In the near fu­ture, smart­phones­may work like your brain

HWM (Malaysia) - - FEATURE -

By For sev­eral years now, chip­maker Qual­comm and Brain Corp, a sep­a­rate com­pany that it has in­vested in, have been work­ing on a com­bined hard­ware and soft­ware plat­form that at­tempts to mimic the pro­cesses of the hu­man brain. They call the plat­form, ‘Ze­roth’.

The goal of Ze­roth is for it to do more than just per­form pre-pro­grammed tasks. It will also be able to learn and adapt ac­cord­ingly. The tech­nol­ogy driv­ing Ze­roth is based on a fam­ily of al­go­rithms called Deep Learn­ing, which has also been used by Google sub­sidiary Deep­Mind to pro­gram a com­puter to play Atari video games at a su­per­hu­man level. Deep Learn­ing soft­ware is loosely mod­elled on the way the hu­man brain works: it can be trained to rec­og­nize cer­tain ob­jects in images by pro­cess­ing many ex­am­ple pho­tos through a net­work of ar­ti­fi­cial ‘neu­rons’ ar­ranged into hi­er­ar­chi­cal lay­ers. In other words, it rec­og­nizes images by com­par­ing them to other images in its mem­ory, sim­i­lar to the way you can iden­tify ob­jects, even if you’ve never seen that same ex­act ob­ject be­fore, based on your own past ex­pe­ri­ences.

At MWC 2015 last month, Qual­comm showed off the first work­ing pro­to­type of Ze­roth. When in­te­grated into a smart­phone, it demon­strated how Ze­roth could im­prove a smart­phone’s cam­era app by be­ing able to suc­cess­fully rec­og­nize in­di­vid­ual el­e­ments of what you’re tak­ing a pic­ture of, such as food, a city sky­line or a group of friends, even if it had no prior ex­pe­ri­ence with the ac­tual sub­ject mat­ter you’re shoot­ing. The app would then au­to­mat­i­cally ad­just the cam­era set­tings to en­sure you take the best pos­si­ble pic­ture. The app was even ca­pa­ble of live tag­ging the names of your friends in real time by ref­er­enc­ing pre­vi­ous pho­tos you’ve taken of them.

Qual­comm also said that, as well as pro­cess­ing images, the Ze­roth soft­ware could al­low phones to rec­og­nize speech or other sounds, and to learn to spot pat­terns of ac­tiv­ity from a de­vice’s sen­sors. Com­put­ing at this level, which re­quires more than just sim­ple task op­er­a­tions, gen­er­ally re­quires Cloud Com­put­ing to ac­com­plish, but Qual­comm says that all such com­pu­ta­tion will be per­formed on the phone it­self. By keep­ing all op­er­a­tions at a lo­cal level, not only will Ze­roth re­move the re­quire­ment of In­ter­net con­nec­tiv­ity, it will also en­able faster feed­back and ac­tion to be taken from data re­ceived from the de­vice’s sen­sors. Qual­comm has sug­gested that one of the first prac­ti­cal ap­pli­ca­tions of Ze­roth’s pre­dic­tive learn­ing skills will be ex­tend­ing de­vice bat­tery life by track­ing the way a per­son uses their phone and learn­ing when it can safely power down to save en­ergy with­out af­fect­ing the user ex­pe­ri­ence.

The Ze­roth soft­ware is be­ing de­vel­oped to launch with Qual­comm’s Snap­dragon 820 sys­tem-on-a-chip, which will en­ter pro­duc­tion later this year, mean­ing it should be avail­able in con­sumer de­vices in early 2016. Qual­comm has also an­nounced that as well as smart­phone man­u­fac­tur­ers, the Snap­dragon 820 and the Ze­roth soft­ware will be aimed at man­u­fac­tur­ers of ar­ti­fi­cial in­tel­li­gence drones and ro­bots. Hope­fully, the plat­form lives up to its name - it’s taken from science fic­tion au­thor Isaac Asimov’s ‘Ze­roth Law of Ro­bot­ics’: “Ro­bots must not harm hu­man­ity.”

Newspapers in English

Newspapers from Malaysia

© PressReader. All rights reserved.