Ukufunda ngomshini ekuthuthukisweni kwamaselula: amathemba nokwabiwa kwezindawo

Sawubona, Habr!

Akukho esingakwengeza esihlokweni se-athikili kusaziso sethu sangaphambilini - ngakho-ke wonke umuntu umenywa ekati ngokushesha. Funda futhi ubeke amazwana.

Ochwepheshe bezokuthuthukiswa kwamaselula bazozuza ezinguqukweni eziguquguqukayo ezilethwa namuhla. ukufunda komshini kumadivayisi. Iphuzu liwukuthi lobu buchwepheshe buthuthukisa kangakanani noma yiluphi uhlelo lokusebenza lweselula, okungukuthi, luhlinzeka ngezinga elisha lokusebenziseka kalula kubasebenzisi futhi likuvumela ukuthi usebenzise ngenkuthalo izici ezinamandla, isibonelo, ukunikeza izincomo ezinembe kakhulu, ngokusekelwe ku-geolocation, noma thola ngokushesha izifo zezitshalo.

Lokhu kuthuthukiswa okusheshayo kokufunda komshini ophathwayo kuyimpendulo ezinkingeni eziningi ezivamile esiye sabhekana nazo ekufundeni komshini wakudala. Eqinisweni, konke kusobala. Ngokuzayo, izinhlelo zokusebenza zeselula zizodinga ukucutshungulwa kwedatha ngokushesha kanye nokwehliswa okwengeziwe kokubambezeleka.

Kungenzeka ukuthi ubusuzibuza ukuthi kungani Izinhlelo zokusebenza zeselula ezinikwe amandla nge-AI,ayikwazi ukumane isebenzise okucatshangwayo efwini. Okokuqala, ubuchwepheshe bamafu buncike kumanodi amaphakathi (cabanga isikhungo esikhulu sedatha esinakho kokubili ukugcinwa kwedatha okubanzi namandla amakhulu ekhompyutha). Le ndlela emaphakathi ayikwazi ukuphatha isivinini sokucubungula esanele ukuze idale okuhlangenwe nakho okushelelayo kweselula okunikwa amandla ukufunda komshini. Idatha kufanele icutshungulwe phakathi nendawo bese ibuyiselwa kumadivayisi. Le ndlela idinga isikhathi, imali futhi ayiqinisekisi ubumfihlo bedatha ngokwayo.

Ngakho-ke, ngemva kokuchaza lezi zinzuzo ezibalulekile zokufunda komshini weselula, ake sibhekisise ukuthi kungani inguquko yokufunda komshini ivela phambi kwamehlo ethu kufanele ibe nentshisekelo kuwe siqu njengonjiniyela weselula.

Yehlisa Ukubambezeleka

Abathuthukisi bezinhlelo zokusebenza zeselula bayazi ukuthi ukubambezeleka okwenyukile kungaba uphawu olumnyama kuhlelo, kungakhathaliseki ukuthi izici zalo zinhle kangakanani noma ukuthi umkhiqizo unedumela elingakanani. Ngaphambilini, kumadivayisi we-Android kwakukhona Ukubambezeleka okukhulu ezinhlelweni eziningi zevidiyo, ngenxa yokuthi ukubukwa kwevidiyo nomsindo kuvame ukuvela kungavumelaniswa. Ngokufanayo, iklayenti lenkundla yezokuxhumana eline-latency ephezulu lingenza ukuxhumana kube ukuhlukunyezwa kwangempela kumsebenzisi.

Ukusebenzisa ukufunda ngomshini kudivayisi kuya ngokuya kubaluleke kakhulu ngenxa yezinkinga zokubambezeleka ezifana nalezi. Cabanga ukuthi izihlungi zezithombe zisebenza kanjani kumanethiwekhi omphakathi, noma izincomo zezindawo zokudlela ezisuselwa kundawo. Kulezi zinhlelo zokusebenza, ukubambezeleka kufanele kube kuncane ukuze kusebenze ezingeni eliphezulu kakhulu.

Njengoba kushiwo ngenhla, ukucutshungulwa kwamafu kwesinye isikhathi kungahamba kancane, futhi unjiniyela ufuna ukubambezeleka kusondele kuqanda ukuze amakhono omshini wohlelo lokusebenza lweselula asebenze kahle. Ukufunda ngomshini kumadivayisi kuvula amandla okucubungula idatha angehlisa ngempela ukubambezeleka kuze kube yiqanda.

Abakhiqizi bama-smartphone kanye neziqhwaga zemakethe yezobuchwepheshe kancane kancane baqala ukukubona lokhu. Isikhathi eside, i-Apple yahlala ingumholi kulo mkhakha, ithuthuka ama-chips athuthukile ngokwengeziwe kuma-smartphones asebenzisa uhlelo lwayo lwe-Bionic, olusebenzisa i-Neural Engine, esiza ukushayela amanethiwekhi e-neural ngqo kudivayisi, kuyilapho ifinyelela isivinini esimangalisayo.

I-Apple isaqhubeka nokuthuthukisa i-Core ML, inkundla yayo yokufunda umshini yezinhlelo zokusebenza zeselula, isinyathelo ngesinyathelo; emtatsheni wezincwadi I-TensorFlow Lite ukwesekwa okwengeziwe kwama-GPU; I-Google iyaqhubeka nokwengeza izici ezilayishwe kuqala kungxenyekazi yayo yokufunda ngomshini i-ML Kit. Usebenzisa lobu buchwepheshe, ungathuthukisa izinhlelo zokusebenza ezikuvumela ukuthi ucubungule idatha ngesivinini sombani, ususe noma yikuphi ukubambezeleka futhi unciphise inani lamaphutha.

Le nhlanganisela yokunemba nesipiliyoni somsebenzisi esingenazihibe iyimethrikhi eyinhloko okufanele abathuthukisi bohlelo lokusebenza lweselula bayicabangele lapho behlanganisa amakhono okufunda omshini ezinhlelweni zabo zokusebenza. Futhi ukuqinisekisa ukusebenza okunjalo, kuyadingeka hambisa ukufunda komshini kumadivayisi.

Ukuvikeleka okuthuthukisiwe nobumfihlo

Enye inzuzo enkulu ye-edge computing engakwazi ukweqiwa ukuthi ithuthukisa kangakanani ukuphepha komsebenzisi kanye nobumfihlo. Ukuqinisekisa ukuphepha nobumfihlo bedatha kuhlelo lokusebenza kuyingxenye ebalulekile yemisebenzi yonjiniyela, ikakhulukazi kucatshangelwa isidingo sokuhambisana ne-GDPR (General Data Protection Regulation), imithetho emisha yaseYurophu, ngokungangabazeki ezothinta umkhuba wokuthuthukiswa kweselula. .

Ngoba idatha ayidingi ukuthunyelwa phezulu noma emafini ukuze icutshungulwe, izigebengu ze-inthanethi azikwazi ukuxhaphaza noma yibuphi ubungozi obudalwe phakathi nesigaba sokudlulisa; ngakho-ke, ubuqotho bedatha buyagcinwa. Lokhu kwenza kube lula ngabathuthukisi bohlelo lokusebenza ukuthi bathobelane nemithetho yezokuphepha yedatha ye-GDPR.

Ukufunda ngomshini kumadivayisi nakho kunika amandla ukuhlukaniswa, ngendlela efanayo ne-blockchain. Ngamanye amazwi, kuba nzima kakhulu kubaduni ukuba baqalise ukuhlasela kwe-DDoS kunethiwekhi exhunyiwe yamadivayisi afihliwe kunokuba benze ukuhlasela okufanayo kuseva emaphakathi. Lobu buchwepheshe bungaba wusizo lapho usebenza ne-drones kanye nokuqapha ukuthotshelwa kwemithetho.

Ama-smartphone chips ashiwo ngenhla avela ku-Apple futhi asiza ukuthuthukisa ukuphepha komsebenzisi nobumfihlo - isibonelo, angasebenza njengesisekelo se-Face ID. Lesi sici se-iPhone sinikwa amandla inethiwekhi ye-neural efakwe kumadivayisi aqoqa idatha kusuka kuzo zonke izethulo ezihlukene zobuso bomsebenzisi. Ngakho-ke, ubuchwepheshe busebenza njengendlela yokuhlonza enembe kakhulu nethembekile.

Lezi kanye nezingxenyekazi zekhompuyutha ezintsha ezinikwe amandla i-AI zizovula indlela yokusebenzelana okuphephile kwabasebenzisi nama-smartphone. Eqinisweni, onjiniyela bathola isendlalelo esengeziwe sokubethela ukuvikela idatha yomsebenzisi.

Alukho uxhumano lwe-inthanethi oludingekayo

Ukubambezeleka kunezinkinga eceleni, ukuthumela idatha kumafu ukuze kucutshungulwe futhi kuthathwe iziphetho kudinga uxhumano lwe-inthanethi oluhle. Ngokuvamile, ikakhulukazi emazweni athuthukile, asikho isidingo sokukhononda nge-Internet. Kodwa yini okufanele uyenze ezindaweni lapho ukuxhumana kubi kakhulu? Uma ukufunda komshini kusetshenziswa kumadivayisi, amanethiwekhi e-neural ahlala emafonini ngokwawo. Ngakho, umthuthukisi angasebenzisa ubuchwepheshe kunoma iyiphi idivayisi nanoma yikuphi, ngaphandle kokunaka ikhwalithi yokuxhumeka. Ngaphezu kwalokho, le ndlela iholela ukwenza intando yeningi amakhono e-ML.

Ezempilo ingenye yezimboni ezingazuza ikakhulukazi ekufundeni komshini okudivayisi, njengoba abathuthukisi bezokwazi ukudala amathuluzi ahlola izimpawu ezibalulekile noma ahlinzeke ngokuhlinzwa kwerobhothi ngaphandle kokuxhumeka kwe-inthanethi. Lobu buchwepheshe buzoba wusizo kubafundi abafuna ukufinyelela ezintweni zokufunda ngaphandle koxhumano lwe-inthanethi - isibonelo, ngenkathi besemhubheni wezokuthutha.

Ekugcineni, ukufunda ngomshini kumadivayisi kuzonikeza onjiniyela ngamathuluzi okudala amathuluzi azozuzisa abasebenzisi emhlabeni wonke, ngaphandle kokunaka isimo sabo sokuxhuma ku-inthanethi. Uma kucatshangelwa ukuthi amandla ama-smartphones amasha azoba namandla okungenani njengawamanje, abasebenzisi bazokhohlwa ngezinkinga zokubambezeleka lapho besebenza nohlelo lokusebenza ungaxhunyiwe ku-inthanethi.

Ukunciphisa izindleko zebhizinisi lakho

Ukufunda ngomshini kumadivayisi nakho kungakongela imali eshisiwe ngokungakhokhi osonkontileka bangaphandle ukuze basebenzise futhi bagcine izixazululo eziningi. Njengoba kushiwo ngenhla, ezimweni eziningi ungenza ngaphandle kokubili ifu kanye Internet.

I-GPU nezinsizakalo zamafu eziqondene ne-AI yizixazululo ezibiza kakhulu ezingathengwa. Uma usebenzisa amamodeli kudivayisi yakho, awudingi ukukhokhela wonke lawa maqoqo, ngenxa yokuthi namuhla kunama-smartphones athuthuke kakhulu afakwe ama-neuromorphic processors (NPU).

Ngokugwema iphupho elibi lokucutshungulwa kwedatha elisindayo okwenzeka phakathi kwedivayisi namafu, wonga kakhulu; Ngakho-ke, kunenzuzo enkulu ukusebenzisa izixazululo zokufunda ngomshini kumadivayisi. Ngaphezu kwalokho, wonga imali ngoba izidingo zomkhawulokudonsa wohlelo lwakho lokusebenza zehliswe kakhulu.

Onjiniyela ngokwabo nabo bonga okuningi enqubweni yentuthuko, ngoba akudingeki bahlangane futhi bagcine ingqalasizinda yamafu eyengeziwe. Ngokuphambene nalokho, kungenzeka ukuzuza okwengeziwe ngeqembu elincane. Ngakho-ke, ukuhlelwa kwabasebenzi emaqenjini okuthuthukisa kusebenza kakhulu.

isiphetho

Ngokungangabazeki, ngawo-2010, ifu laba isibusiso sangempela, lenza ukucutshungulwa kwedatha kube lula. Kodwa ubuchwepheshe obuphezulu buthuthuka ngokuphawulekayo, futhi ukufunda ngomshini kumadivayisi kungase kube indinganiso ye-de facto hhayi kuphela emkhakheni wokuthuthukiswa kweselula, kodwa naku-inthanethi Yezinto.

Ngokubambezeleka okuncishisiwe, ukuvikeleka okuthuthukisiwe, amandla angaxhunyiwe ku-inthanethi, kanye nezindleko eziphansi zizonke, akumangazi ukuthi abadlali abakhulu ekuthuthukisweni kwamaselula babheja kakhulu ngobuchwepheshe. Abathuthukisi bezinhlelo zokusebenza zamaselula kufanele nabo bakubhekisise ukuze bahambisane nezikhathi.

Source: www.habr.com