I-AI, abantwana besikolo kunye namabhaso amakhulu: indlela yokwenza umatshini wokufunda kwibanga lesi-8

Hayi Habr!

Singathanda ukuthetha malunga nendlela engaqhelekanga yokufumana imali yabakwishumi elivisayo njengokuthatha inxaxheba kwii-hackathons. Oku kuluncedo ngokwasezimalini kwaye kukuvumela ukuba usebenzise ulwazi olufumene esikolweni nangokufunda iincwadi ezikrelekrele.

Umzekelo olula ngunyaka ophelileyo we-Artificial Intelligence Academy hackathon yabantwana besikolo. Abathathi-nxaxheba balo kwafuneka baqikelele umphumo womdlalo weDota 2. Ophumelele ukhuphiswano nguAlexander Mamaev, umfundi weshumi waseChelyabinsk. I-algorithm yakhe ichaze ngokuchanekileyo iqela eliphumeleleyo lokulwa. Ngenxa yoko, uAlexander wafumana imali eninzi yebhaso - i-ruble eyi-100 lamawaka.

Indlela u-Alexander Mamaev wasebenzisa ngayo imali yebhaso, luphi ulwazi umfundi ongenalo ukusebenza kunye ne-ML, kwaye yintoni isalathiso kwintsimi ye-AI ecinga ukuba inomdla kakhulu - umfundi uxelele udliwano-ndlebe.

— Khawusixelele ngawe, uye waba nomdla njani kwi-AI? Ngaba kwakunzima ukungena kumxholo?
— Ndineminyaka eli-17 ubudala, ndiyagqiba kulo nyaka yaye ndisandul’ ukufuduka eChelyabinsk ndiye eDolgoprudny, ekufutshane naseMoscow. Ndifunda eKapitsa Physics and Technology Lyceum, esi sesinye sezona zikolo zibalaseleyo kwingingqi yaseMoscow. Ndiyakwazi ukuqasha indawo yokuhlala, kodwa ndihlala kwisikolo sokuhlala esikolweni, kungcono kwaye kulula ukunxibelelana nabantu abavela kwi-lyceum.

Ukuqala kwam ukuva nge-AI kunye ne-ML mhlawumbi ngo-2016, xa uPrisma ebonakala. Emva koko ndandikwibanga lesi-8 kwaye ndandisenza inkqubo ye-olympiad, ndaya kwii-olympiads kwaye ndafumanisa ukuba sasine-ML meetups edolophini. Ndandinomdla wokuyifumanisa, ukuqonda indlela esebenza ngayo, kwaye ndaqalisa ukuya apho. Apho ndafunda iziseko okokuqala, emva koko ndaqala ukuyifunda kwi-Intanethi, kwiikhosi ezahlukeneyo.

Ekuqaleni, kwakukho kuphela ikhosi evela kuKonstantin Vorontsov ngesiRashiya, kwaye indlela yokufundisa yayingqongqo: iqulethe amagama amaninzi, kwaye kukho iifomyula ezininzi kwiinkcazo. Kumfundi webanga lesibhozo oku kwakunzima kakhulu, kodwa ngoku, ngokuchanekileyo ngenxa yokuba ndahamba kwisikolo esinjalo ekuqaleni, imimiselo ayibangeli ubunzima kum ekusebenzeni kwiingxaki zokwenyani.

— Kufuneka wazi kangakanani imathematika ukuze usebenze nge-AI? Ngaba kukho ulwazi olwaneleyo oluvela kwikharityhulam yesikolo?
— Ngeendlela ezininzi, i-ML isekelwe kwiikhonsepthi ezisisiseko zesikolo kumabakala 10-11, ialjebra esisiseko ehambelana nomgca nokwahlulahlula. Ukuba sithetha ngemveliso, malunga neengxaki zobugcisa, ngoko ngeendlela ezininzi imathematika ayidingeki; iingxaki ezininzi zisonjululwa ngokulula kunye nempazamo. Kodwa ukuba sithetha ngophando, xa iteknoloji entsha yenziwa, ngoko akukho ndawo ngaphandle kwemathematika. IMathematika iyadingeka kwinqanaba elisisiseko, ubuncinci ukwazi indlela yokusebenzisa imatriksi okanye, xa sithelekisa, ukubala izinto eziphuma kuzo. Akukho kubaleka imathematika apha.

- Ngokombono wakho, ngaba nawuphi na umfundi onengqondo yohlalutyo lwendalo angazisombulula iingxaki zeML?
- Ewe. Ukuba umntu uyayazi into elele entliziyweni ye-ML, ukuba uyazi indlela idatha eyakhiwe ngayo kwaye uyayiqonda amaqhinga asisiseko okanye i-hacks, akayi kufuna izibalo, kuba ezininzi izixhobo zomsebenzi sele zibhaliwe ngabanye abantu. Konke kuhla ekufumaneni iipateni. Kodwa yonke into, ngokuqinisekileyo, ixhomekeke kumsebenzi.

— Yintoni eyona nto inzima kakhulu ekusombululeni iingxaki zeML kunye namatyala?
— Umsebenzi ngamnye omtsha yinto entsha. Ukuba ingxaki ibisele ikhona ngendlela efanayo, bekungayi kufuneka ukuba isonjululwe. Akukho algorithm yendalo yonke. Kukho uluntu olukhulu lwabantu abaqeqesha izakhono zabo zokusombulula iingxaki, baxelele indlela abazisombulule ngayo iingxaki, kwaye bachaze amabali oloyiso lwabo. Kwaye kunomdla kakhulu ukulandela ingqiqo yabo, izimvo zabo.

— Zeziphi iimeko kunye neengxaki onomdla kakhulu ekuzisombululeni?
- Ndisebenza ngokukhethekileyo kwiilwimi zokudibanisa, ndinomdla kwiitekisi, iingxaki zokuhlela, ii-chatbots, njl.

- Ngaba uhlala uthatha inxaxheba kwii-hackathons ze-AI?
- IiHackathons, enyanisweni, yinkqubo eyahlukileyo yeOlympiads. I-Olympiad ineseti yeengxaki ezivaliweyo, ezineempendulo ezaziwayo ekufuneka umthathi-nxaxheba aziqikelele. Kodwa kukho abantu abangalunganga kwimisebenzi evaliweyo, kodwa bakrazula wonke umntu kwindawo evulekileyo. Ngoko unokuvavanya ulwazi lwakho ngeendlela ezahlukeneyo. Kwiingxaki ezivulekileyo, itekhnoloji ngamanye amaxesha yenziwa ukusuka ekuqaleni, iimveliso ziphuhliswa ngokukhawuleza, kwaye nabaququzeleli bahlala bengayazi impendulo echanekileyo. Sihlala sithatha inxaxheba kwii-hackathons, kwaye ngale nto sinokufumana imali. Oku kunomdla.

- Unokufumana imali engakanani kule nto? Uyisebenzisa njani imali yakho yebhaso?
- Mna nomhlobo wam sathatha inxaxheba kwi-hackathon ye-VKontakte, apho senza isicelo sokukhangela imizobo eHermitage. Iseti ye-emojis kunye nee-emoticon ziboniswe kwisikrini sefowuni, kwakuyimfuneko ukufumana umfanekiso usebenzisa le seti, ifowuni yayibhekiselwe kumfanekiso, yaqatshelwa ngokusebenzisa i-neural networks kwaye, ukuba impendulo yayichanekile, amanqaku anikezelwa. Sasonwabile kwaye sinomdla ukuba sikwazile ukwenza isicelo esivumela ukuba siqaphele umzobo kwisixhobo esiphathwayo. Besibambekile kwindawo yokuqala, kodwa ngenxa yenkqubo esemthethweni saphoswa ngumvuzo we-500 lamawaka eeruble. Lihlazo, kodwa ayisiyiyo eyona nto iphambili leyo.

Ukongezelela, wathatha inxaxheba kukhuphiswano lweSberbank Data Science Journey, apho wathatha indawo ye-5 waza wafumana i-ruble eyi-200 lamawaka. Kowokuqala bahlawule isigidi, kowesibini amawaka angama-500. Iimali zamabhaso ziyahluka, kwaye ngoku ziyanda. Ukuba phezulu, unokufumana i-100 ukuya kwi-500 lamawaka. Ndigcina ibhaso lemali yemfundo, eli ligalelo lam kwikamva, imali endiyichitha kubomi bemihla ngemihla, ndiyazisebenzela.

- Yintoni eyona nto inomdla - iihackathons zomntu okanye iqela?
— Ukuba sithetha ngokuphuhlisa imveliso, kufuneka ibe liqela; umntu omnye akanakuyenza loo nto. Uya kusuka adinwe kwaye afune inkxaso. Kodwa ukuba sithetha, umzekelo, malunga ne-hackathon ye-AI ye-Academy, ngoko umsebenzi ulinganiselwe, akukho mfuneko yokwenza imveliso. Umdla wakhona wahlukile - ukudlula omnye umntu ophuhlayo kule ndawo.

— Uceba ukuphuhlisa njani ngakumbi? Uwubona njani umsebenzi wakho?
- Ngoku injongo ephambili kukulungiselela umsebenzi wakho onzulu wesayensi, uphando, ukuze ubonakale kwiinkomfa ezikhokelayo njengeNeurIPS okanye i-ICML - iinkomfa zeML ezenzeka kumazwe ahlukeneyo ehlabathi. Umbuzo wekhondo uvulekile, jonga indlela iML ephuhlise ngayo kule minyaka mihlanu idlulileyo. Itshintsha ngokukhawuleza, ngoku kunzima ukuqikelela ukuba kuya kwenzeka ntoni emva koko. Kwaye ukuba sithetha ngeengcinga kunye nezicwangciso ngaphandle komsebenzi wesayensi, ngoko mhlawumbi ndingazibona kuhlobo oluthile lweprojekthi yam, isiqalo kwintsimi ye-AI kunye ne-ML, kodwa oku akuqinisekanga.

- Ngokombono wakho, yeyiphi imida yetekhnoloji ye-AI?
- Ewe, ngokubanzi, ukuba sithetha nge-AI njengento enobukrelekrele, inkqubo yedatha, ngoko, kwixesha elizayo, kuya kuba luhlobo oluthile lolwazi lwehlabathi elisingqongileyo. Ukuba sithetha nge-neural networks kwi-computational linguistics, umzekelo, sizama ukwenza imodeli yendawo ethile, umzekelo, ulwimi, ngaphandle kokunika imodeli ukuqonda umxholo malunga nehlabathi lethu. Okokuthi, ukuba siyakwazi ukubandakanya oku kwi-AI, siya kukwazi ukudala iimodeli zengxoxo, iibhothi zokuxoxa ezingayi kukwazi nje imizekelo yolwimi, kodwa ziya kuba nembono kwaye zazi iinyani zesayensi. Kwaye le nto ndingathanda ukuyibona kwixesha elizayo.

Ngendlela, i-Academy of Artificial Intelligence ngoku igaya abantwana besikolo kwi-hackathon entsha. Imali yebhaso nayo ininzi, kwaye umsebenzi walo nyaka unomdla ngakumbi - kuya kufuneka ukuba wakhe i-algorithm eqikelela amava omdlali ngokusekelwe kwizibalo zomdlalo omnye weDota 2. Ukuze ufumane iinkcukacha, yiya esi sixhobo.

umthombo: www.habr.com