Sei sulla pagina 1di 5

I INTRODUCTION TO ARTIFICIAL INTELLIGENCE

Artifcial Intelligence is a combination of computer science p!"siolog" an# p!ilosop!"$ Artifcial


Intelligence %AI& is t!e area of computer science focusing on creating mac!ines t!at can engage on
be!a'iors t!at !umans consi#er intelligent$
T!e abilit" to create intelligent mac!ines !as intrigue# !umans since ancient times an# to#a" (it!
t!e a#'ent of t!e computer an# )* "ears of researc! into AI programming tec!ni+ues t!e #ream
of smart mac!ines is becoming a realit"$
Intelligence is t!e abilit" to t!in, to imagine creating memori-ing an# un#erstan#ing
recogni-ing patterns ma,ing c!oices a#apting to c!ange an# learn from e.perience$ T!is is t!e
branc! of computer science concerne# (it! ma,ing computers be!a'e li,e !umans$ /ence it is
calle# as 0Artifcial Intelligence0 123$
II$ INTRODUCTION TO NEURAL NET4OR56
Neural Net(or,s basicall" aim at mimic,ing t!e structure an# functioning of t!e !uman brain to
create intelligent be!a'ior$ Researc!ers are attempting to buil# a silicon7base# electronic net(or,
t!at is mo#ele# on t!e (or,ing an# form of t!e !uman brain8
Our brain is a net(or, of billions of neurons eac! connecte# (it! t!e ot!er$ At an in#i'i#ual le'el
a neuron !as 'er" little intelligence in t!e sense t!at it operates b" a simple set of rules
con#ucting electric signals t!roug! its net(or,$ /o(e'er t!e combine# net(or, of all t!ese
neurons creates intelligent be!a'ior t!at is unri'ale# an# unsurpasse#$
/o( t!e !uman brain (or,s it learns to reali-e patterns an# remembers t!em$ 6imilarl" t!e
neural net(or,s #e'elope# !a'e t!e abilit" to learn patterns an# remember$ T!is approac! !as its
limitations #ue to t!e scale an# comple.it" of #e'eloping an e.act replica of a !uman brain as
t!e neurons number in billions8 Currentl" t!roug! simulation tec!ni+ues people create 'irtual
neural net(or,s 193$
I III$ A::LICATION6 OF NEURAL NET4OR56
Evolu Evolution of Neural Networks for Control of Pursuit & Evasion:


Figure$2 ;o'ie se+uences illustrate be!a'iour generate# b" #"namical recurrent neural net(or,
controllers$
T!e follo(ing ;:EG mo'ie se+uences illustrate be!a'ior generate# b" #"namical recurrent
neural net(or, controllers co7e'ol'e# for pursuit an# e'asion capabilities$
From an initial population of ran#om net(or, #esigns successful #esigns in eac! generation
are selecte# for repro#uction (it! recombination mutation an# gene #uplication$
1
6election is base# on measures of !o( (ell eac! controller performs in a number of pursuit7
e'asion contests$ In eac! contest a pursuer controller an# an e'a#er controller are pitc!e#
against eac! ot!er controlling simple <<'isuall" gui#e#00 97#imensional autonomous 'irtual
agents$
=ot! t!e pursuer an# t!e e'a#er !a'e limite# amounts of energ" (!ic! is use# up in
mo'ement so t!e" !a'e to e'ol'e to mo'e economicall"$ Eac! contest results in a time7series
of position an# orientation #ata for t!e t(o agents$ T!ese time7 series are t!en fe# into a
custom >7D mo'ie generator$ It is
AKGEC JOURNAL OF TECNOLOG!" 'ol$2 no$9
important to note t!at alt!oug! t!e c!ase be!a'iors are genuine #ata t!e >D structures
surface p!"sics an# s!a#ing are all purel" for illustrati'e e?ect$
Learnin# t$e %istri&ution of O&'e(t Tra'e(tories for Event Re(o#nition: T!is (or, is about
t!e mo#eling of o&'e(t &e$aviors using #etaile# learnt statistical mo#els$ T!e tec!ni+ues
being #e'elope# (ill allo( mo#els of c!aracteristic ob@ect be!a'iors to be learnt from t!e
continuous obser'ation of long image se+uences.
It is !ope# t!at t!ese mo#els of c!aracteristic be!a'iors (ill !a'e a number of uses
particularl" in automate# sur'eillance an# e'ent recognition, allo(ing t!e sur'eillance
problem to be approac!e# from a lo(er le'el (it!out t!e nee# for !ig!7le'el sceneAbe!a'ioral
,no(le#ge$
Ot!er possible uses inclu#e t!e ran#om generation of realistic loo,ing ob@ect be!a'ior for use
in Birtual Realit" an# long7term pre#iction of ob@ect be!a'iors to ai# occlusion reasoning in
ob@ect trac,ing$


Figure 9C Ob@ect Tra@ectories for E'ent Recognition
T!e mo#el is learnt in an unsuper'ise# manner b" trac,ing ob@ects o'er long image se+uences
an# is base# on a combination of a neural net(or, implementing Bector Duanti-ation an# a t"pe
of neuron (it! s!ort7term memor" capabilities$
;o#els of t!e tra@ectories of pe#estrians !a'e been generate# an# use# to assess t!e t"picalit"
of ne( tra@ectories %allo(ing t!e i#entifcation of inci#ents of interest0 (it!in t!e scene&
pre#ict future ob@ect tra@ectories an# ran#oml" generate ne( tra@ectories$
2
Ra)iosit* for +irtual Realit* ,*ste-s .RO+ER/: T$e s"nt!esis of actual an# computer generate#
p!oto7realistic images !as been t!e aim of artists an# grap!ic #esigners for man" #eca#es$
6ome of
t!e most realistic images (ere generate# using ra#iosit" tec!ni+ues$
Unli,e ra" tracing ra#iosit" mo#els t!e actual interaction bet(een t!e lig!ts an# t!e
en'ironment$ In p!oto realistic Birtual Realit" %BR& en'ironments t!e nee# for +uic, fee#bac,
base# on user actions is crucial.
It is generall" recogni-e# t!at tra#itional implementation of ra#iosit" is computationall" 'er"
e.pensi'e an# t!erefore not feasible for use in BR s"stems (!ere practical #ata sets are of !uge
comple.it".
T!ere are t(o ne( met!o#s an# se'eral !"bri# tec!ni+ues to t!e ra#iosit" researc! communit"
on using ra#iosit" in BR applications$ On t!e left column E"b" (al,t!roug! an# a 'irtual space
are frst intro#uce# an# on t!e left$ On t!e rig!t one met!o# using Neural Net(or, tec!nolog"
is s!o(n$
Figure$> Ra#iosit" for Birtual Realit" 6"stemsC
Learning from E.ample$

Autono-ous 0alker & swi--in# Eel: T!e researc! in t!is area in'ol'es combining biolog",
mec!anical engineering an# information tec!nolog" in or#er to #e'elop t!e tec!ni+ues
necessar" to buil# a #"namicall" stable legge# 'e!icle controlle# b" a neural net(or,. T!is
(oul# incorporate comman# signals sensor" fee#bac, an# reEe. circuitr" in or#er to pro#uce
t!e #esire# mo'ement$
Figure F$ Autonomous (al,er$
3
Figure )$ 6imulation of t!e s(imming lampre"
%eel7li,e sea creature& #ri'en b" a neural net(or,.
Fa(ial Ani-ation: Facial animations create# using !ierarc!ical =7spline as t!e un#erl"ing
surface representation$ Neural net(or,s coul# be use for learning of eac! 'ariation in t!e
face e.pressions for animate# se+uences$ T!e %mas,& mo#el (as create# in 6oftImage an#
is an earl" protot"pe for t!e c!aracter G;ouseG in t!e HTBAA=C tele'isions series GRe=ootG
%T!e" #o not use !ierarc!ical splines for Reboot8&$
T!e original stan#ar# bicubic =7spline (as importe# to t!e GDragonG e#itor an# a !ierarc!"
automaticall" constructe#$ T!e surface (as attac!e# to a @a( to allo( it to open an# close
t!e mout!$ Groups of control 'ertices (ere t!en mo'e# aroun# to create 'arious facial
e.pressions$ T!ree of t!ese e.pressions (ere c!osen as ,e" s!apes t!e spline surface (as
e.porte# bac, to 6oftImage an# t!e ,e" s!apes (ere interpolate# to create t!e fnal
animation

Figure$I

Figure$J Facial animations using !ierarc!ical =7spline$
ARTIFICIAL INTELLIGENCE AND NEURAL NET4OR56
IB$A::LICATION6 OF ARTIFICIAL
INTELLIGENCE
%ata 1inin#: Disco'er" of meaningful patterns %,no(le#ge& from large 'olumes of #ata$
4
E23ert ,*ste-s: A computer program for #ecision ma,ing t!at simulates t!oug!t
process of a !uman e.pert$
Neural Networks: Tool base# on t!e brain analog"$
Fu44* Lo#i(: T!eor" of appro.imate reasoning$
Arti5(ial LifeC E'olutionar" Computation 6(arm Intelligence$
Evolutionar* Co-3utation6
Arti5(ial 7--une ,*ste-: A computer program base# on t!e biological immune s"stem$
A7 also Use) in : :attern recognition Optical c!aracter recognition /an#(riting recognition
6peec! recognition Face recognition Artifcial Creati'it" Computer 'ision Birtual realit" an#
Image processing Diagnosis %artifcial intelligence& Game t!eor" an# 6trategic planning
Game artifcial intelligence an# Computer game Natural language processing Translation
an# C!atter bots$
5

Potrebbero piacerti anche