I-Meteorology njengenzululwazi iqhubela phambili enkosi kuphuhliso lwetekhnoloji. Okwangoku, kukho iinkqubo ezininzi zekhompyuter ezinokwazi ukuqikelela ngokuthe ngqo ukuba iza kunetha nini kwaye phi. Inkampani ye DeepMind uvelise ubukrelekrele bokuzenzela obunakho ukuqikelela phantse ngokuthe ngqo ukuba iza kunetha nini kwaye phi. Le nkampani isebenze kunye neemeteorologists zase-UK ukwenza imodeli ebhetele yokwenza uqikelelo lwexesha elifutshane kuneenkqubo ezikhoyo.
Kweli nqaku siza kukuxelela yonke into ekufuneka uyazi malunga nengxowa yeRobleda, itekhnoloji yemozulu yokuqikelela kwinkampani ye-DeepMind.
Uqikelelo lwemozulu
I-DeepMind, inkampani esekwe eLondon yobukrelekrele bokuzenzelaUqhubeka nomsebenzi wakhe wokufunda ngokunzulu kwiingxaki ezinzima zesayensi. I-DeepMind iphuhlise isixhobo sokufunda esinzulu esibizwa ngokuba yi-DGMR ngokusebenzisana ne-Met National Service yase-Bhritane, enokuchaza ngokuchanekileyo amathuba emvula kwimizuzu engama-90 ezayo. Ngomnye wemiceli mngeni enzima kuqikelelo lwemozulu.
Xa kuthelekiswa nezixhobo ezikhoyo, uninzi lweengcali zikholelwa ukuba uqikelelo lwe-DGMR lolona luhamba phambili kwizinto ezininzi, kubandakanya ukuqikelela indawo, uluhlu, intshukumo kunye nokuqina kwemvula, i-89% yexesha. Isixhobo esitsha se-DeepMind sivula isitshixo esitsha kwi-biology izazinzulu ebezama ukusombulula amashumi eminyaka.
Nangona kunjalo, nkqu nokuphuculwa okuncinci kuqikelelo kubalulekile. Ukuqikelela imvula, ngakumbi imvula enkulu, kubalulekile kumashishini amaninzi, ukusuka kwimisebenzi yangaphandle ukuya kwiinkonzo zokuhamba ngenqwelomoya kunye nongxamiseko. Kodwa ukuyilungisa kunzima. Ukufumanisa ukuba amanzi angakanani esibhakabhakeni kwaye nini kwaye iya kuwela phi kuxhomekeke kwiinkqubo ezininzi zemozulu, njengotshintsho lobushushu, ukwakheka kwamafu, kunye nomoya. Zonke ezi zinto zinzima ngokwaneleyo ngokwazo, kodwa zinzima ngakumbi xa zidityanisiwe.
Eyona teknoloji ibalaseleyo yokuqikelela isebenzisa inani elikhulu leemodeli zekhompyuter zefizikiki yomoya. Ezi zilungele uqikelelo lwexesha elide, kodwa azilunganga kakhulu ekuqikeleleni okuza kwenzeka kule yure izayo. Oku kubizwa ngokuba luqikelelo kwangoko.
Uphuhliso olunzulu
Iindlela zokufunda ezinzulu zangaphambili ziye zaphuhliswa, kodwa ezi ndlela zihlala zisebenza kakuhle kwicala elinye, njengokuqikelela indawo, kunye nokuchaphazela enye, njengokuxela kwangaphambili. Idatha yeRadar yemvula enzima enceda ukuqikelela imvula kwangoko ihlala ingumceli mngeni omkhulu kwizazi zemozulu.
Iqela le-DeepMind lisebenzise idatha ye-radar ukuqeqesha i-AI yabo. Amazwe amaninzi kunye nemimandla ihlala ipapasha iifoto zemilinganiselo ye-radar elandelela ukwakheka kwamafu kunye nokuhamba imini yonke. Umzekelo, e-UK, ukufundwa okutsha kuthunyelwa rhoqo kwimizuzu emihlanu. Ngokubeka ezi zinto zidibeneyo, unokufumana ividiyo yokuma eshukumayo ehlaziyiweyo ebonisa indlela ipateni yemvula yelizwe etshintsha ngayo.
Abaphandi bathumela le datha kuthungelwano olunzulu lwesizukulwane esifana neGAN, eyi-AI eqeqeshiweyo enokuvelisa iisampulu zedatha ezintsha ezifanayo kakhulu nedatha yokwenyani esetyenziswe kuqeqesho. I-GAN isetyenziselwe ukuvelisa ubuso bobuxoki, kubandakanya iRembrandt engeyiyo. Kule meko, i-DGMR (emele "Imodeli yemvula enzulu") ifundile ukuvelisa izifinyezo ze-radar ezingezizo eziqhubeka ngokulandelelana komlinganiso.
Uvavanyo lwe-DeepMind AI
UShakir Mohamed, okhokele uphando kwi-DeepMind, uthe oku kuyafana nokubukela iifreyimu ezimbalwa kumdlalo bhanyabhanya kunye nokuqikelela ukuba kuzakwenzeka ntoni emva koko. Ukuvavanya le ndlela, iqela libuze ama-meteorologists angama-56 avela kwi-Bureau of Meteorology (ababengabandakanyekanga emsebenzini) ukuba bajonge kwiimpawu eziphambili zomzimba kunye neqela labachasi.
I-89% yabantu ithe bakhetha iziphumo ezinikezwe yi-DGMR. Ialgorithms yokufunda ngomatshini ngokubanzi izama ukwandisa umlinganiso olula wokuba zilungile kangakanani iingqikelelo zakho. Nangona kunjalo, uqikelelo lwemozulu lunezinto ezininzi ezahlukeneyo. Mhlawumbi ukubikezela kuye kwafumana amandla emvula angalunganga kwindawo efanelekileyo, okanye esinye isibikezelo sifumene indibaniselwano yamandla kodwa ichanekile kwindawo engeyiyo, njalo njalo.
I-DeepMind ithe izakukhupha ubume bazo zonke iiproteni ezaziwa yisayensi. Inkampani isebenzise iprotein yayo yeAlphaFold yokusonga ubukrelekrele bokuvelisa ukwenza iiproteni zomntu, kunye negwele, iimpukane zeziqhamo kunye neempuku.
Intsebenziswano phakathi kwe-DeepMind kunye ne-Met Office ngumzekelo olungileyo wokusebenza nabasebenzisi bokugqibela ukugqibezela uphuhliso lwe-AI. Ngokucacileyo olu luvo olulungileyo, kodwa amaxesha amaninzi akwenzeki. Iqela lisebenze kwiprojekthi iminyaka eliqela kwaye igalelo elivela kwiingcali ezivela kwiZiko leMeteorology liyile iprojekthi. USuman Ravuri, isazinzulu sophando kwi-DeepMind, uthe: "Ikhuthaza ukuphuhliswa kwemodeli yethu ngendlela eyahlukileyo kunokuphunyezwa kwethu." "Ngaphandle koko, besinokwenza imodeli ebingayi kuba luncedo kakhulu ekugqibeleni."
I-DeepMind ikwanomdla wokubonisa ukuba i-AI yayo inezicelo ezisebenzayo. I-Shakir, i-DGMR kunye ne-AlphaFold ziyinxalenye yebali elifanayo: inkampani isebenzisa iminyaka yabo yamava ukusombulula iipuzzle. Mhlawumbi esona sigqibo sibaluleke kakhulu apha kukuba i-DeepMind igqibile ukudwelisa iingxaki zehlabathi zenzululwazi.
Inkqubela phambili kuqikelelo lwemozulu
Uqikelelo lwemozulu kufuneka luxhaswe luphuhliso lwetekhnoloji njengoko sisondela kwaye sisondela ekuqondeni ngokupheleleyo ukuba umoya wethu usebenza njani. Amaxesha amaninzi umntu kunye nokubala kwakhe kunokuxhomekeka kwiimpazamo eziqhelekileyo ezinokuphetshwa ngophuhliso lobukrelekrele bokuzenzela.
Uqikelelo lwemozulu ngundoqo ekubeni ngumntu kuba singasebenzisa ithuba elininzi ubutyebi bamanzi obusebenza kakuhle ngakumbi kwaye kuthintelwe iintlekele ezithile kwisaqhwithi nakwiimvula ezinkulu. Ngesi sizathu, oochwephesha bemozulu bayavuma ukuba baphuhlise iiprojekthi zobukrelekrele bokuqikelela imvula.
Ndiyathemba ukuba ngolu lwazi unokufunda ngakumbi malunga neprojekthi ye-DeepMind kunye neempawu zayo.