Computer Shopper – September 2019

118 SEPTEMBER 2019 |COMPUTERSHOPPER|ISSUE 379

Here we take alook at someof

thesenewinitiatives with aviewto

understandinghowANNswork,

what’sbeen achieved, and howthey

mightaffect the real world, at atime

whenmanyindustryexpertsbelieve

that conventional digital computers

areapproachingafundamental

performancelimit.

Theneuraladvantage

Beforelookinganyfurtherat this

potentially newapproach to

computation,it makes senseto

considerwhywemightwantto

mimicbiologyinthis way. Afterall,

acomparisonbetween the power

of the humanbrain and that of a

digital computer mightnot appear

to presentour brains in a

particularlyfavourable light.

Shakuntala Devi wasnicknamed

the Human Computer.She held the

world record formultiplyingtwo

randomlyselected 13-digitnumbers

–specifically,7,686,369,774,870and

2,465,099,745,779,to givethe

correct answerof18,947,668,177,995,

426,462,773,730duringher record

attempt–injust 28 seconds.Most

peoplewould take significantly longer,

even usingapen and paper–several

minutes would probablybe more

typical forthe vast majorityof people.

By wayofcontrast, an ordinary

desktopPCwould be able to perform

severalbillion such operationsin a

second, and if youweretoshell out on

an 18-corevariantofIntel’smonster

Core i9 X-serieschips,the figure

would be closerto atrillion.

Paradoxically,this comparison

doesn’t alwayshold true. Asurprising

assertion,perhaps?Well, just think

back to the last time youfelt like

tearingyour hair out, whiletryingfor

the umpteenth time to tell an

automated telephoneanswering

systemwhat youcould have

communicated to ahumanoperator

in seconds.Nowbear in mindthat the

currentstate-of-play in speech

recognition,flawed as it may be, is the

result of over60years of research.

Speech recognitionis just one

exampleof awhole range of

problems,generally referred to as

pattern recognition,in whichdigital

computers struggle,yethumans

excel. What’smore, in taskssuch as

face recognition,readinghandwritten

text, and manyaspectsof the

much-hypedfield of artificial

intelligence, programmershave

difficultyin comingup with alogical

sequenceofinstructionsthat a

computer canfollow, whilehumans

do this intuitively,havinglearnedfrom

experience. This is all the more

impressivewhen we bear in mindthat

the inputdatafor all thesetasksisso

variable.Rarely do twopeople

pronouncethe sameword exactly the

same,the samesentencewritten by

twoindividualscanlook remarkably

different, and aparticularhuman

face canappeardifferentdepending

on whether that personis smilingor

frowning, whether their face is

partiallyobscured by ahat, by the

presenceofmakeup, and even on

external conditionssuch as lighting.

Brainpower

The ANN approach to computingis

electronic rather than biological. Even

so, because the basicarchitectureis

inspired by animalbrains, we’ll start

by delvinginto the brain’s workings.

The processingpowerofthe brain

comes from neurons that are

connected to other neurons. This may

suggestasimilaritywith today’s

processors,whichcontains multiple

coresthat areable to exchangedata

with each other,but in reality things

couldn’t be moredifferent. First, those

neurons aremuch simplerthan even

the mostbasicdigital processorcore

has ever been.Whileaprocessor core

ABOVE:Under-

standingthe

workingsofthe

brainis atthe

forefrontof

researchinto

up-and-coming

computer

architectures

RIGHT:The

biologicalneuron

maybeasimple

computingelement,

butit makesupfor

thatwithitsvast

numbersand

interconnections

Photo:Gerry Shaw