117
I
fyou take morethan apassinginterest in the workingsof
your PC, you’renodoubtaware that computers work by
executinginstructionssequentially on digital data. In fact,
this method of working is so muchtakenfor grantedthat
unconventional computingarchitectures arecommonlythought
of as the sole domainof sciencefiction.Yetalternatives have
existed in the past and, if someof today’sresearch is at all
indicative, the futureofcomputingmightbetechnologically
morevariedthan we’veseen forthe last half century.
Onesuch alternativetechnologyisthe neural network.
This mightconjureupFrankenstein-likeimagesof brains with
electrodes attached,but mostcurrentresearch involves the
developmentofartificialneural networks(ANNs),inwhich
electronic circuitrysimulates the operationof the humanbrain.
Paradoxically,this is notanew wayofworking;farfromit.
Research into ANNscan traceits rootsback to the earlydays
of electronic computingand, overthe decades,it has enjoyed
occasionalresurgences of interest.
This mightsuggestthat it’saniche technology, with afew
obsessivesupporters, but one that is highlyunlikely to havea
majorimpacton the futureofcomputing. However,when we
discoverthat Intel, Microsoft and Googleall haveANN research
programmes,we mightjust be prepared to entertain the
thoughtthat neural computingcould, at last, be comingof age.MikeBedfordinvestigatesresearchintoanalternative
model ofcomputationthatoffersthepotentialofmassive
performancegainsbymimickingthehumanbrain
CEREBRAL
COMPUTATION