The im­pact of deep learn­ing

The Star Malaysia - Star2 - - TECHNOLOGY -

By tAt­SuYA KiMurA and YOMi­uri SHiM­Bun

THE AI ( ar­ti­fi­cial in­tel­li­gence) tech­nol­ogy called deep learn­ing has the po­ten­tial to bring about revo­lu­tion­ary changes to so­ci­ety.

The self- learn­ing tech­nol­ogy is al­ready be­ing used not only for games but also for med­i­cal di­ag­noses, self- driv­ing cars and con­trol­ling ro­bots.

Un­til re­cently, it was dif­fi­cult for con­ven­tional com­put­ers to recog­nise, say, a cat in a pic­ture, some­thing hu­mans can eas­ily do. In the past, to en­able a com­puter to in­stantly recog­nise a cat, it was nec­es­sary for pro­gram­mers to in­put de­tailed fea­tures of a cat.

Now, deep learn­ing al­lows com­put­ers to over­come this weak­ness. By scan­ning large quan­ti­ties of data, com­put­ers can rapidly learn by them­selves to recog­nise fea­tures of cats – such as their whiskers and ears – with an ac­cu­racy that sur­passes that of hu­mans.

By analysing 160,000 games played by pro­fes­sional Go play­ers and record­ing the 30 mil­lion moves in games played by hu­man ex­perts, the com­puter pro­gram Al­phaGo learned moves that en­abled it to de­feat a pro­fes­sional player.

When a hu­man player places a Go stone, Al­phaGo recog­nises where the stone was placed and then de­ter­mines its next move based on what will give it the high­est prob­a­bil­ity for vic­tory.

Not only does Al­phaGo mon­i­tor the en­tire board and pre­dict the pat­terns of how a hu­man player will re­spond to its moves, but it can also make ef­fec­tive coun­ter­ing moves even in the event of un­ex­pected de­vel­op­ments in a match.

Though com­put­ers can­not im­me­di­ately trans­fer the ex­pe­ri­ence they ob­tain by play­ing Go to other fields in the way hu­man be­ings can, there is no doubt that deep learn­ing will be ap­plied even more in such fields as im­age and voice recog­ni­tion in the fu­ture.

An­other ap­pli­ca­tion of the tech­nol­ogy is ex­pected to en­able self- driv­ing cars to avoid hit­ting pedes­tri­ans and other cars, with Toy­ota Mo­tor Corp un­der­tak­ing joint re­search with a do­mes­tic start- up com­pany.

Deep learn­ing tech­nol­ogy also can be used for med­i­cal di­ag­noses to spot can­cer, among other med­i­cal prob­lems, in im­ages. Also, se­cu­rity soft­ware us­ing deep learn­ing can help de­tect sus­pi­cious per­sons in street im­ages.

It will play a big part in the fu­ture as com­pa­nies and gov­ern­ments have stepped up ef­forts in AI re­search. – Ja­pan News/ Asia News Net­work

— The Yomi­uri Shim­bun

Newspapers in English

Newspapers from Malaysia

© PressReader. All rights reserved.