B. VERMA SIR TIT Theoretical and Practical Concpts

8/18/2019 B. VERMA SIR TIT Theoretical and Practical Concpts

1/53


2/53

ARTIFICIAL INTELLIGENCE

AI s the *ran!h "$ !"mputer s!en!e that s !"n!erned +ththe aut"mat"n "$ Inte%%&ent *eha,"r.

Inte%%&en!e s n"t ,ery +e%% de-ned and there$"re has *een%ess underst""d.

Tas ass"!ated +th nte%%&en!e su!h as %earnn ntut"n!reat,ty and nter$eren!e a%% seem t" ha,e *een parta%%yunderst""d.

AI n ts /uest t" des&n nte%%&ent systems has $anned "utt" en!"mpass a num*er "$ te!hn"%"&es n ts $"%d.

O$ these te!hn"%"&es NN#FL# GA are pred"mnant%y n"+nas S"$t C"mputn&


3/53

What is Soft Computing

Soft computing is a consortium of methodologies that workssynergistically and provides, in one form or another, flexible

information processing capability for handling real-life ambiguous

situations.

Its aim is to exploit the tolerance for imprecision, uncertainty,approximate reasoning, and partial truth in order to achieve

tractability!olynomial complexity", robustness, and low-cost

solutions.

#he guiding principle is to devise methods of computation that lead

to an acceptable solution at low cost by seeking for an

approximate solution to an imprecisely$precisely formulated

problem.


4/53

Soft Computing #echnologies

- Fuzzy Sets provide a natural framework for theprocess in dealing with uncertainty.

- Neural Networks are widely used for classificationand rule generation.

- Genetic Algorithms %&s" are involved in various

optimi'ation and search processes.


5/53

GENETIC ALGORITH0S

GA are a $am%y "$ !"mputat"na% m"de%s nspred *y &enet!e,"%ut"n. The *as! dea s that ea!h 1nd,dua%2 "$ an e,"%,n&p"pu%at"n en!"des a !anddate s"%ut"n 'e.&. a pred!t"n ru%e) t" a&,en pr"*%em 'e.&. !%ass-!at"n).

GAs are adapt,e# r"*ust# e3!ent# and &%"*a% sear!h meth"ds#suta*%e n stuat"ns +here the sear!h spa!e s %ar&e. They "ptm4ea -tness $un!t"n# !"rresp"ndn& t" the pre$eren!e !rter"n # t"arr,e at an "ptma% s"%ut"n usn& !ertan &enet! "perat"rs.

GA has *een su!!ess$u%%y app%ed t" pr"*%ems that are d3!u%t t"s"%,e usn& !"n,ent"na% te!hn/ues su!h as s!hedu%n& pr"*%ems#tra,e%n& sa%espers"n pr"*%em# net+"r r"utn& pr"*%ems and-nan!a% maretn& et!.


6/53

SI0PLE GENETIC ALGORITH0

Step 56 Initiali'e a population ! of n elements each asa potential solution.

Step ()*ntil a specified termination condition is

satisfied) (a) *se a fitness function to evaluate each element of the

current population. If an element passes the fitness criteria,

it remains in !.

(b) #he population now contains m elements. *se geneticoperators to create new elements. &dd the new elements

to the population.


7/53

ADVANTAGES OF GA

Easy t" understand Genera% purp"se# r"*ust sear!h te!hn/ue 7e a%+ays &et an ans+er and t &ets *etter +th tme Inherent%y para%%e% and eas%y dstr*uted

Supp"rts mu%t "*8e!t,e "ptm4at"n G""d $"r n"sy en,r"nment 0"du%ar# separate $r"m app%!at"n Smp%e# P"+er$u%# Adapt,e# Para%%e% Guarantee near "ptmum s"%ut"ns.

G,e s"%ut"ns "$ un9appr"(mated $"rm "$ pr"*%em. Fner &ranu%arty "$ sear!h spa!es.


8/53

CO0PONENTS OF A GA

A pr"*%em t" s"%,e# and ...En!"dn& te!hn/ue 'gene, chromosome)

Inta%4at"n pr"!edure (creation)

E,a%uat"n $un!t"n (environment)

Se%e!t"n "$ parents (reproduction)Genet! "perat"rs (mutation, recombination)

Parameter settn&s (practice and art)


9/53

Representat"n "$ nd,dua%s

: Strn& "$ parameters 'genes) :chromosome

e&. F'p#/#r#s#t)6 p q r s t : 7e !an use *t strn arrays# trees# %sts"r any "ther "*8e!ts

: Chr"m"s"me represent a !anddate

s"%ut"n: Bt9strn& representat"n 'E()6

1 0 0 1 1 0 1 0 1 1 0 1 1 0 0


10/53

; Standard GAs use *nary strn&s "$ -(ed

%en&th.

; Can *e eas%y m"d-ed $"r any -ntea%pha*ets.

F"r e(amp%e# !an use 5< sym*"%s =?.

E(amp%e s"%ut"n6 @5>5.

; Can use %etters and num*ers.

E(amp%e s"%ut"n6 THEANS7ERIS@


11/53

+O!ta% En!"dn&

+He(ade!ma% En!"dn&+Permutat"n En!"dn&

' E( Tra,e%%n& Sa%esman Pr"*%em)

+Va%ue En!"dn&

'Ths en!"dn& Re/ures t" m"d$y the

Genet! "perat"rs E( -ndn& +e&hts "$ NN)+ Tree En!"dn&

' 0an%y sed $"r &enet! pr"&rammn&' Lsp) )

Other Types "$ En!"dn&s "$Parameters


12/53

EA0PLE BINAR9CODED REPRESENTATIONS

F"r E(amp%e# %ets say that +e are tryn& t" "ptm4ethe $"%%"+n& $un!t"n# f(x) = x 2 $"r ≥ ( ≥ 5

I$ +e +ere t" use *nary9!"ded representat"ns +e+"u%d -rst need t" de,e%"p a mappn& $un!t"n $"rm"ur &en"type representat"n '*nary strn&) t" "urphen"type representat"n '"ur CS). Ths !an *e d"ne

usn& the $"%%"+n& mappn& $un!t"n6 d(ub,lb,l,chrom) = (ub-lb) decode(chrom)/2l-1 + lb


13/53

GENETIC ALGORITH0S6BINAR9CODED REPRESENTATIONS

d(ub,lb,l,c) = (ub-lb) decode(c)/2l -1 + lb # +here ub # lb 5# l the %en&th "$ the !hr"m"s"me n *ts

c the !hr"m"s"me The parameter# %# determnes the a!!ura!y 'and

res"%ut"n "$ "ur sear!h). 7hat happens +hen % s n!reased '"r de!reased)


14/53

T7O APPROACHES TO OTHERREPRESENTATIONS

+ 0ap t" *nary representat"n

0ap the "ther representat"n t" *nary strn&s "$-(ed9%en&th and use standard *nary9strn& GAs.

; 0"d$y "perat"rs '!r"ss",er and mutat"n)

0"d$y "perat"rs t" mae them +"r $"r the"ther representat"n. Can *e d"ne *y des&nn&a%ternat,es $"r standard !r"ss",er and mutat"n


15/53

Sur,,a% "$ the -ttest 'Fitness function): numer!a% 1-&ure "$ mert2Jut%ty measure "$

an nd,dua%

: Trade"K am"n&st a mu%tp%e e,a%uat"n !rtera

: E3!ent e,a%uat"n

: Ftness Fun!t"n s "$ten der,ed $r"m"*8e!t,e $un!t"n

: F') $'() $"r 0a(m4at"n pr"*%em F'() 5J$'() $"r mnm4at"n pr"*%em $ $'() <

F'() 5J '5 M $'()) $ $'() <

Ftness Fun!t"n


16/53

CREATING NE7 GENERATION OF OFFSPRING


17/53

OPERATORS6 SELECTION

; FITNESS PROPORTIONATE SELECTION 'FIJF ); N0BER OF !"#$%&'! '* F# INDIVIDALS

R"u%ette9+hee% se%e!t"n

: +hee% spa!ed n pr"p"rt"n t" -tness ,a%ues

: N 'p"p s4e) spns "$ the +hee%


18/53

T"urnament Se%e!t"n

+ Lst a%% nd,dua% +th -tness ,a%ue+

Se%e!t any t+" nd,dua% at rand"m+ Tae the "ne +th h&her -tness ,a%ue+ Repeat the same $"r se%e!tn& $u%% s4e"$ p"pu%at"n.


19/53


20/53

RAN SELECTION

R"u%ette +hee% has a pr"*%em +hen the -tness,a%ues "$ nd,dua% dKers ,ery mu!h. I$ *est!hr"m"s"me has -tness >


21/53


22/53


23/53


24/53


25/53

7H CROSSOVER AND 0TATION

Cr"ss",er Pr"du!es ne+ s"%ut"ns +h%e

remem*ern& the !hara!terst!s "$ "%d

s"%ut"ns Parta%%y preser,es dstr*ut"n "$ strn&s

a!r"ss s!hemas

0utat"n Rand"m%y &enerates ne+ s"%ut"ns

+h!h !ann"t *e pr"du!ed $r"m e(stn&p"pu%at"n

A,"ds %"!a% "ptmum


26/53

GENETIC ALGORITH0S6EA0PLE

The SGA $"r "ur e(amp%e +%% use6 A p"pu%at"n s4e "$ # A !r"ss",er usa&e rate "$ 5.


27/53

GENETIC ALGORITH0S6AN EA0PLE RN 'B HAND)

Rand"m%y Generate an Inta% P"pu%at"n

Gen"type Phen"type Ftness

Pers"n 56 5


28/53


E,a%uate P"pu%at"n at t<


Pers"n 56 5


29/53


Se%e!t S( Parents sn& the R"u%ette 7hee%


Pers"n 6


30/53


Create OKsprn& 5 sn& Sn&%e9P"nt Cr"ss",er


Pers"n 6


31/53


Create OKsprn& Q @


Pers"n 6 5


32/53


Create OKsprn&


Pers"n 6


33/53


E,a%uate the OKsprn&


Ch%d 5 6


34/53


P"pu%at"n at t< Gen"type Phen"type Ftness

Pers"n 56 5


35/53


P"pu%at"n at t5


Pers"n 56


36/53


The Pr"!ess "$6 Se%e!tn& s( parents# A%%"+n& the parents t" !reate s( "Ksprn 0utatn& the s( "Ksprn

E,a%uatn& the "Ksprn and Rep%a!n& the parents +th the "Ksprn&

Is repeated unt% a st"ppn& !rter"n has *eenrea!hed.


37/53

C/0%/C/

pr"&ress"n t"+ards un$"rmty np"pu%at"n

premature !"n,er&en!e

99'%"!a% "ptma)


38/53


39/53


40/53

HO7 GA 7ORS I.E. SCHE0AS

P"pu%at"n Strn&s ",er a%pha*et =


41/53

HPER9PLANE 0ODEL

Sear!h spa!e A hyper9!u*e n L dmens"na% spa!e

Ind,dua%s Vert!es "$ hyper9!u*e

S!hemas Hyper9p%anes $"rmed *y ,ert!es


42/53

SA0PLING HPER9PLANES

L"" $"r hyper9p%anes 's!hemas) +th &""d-tness ,a%ue nstead "$ ,ert!es 'nd,dua%s) t"redu!e sear!h spa!e

Ea!h ,erte( 0em*er "$ QL hyper9p%anes Samp%es hyper9p%anes

A,era&e Ftness "$ a hyper9p%ane !an *eestmated *y samp%n& -tness "$ mem*ers np"pu%at"n

Se%e!t"n retans hyper9p%anes +th &""d


43/53

SCHE0A THEORE0 S!hema Order O'H)

S!hema "rder# O'.) # s the num*er "$ n"n &enes ns!hema H.

E.&. O'55) There$"re s!hema H +%% represent %9"'H) nd,dua%s.

S!hema De$nn& Len&th W'H) S!hema De$nn& Len&th# W'H)# s the dstan!e *et+een

-rst and %ast n"n &ene n s!hema H E.&. W'55) @ : 5 Q

S!hemas +th sh"rt de-nn& %en&th# %"+ "rder +th-tness a*",e a,era&e p"pu%at"n are $a,"red *y GAs


44/53

S!hema The"rem Termn"%"&y 'C"nt.)6 Let m'H#t) den"ted the num*er "$ nstan!es "$ H that are n

the p"pu%at"n at tme t. Let $'H#t) den"te the a,era&e -tness "$ the nstan!es "$ H

that are n the p"pu%at"n at tme t.

Let $ a,&'t) represent the a,era&e -tness "$ the p"pu%at"n at

tme t.

Let p! and pm represent the sn&%e9p"nt !r"ss",er and

mutat"n rates.

A!!"rdn& t" the S!hema The"rem there +%% *e6 m'H#tM5) m'H#t) $'H#t)J$ a,&'t) nstan!es "$ H n the ne(t

p"pu%at"n $ H has an a*",e a,era&e -tness.


45/53

S!hema The"rem Termn"%"&y 'C"nt.)6 A!!"rdn& t" the S!hema The"rem there +%% *e6 m(H,t+1) = m(H,t) f(H,t)/f avg(t) nstan!es "$ H n the ne(t p"pu%at"n $ H has an a*",e

a,era&e -tness. I$ +e %et $'H#t) $ a,&'t) M ! $ a,&'t)# $"r s"me ! X


46/53

FOR0AL STATE0ENT

Se%e!t"n pr"*a*%ty

Cr"ss",er pr"*a*%ty

0utat"n pr"*a*%ty

E(pe!ted num*er "$ mem*ers "$ as!hema

f

t H f t H mt H m E

),(),())1,(( =+

1

)(

)( −= L H

ccrossover ph P δ

mmutation p H h P )()( Ο=

))(1

)(1(

),(),()1,(( H p

L

H p

f

t H f t H mt H m E mc Ο−

−−=+

δ


47/53

AREA OF APPLICATION

GAs !an *e used +hen6 N"n9ana%yt!a% pr"*%ems. N"n9%near m"de%s. n!ertanty. Lar&e state spa!es.


48/53

NON9ANALTICAL PROBLE0S

Ftness $un!t"ns may n"t *ee(pressed ana%yt!a%%y a%+ays.

D"man spe!-! n"+%ed&e may n"t*e !"mputa*%e $r"m -tness $un!t"n.

S!ar!e d"man n"+%ed&e t" &ude thesear!h.


49/53

NON9LINEAR 0ODELS

S"%ut"ns depend "n startn& ,a%ues.

N"n : %near m"de%s may !"n,er&e

t" %"!a% "ptmum.

Imp"se !"ndt"ns "n -tness

$un!t"ns su!h as !"n,e(ty# et!.


50/53

NCERTAINT

N"sy J appr"(mated -tness$un!t"ns.

Chan&n& parameters.

Chan&n& -tness $un!t"ns.


51/53

LARGE STATE SPACES

Heurst!s $"!us "n%y "n the mmedatearea "$ nta% s"%ut"ns.

State9e(p%"s"n pr"*%em6 num*er "$ stateshu&e "r e,en n-nteY T"" %ar&e t" *ehand%ed.

State spa!e may n"t *e !"mp%ete%yunderst""d.


52/53

GA APPLICATION EA0PLES

Fun!t"n "ptm4ers: d3!u%t# ds!"ntnu"us# mu%t9m"da%# n"sy $un!t"ns

C"m*nat"ra% "ptm4at"n

: %ay"ut "$ VLSI !r!uts# $a!t"ry s!hedu%n tra,e%n& sa%esman

pr"*%em Des&n and C"ntr"%

: *rd&e stru!tures# neura% net+"rs# !"mmun!at"n net+"rs

des&nZ !"ntr"% "$ !hem!a% p%ants# ppe%nes 0a!hne %earnn&

: !%ass-!at"n ru%es# e!"n"m! m"de%n s!hedu%n& strate&es P"rt$"%" des&n# "ptm4ed tradn& m"de%s# dre!t maretn&

m"de%s# se/uen!n& "$ TV ad,ertsements# adapt,e a&ents# datamnn et!


53/53

7HEN NOT TO SE GAY

C"nstraned mathemat!a%"ptm4at"n pr"*%ems espe!a%%y+hen there are $e+ s"%ut"ns.

C"nstrants are d3!u%t t"n!"rp"rate nt" a GA.

Guded d"man sear!h s p"ss*%e

and e3!ent.

B. VERMA SIR TIT Theoretical and Practical Concpts

Documents