I-AI, izingane zesikole kanye nemiklomelo emikhulu: ungakwenza kanjani ukufunda ngomshini ebangeni lesi-8

Sawubona Habr!

Sithanda ukukhuluma ngendlela engavamile kangaka yokuthola imali yentsha njengokuhlanganyela kuma-hackathons. Lokhu kokubili kuzuzisa ngokwezimali futhi kukuvumela ukuba usebenzise ulwazi oluzuzwe esikoleni nokufunda izincwadi ezihlakaniphile.

Isibonelo esilula i-hackathon ye-Artificial Intelligence Academy yangonyaka odlule yezingane zesikole. Abahlanganyeli bayo kwakudingeka babikezele umphumela womdlalo we-Dota 2. Umnqobi walo mncintiswano kwakungu-Alexander Mamaev, umfundi webanga leshumi waseChelyabinsk. I-algorithm yakhe yanquma ngokunembe kakhulu iqembu eliwinile empini. Ngenxa yalokhu, u-Alexander wathola umklomelo omkhulu wemali - ruble ayizinkulungwane eziyi-100.

Indlela u-Alexander Mamaev ayisebenzisa ngayo imali yomklomelo, yiluphi ulwazi umfundi alukho ukuze asebenze ne-ML, futhi yisiphi isiqondiso emkhakheni we-AI asibheka njengesithakazelisa kakhulu - umfundi utshele inhlolokhono.

— Sitshele ngawe, waba kanjani nesithakazelo ku-AI? Ingabe bekunzima ukungena esihlokweni?
— Ngineminyaka engu-17 ubudala, ngiqeda isikole kulo nyaka, futhi muva nje ngisuka eChelyabinsk ngiya eDolgoprudny, eseduze neMoscow. Ngifunda eKapitsa Physics and Technology Lyceum, lesi ngesinye sezikole ezihamba phambili esifundeni saseMoscow. Ngangikwazi ukuqasha indawo yokuhlala, kodwa ngihlala esikoleni okuhlalwa kuso esikoleni, kungcono futhi kulula ukuxhumana nabantu abavela e-lyceum.

Isikhathi sokuqala lapho ngizwa nge-AI ne-ML mhlawumbe kwakungu-2016, lapho uPrisma ebonakala. Ngabe sengifunda ibanga lesi-8 futhi ngenza izinhlelo ze-olympiad, ngaya kuma-olympiad ngathola ukuthi sasiba ne-ML meetups edolobheni. Nganginentshisekelo yokuyihlola, ngiqonde ukuthi isebenza kanjani, futhi ngaqala ukuya lapho. Lapho ngafunda izisekelo okokuqala, ngabe sengiqala ukuyifunda ku-inthanethi, ezifundweni ezahlukahlukene.

Ekuqaleni, kwakukhona kuphela inkambo kaKonstantin Vorontsov ngesiRashiya, futhi indlela yokufundisa yayiqinile: yayiqukethe amagama amaningi, futhi kwakukhona amafomula amaningi ezincazelweni. Kumfundi webanga lesishiyagalombili lokhu kwakunzima kakhulu, kodwa manje, ngokuqondile ngenxa yokuthi ngadlula esikoleni esinjalo ekuqaleni, imigomo ayifaki ubunzima kimi ekusebenzeni ezinkingeni zangempela.

- Zingakanani izibalo okudingeka uzazi ukuze usebenze ne-AI? Ingabe lukhona ulwazi olwanele oluvela kukharikhulamu yesikole?
— Ngezindlela eziningi, i-ML isekelwe emiqondweni eyisisekelo yesikole emabangeni 10-11, i-algebra yomugqa eyisisekelo kanye nokuhlukaniswa. Uma sikhuluma ngokukhiqiza, ngezinkinga zobuchwepheshe, ngakho-ke ngezindlela eziningi izibalo azidingeki; izinkinga eziningi zixazululwa ngokuzama nangephutha. Kodwa uma sikhuluma ngocwaningo, lapho kwakhiwa ubuchwepheshe obusha, akukho ndawo ngaphandle kwezibalo. Izibalo ziyadingeka ezingeni eliyisisekelo, okungenani ukwazi ukusebenzisa i-matrix noma, uma kuqhathaniswa, ubale okuphuma kokunye. Azikho izibalo zokubaleka lapha.

- Ngokubona kwakho, ingabe kukhona umfundi onengqondo ehlaziyayo yemvelo angakwazi ukuxazulula izinkinga ze-ML?
- Yebo. Uma umuntu azi ukuthi yini elele enhliziyweni ye-ML, uma azi ukuthi idatha yakhiwe kanjani futhi eqonda amaqhinga ayisisekelo noma ama-hacks, ngeke adinge izibalo, ngoba amathuluzi amaningi omsebenzi asevele ebhalwe ngabanye abantu. Konke kuncike ekutholeni amaphethini. Kodwa konke, yebo, kuncike emsebenzini.

- Iyiphi into enzima kakhulu ekuxazululeni izinkinga namacala e-ML?
- Umsebenzi omusha ngamunye uyinto entsha. Ukube inkinga ibivele ikhona ngendlela efanayo, bekungeke kudingeke ukuthi ixazululwe. Ayikho i-algorithm yendawo yonke. Kunomphakathi omkhulu wabantu abaqeqesha amakhono abo okuxazulula izinkinga, abasho ukuthi bazixazulule kanjani izinkinga, futhi bachaze izindaba zokunqoba kwabo. Futhi kuyathakazelisa kakhulu ukulandela ingqondo yabo, imibono yabo.

— Yiziphi izimo nezinkinga othanda kakhulu ukuzixazulula?
- Ngisebenza ngokukhethekile kulimi lwekhompyutha, nginentshisekelo emibhalweni, izinkinga zokuhlukaniswa, ama-chatbots, njll.

- Ingabe uvame ukubamba iqhaza kuma-hackathons e-AI?
- Ama-Hackathons, empeleni, awuhlelo oluhlukile lwama-Olympiad. I-Olympiad inesethi yezinkinga ezivaliwe, ezinezimpendulo ezaziwayo umhlanganyeli okufanele aziqagele. Kodwa kunabantu abangasebenzi kahle emisebenzini evaliwe, kodwa bahlukanisa wonke umuntu ngokuvuleka. Ngakho ungakwazi ukuhlola ulwazi lwakho ngezindlela ezahlukene. Ezinkingeni ezivulekile, ngezinye izikhathi ubuchwepheshe budalwa kusukela ekuqaleni, imikhiqizo ithuthukiswa ngokushesha, futhi ngisho nabahleli ngokuvamile abazi impendulo efanele. Sivame ukubamba iqhaza kuma-hackathons, futhi ngalokhu singathola imali. Kuyathakazelisa lokhu.

- Ungazuza malini kulokhu? Uyisebenzisa kanjani imali yakho yomklomelo?
- Mina nomngane wami sabamba iqhaza ku-hackathon ye-VKontakte, lapho senza khona isicelo sokufuna imidwebo eHermitage. Isethi yama-emojis nezithonjana zavezwa esikrinini sefoni, kwakudingeka kutholwe isithombe usebenzisa le sethi, ifoni yakhonjwa esithombeni, yaqashelwa kusetshenziswa amanethiwekhi emizwa futhi, uma impendulo ilungile, kwanikezwa amaphuzu. Sijabule futhi saba nentshisekelo yokuthi sikwazile ukwenza uhlelo lokusebenza olusivumele ukuthi sibone umdwebo kudivayisi ephathwayo. Besisendaweni yokuqala, kodwa ngenxa yesimo esisemthethweni saphuthelwa umklomelo wama-ruble ayizinkulungwane ezingama-500. Kuyihlazo, kodwa akuyona into esemqoka leyo.

Ngaphezu kwalokho, wabamba iqhaza emncintiswaneni we-Sberbank Data Science Journey, lapho athatha khona indawo yesi-5 futhi wazuza ama-ruble ayizinkulungwane ezingu-200. Kowokuqala bakhokha isigidi, kowesibili izinkulungwane ezingamakhulu amahlanu. Izimali zemiklomelo ziyahlukahluka, futhi manje ziyakhula. Ukuba phezulu, ungathola izinkulungwane eziyi-500 kuye kwezingama-100. Ngilondoloza imali yomklomelo wemfundo, lokhu kungumnikelo wami wekusasa, imali engiyisebenzisa ekuphileni kwansuku zonke, ngiyaziholela.

- Yini ejabulisa kakhulu - ama-hackathon angawodwa noma eqembu?
- Uma sikhuluma ngokwakha umkhiqizo, kufanele kube yiqembu; umuntu oyedwa ngeke akwenze. Uzomane akhathale futhi adinga ukusekelwa. Kodwa uma sikhuluma, isibonelo, nge-hackathon ye-AI ​​Academy, khona-ke umsebenzi ulinganiselwe, asikho isidingo sokudala umkhiqizo. Intshisekelo lapho ihlukile - ukudlula omunye umuntu naye osathuthukayo kule ndawo.

- Uhlela ukuthuthuka kanjani? Uwubona kanjani umsebenzi wakho?
— Manje umgomo oyinhloko uwukulungiselela umsebenzi wakho wesayensi ongathi sína, ucwaningo, ukuze ubonakale ezingqungqutheleni eziholayo njengezingqungquthela ze-NeurIPS noma ze-ICML - ML eziba khona emazweni ahlukahlukene omhlaba. Umbuzo we-career uvulekile, bheka ukuthi i-ML ithuthuke kanjani kule minyaka emi-5 edlule. Ishintsha ngokushesha, manje sekunzima ukubikezela ukuthi kuzokwenzekani ngokulandelayo. Futhi uma sikhuluma ngemibono nezinhlelo ngaphandle komsebenzi wesayensi, mhlawumbe ngingazibona ngisohlotsheni oluthile lwephrojekthi yami, isiqalo emkhakheni we-AI ne-ML, kodwa lokhu akuqinisekile.

- Ngokubona kwakho, yikuphi ukulinganiselwa kobuchwepheshe be-AI?
- Yebo, ngokuvamile, uma sikhuluma nge-AI njengento enohlobo oluthile lobuhlakani, icubungula idatha, khona-ke, esikhathini esizayo esiseduze, kuyoba uhlobo oluthile lokuqwashisa umhlaba osizungezile. Uma sikhuluma ngamanethiwekhi e-neural kuzilimi zekhompyutha, isibonelo, sizama ukwenza imodeli endaweni ethile, isibonelo, ulimi, ngaphandle kokunikeza imodeli ukuqonda komongo mayelana nomhlaba wethu. Okusho ukuthi, uma sikwazi ukuhlanganisa lokhu ku-AI, sizokwazi ukudala amamodeli ezingxoxo, ama-chat bots angeke azi amamodeli olimi kuphela, kodwa azoba nombono futhi azi amaqiniso esayensi. Futhi yilokhu engingathanda ukukubona esikhathini esizayo.

Phela, i-Academy of Artificial Intelligence okwamanje iqashela izingane zesikole i-hackathon entsha. Imali yomklomelo nayo inkulu, futhi umsebenzi walo nyaka uthakazelisa nakakhulu - uzodinga ukwakha i-algorithm ebikezela ulwazi lomdlali ngokusekelwe kwizibalo zomdlalo owodwa we-Dota 2. Ukuze uthole imininingwane, vakashela ku- lesi sixhumanisi.

Source: www.habr.com