph c t p ch c O(n).

Slide 94

4.5 Ch ng minh m nh
M t trong nh ng v n kh quan tr ng c a logic m nh l ch ng minh tnh ng n c a php suy di n (a b). y cng chnh l bi ton ch ng minh th ng g p trong ton h c. V i hai php suy lu n c b n c a logic m nh (Modus Ponens, Modus Tollens) c ng v i cc php bi n i hnh th c, ta cng c th ch ng minh c php suy di n. Tuy nhin, thao tc bi n i hnh th c l r t kh ci t c trn my tnh. Th m ch i u ny cn kh khn v i c con ng i! V i cng c my tnh, c th cho r ng ta s d dng ch ng minh c m i bi ton b ng m t phng php bi t l l p b ng chn tr . Tuy v l thuy t, phng php l p b ng chn tr lun cho c k t qu cu i cng nhng ph c t p c a phng php ny l qu l n, O(2n) v i n l s bi n m nh . Sau y chng ta s nghin c u hai phng php ch ng minh m nh v i ph c t p ch c O(n).

Slide 95

4.5.1 Thu t gi i Vng H o

B1 : Pht bi u l i gi thi t v k t lu n c a v n theo d ng chu n sau : GT1, GT2, ..., GTn KL1, KL2, ..., KLm Trong cc GTi v KLi l cc m nh c xy d ng t cc bi n m nh php n i c b n : , , nh. B2 : Chuy n v cc GTi v KLi c d ng ph V d : p q, (r s), g, p r s, p p q, p r , p (r s), g, s B3 : N u GTi c php th thay th php b ng d u "," N u KLi c php th thay th php b ng d u "," V d : p q , r ( p s) q, s p, q, r, p s q, s

v 3

Slide 96

4.5.1 Thu t gi i Vng H o

B4 : N u GTi c php th tch thnh hai dng con. N u KLi c php th tch thnh hai dng con. V d : p, p q q p, p q v p, q q B5 : M t dng c ch ng minh n u t n t i chung m t m nh c hai pha. V d : p, q q c ch ng minh p, p q p p, q B6 : a) N u m t dng khng cn php n i v php n i c hai v v 2 v khng c chung m t bi n m nh th dng khng c ch ng minh. b) M t v n c ch ng minh n u t t c dng d n xu t t d ng chu n ban u u c ch ng minh. V d : i) p ( p q) q ii) (p q) ( p r) q r
Slide 97

4.5.2 Thu t gi i Robinson

Thu t gi i ny ho t ng d a trn phng php ch ng minh ph n ch ng v php h p gi i Robinson. Phng php ch ng minh ph n ch ng:

Ch ng minh php suy lu n (a b) l ng (v i a l gi thi t, b l k t lu n). Ph n ch ng : gi s b sai suy ra b l ng. i) p ( p q) q ii) (p q) ( p r) q r

Php h p gi i Robinson:

Bi ton c ch ng minh n u a ng v b ng sinh ra m t mu thu n.

Slide 98

4.5.2 Thu t gi i Robinson (ti p)

B1 : Pht bi u l i gi thi t v k t lu n c a v n nh sau : d i d ng chu n

Trong : GTi v KLj c xy d ng t cc bi n m nh v cc php ton : , , B2 : N u GTi c php th thay b ng d u "," N u KLi c php th thay b ng d u "," B3 : Bi n i dng chu n B1 v thnh danh sch m nh nh sau : B4 : N u trong danh sch m nh b c 2 c 2 m nh i ng u nhau th bi ton c ch ng minh. Ng c l i th chuy n sang B5. (a v a g i l hai m nh i ng u nhau)
{ GT1, GT2, ..., GTn , KL1, KL2, ..., KLm }

GT1, GT2, ..., GTn KL1, KL2, ..., KLm

Slide 99

4.5.2 Thu t gi i Robinson (ti p)

B6 : p d ng php h p gi i i) p ( p q) q ii) (p q) ( p r) q r B7 : N u khng xy d ng c thm m t m nh m i no v trong danh sch m nh khng c 2 m nh no i ng u nhau th v n khng c ch ng minh. V d : Ch ng minh r ng ( p q) ( q r) ( r s) ( u s) p u

Slide 100

Chng 5
5.1 M U

My h c

Cc chng tr c th o lu n v bi u di n v suy lu n tri th c. Trong tr ng h p ny gi nh c s n tri th c v c th bi u di n t ng minh tri th c. Tuy v y trong nhi u tinh hu ng, s khng c s n tri th c nh:

K s tri th c c n thu nh n tri th c t chuyn gia lnh v c. C n bi t cc lu t m t lnh v c c th . Bi ton khng c bi u di n t ng minh theo lu t, s ki n hay cc quan h . H c t k hi u: bao g m vi c hnh th c ha, s a ch a cc lu t t ng minh, s ki n v cc quan h . H c t d li u s : c p d ng cho nh ng h th ng c m hnh d i d ng s lin quan n cc k thu t nh m t i u cc tham s . H c theo d ng s bao g m m ng Neural nhn t o, thu t gi i di truy n, bi ton t i u truy n th ng. Cc k thu t h c theo s khng t o ra CSTT t ng minh.
Slide 101

C hai ti p c n cho h th ng h c:

5.2 CC HNH TH C H C
1. H c v t: H ti p nh n cc kh ng nh c a cc quy t nh ng. Khi h t o ra m t quy t nh khng ng, h s a ra cc lu t hay quan h ng m h s d ng. Hnh th c h c v t nh m cho php chuyn gia cung c p tri th c theo ki u tng tc. 2. H c b ng cch ch d n: Thay v a ra m t lu t c th c n p d ng vo tnh hu ng cho tr c, h th ng s c cung c p b ng cc ch d n t ng qut. V d : "gas h u nh b thot ra t van thay v thot ra t ng d n". H th ng ph i t mnh ra cch bi n i t tr u t ng n cc lu t kh d ng. 3. H c b ng qui n p: H th ng c cung c p m t t p cc v d v k t lu n c rt ra t t ng v d . H lin t c l c cc lu t v quan h nh m x l t ng v d m i.
Slide 102

5.2 CC HNH TH C H C (Ti p)

4. H c b ng tng t : H th ng c cung c p p ng ng cho cc tc v tng t nhng khng gi ng nhau. H th ng c n lm thch ng p ng tr c nh m t o ra m t lu t m i c kh nng p d ng cho tnh hu ng m i. 5. H c d a trn gi i thch: H th ng phn tch t p cc l i gi i v d (v k t qu ) nh m n nh kh nng ng ho c sai v t o ra cc gi i thch dng h ng d n cch gi i bi ton trong tng lai. 6. H c d a trn tnh hu ng: B t k tnh hu ng no c h th ng l p lu n u c lu tr cng v i k t qu cho d ng hay sai. Khi g p tnh h ng m i, h th ng s lm thch nghi hnh vi lu tr v i tnh hu ng m i. 7. Khm ph hay h c khng gim st: Thay v c m c tiu t ng minh, h khm ph lin t c tm ki m cc m u v quan h trong d li u nh p. Cc v d v h c khng gim st bao g m gom c m d li u, h c nh n d ng cc c tnh c b n nh c nh t cc i m nh.

Slide 103

V d v CC HNH TH C H C

V d : - H MYCIN - M ng Neural nhn t o - Thu t ton h c Quinland - Bi ton nh n d ng - My chi c car, c t ng

Slide 104

5.3 THU T GI I Quinlan

-

L thu t ton h c theo quy n p dng lu t, a m c tiu. Do Quinlan a ra nm 1979. t ng: Ch n thu c tnh quan tr ng nh t t o cy quy t nh. Thu c tnh quan tr ng nh t l thu c tnh phn lo i B ng quan st thnh cc b ng con sao cho t m i b ng con ny d phn tch tm quy lu t chung.

Slide 105

5.3.1 THU T GI I A. Quinlan

STT 1 2 3 4 5 6 7 8 Size Small Large Large Small Large Large Large Small Nationality German French German Italian German Italian Italian German Family Single Single Single Single Married Single Married Married Conclusion A A A B B B B B
Slide 106

V i m i thu c tnh c a b ng quan st:

Xt vector V: c s chi u b ng s phn lo i

V(Size=Small) = (ASmall, BSmall) ASmall=S quan st A c Size l Small / T ng s quan st c Size=Small BSmall= S quan st B c Size l Small / T ng s quan st c Size=Small
V(Size=Small) = (1/3 , 2/3) V(Size=Large) = (2/5 , 3/5)

V i thu c tnh Nationality

V(Nat = German)= (2/4 , 2/4) V(Nat = French) = (1 , 0) V(Nat = Italian) = (0 , 1)

Thu c tnh Family:

V(Family=Single) = (3/5 ,2/5) V(Family = Married) = (0, 1)

Slide 107

V i m i thu c tnh c a b ng quan st:

Ch cn xt German Thu c tnh Size: V(Size=Small) = (1/2 , 1/2) V(Size=Large) = (1/2 , 1/2) Thu c tnh Family: V(Family=Single) = (1, 0) V(Family=Married) = (0,1) STT 1 2 3 4 Size Small Large Large Small Family Single Single Married Married Conclusion A A B B

Nationality Italian French Single German Married

Slide 108

V i m i thu c tnh c a b ng quan st(ti p)

Nationality Italian French Single German Married

Rule 1: If (Nationality IS Italian) then (Conclusion IS B) Rule 2: If (Nationality IS French) then (Conclusion IS A) Rule 3: If (Nationality IS German) AND (Family IS Single) then (Conclusion IS A) Rule 4: If (Nationality IS German) AND (Family IS Married) then (Conclusion IS B)

Slide 109

5.3.2 Thu t gi i H c theo

Stt 1 2 3 4 5 6 7 8 9 10 Age Old Midle Midle Old New New Midle New Midle Old Competition No Yes No No No No No Yes Yes Yes Type Software Software Hardware Hardware Hardware Software Software Software Hardware Hardware

b t
Profit Down Down Up Down Up Up Up Up Down Down

nh

Slide 110

H c theo
b t nh c a X:

b t

nh(ti p)
k

E ( X ) = - p i log 2 p i
i =1

Tnh Entropy cho m i thu c tnh v ch n thu c tnh c Entropy nh nh t.

E ( C / A ) = - p ( c i , a i ) log 2 p ( c i , a i )
i =1

4 4 2 2 E ( C / Competitio n = No ) = - log 2 - log 2 = 0 . 918 6 6 6 6 1 1 3 3 E ( C / Competitio n = Yes ) = - log 2 - log 2 = 0 . 811 4 4 4 4 E ( C / Competitio n ) = 0 . 6 * 0 .918 + 0 .4 * 0 . 811 = 0 .8752
Slide 111

H c theo

b t
STT 1 2 3 4
Age

nh(ti p)
Competition Yes No No Yes Type Software Hardware Software Hardware Profit Down Up Up Down

Tng t : E(C/Age) = 0.4 E(C/Type) = 1 Age cho nhi u thng tin nh t

Old Down No Up

Milde Competition Yes Down

New Up
Slide 112

H c theo

b t
Age

nh(ti p)

Old Down No Up

Milde Competition Yes Down

New Up

Rule 1: If (Age IS Old) then (Profit IS Down) Rule 2: If (Age IS New) then (Profit IS Up) Rule 3: If (Age IS Midle) And (Competition IS No) then (Profit IS Up) Rule 4: If (Age IS Midle) And (Competition IS Yes) then (Profit IS Down)

Slide 113