Ibhaso elibizwa ngokuba ngu-Ilya Segalovich. Ibali malunga nesayensi yekhompyuter kunye noshicilelo ngexesha lokusungulwa

Ibhaso elibizwa ngokuba ngu-Ilya Segalovich. Ibali malunga nesayensi yekhompyuter kunye noshicilelo ngexesha lokusungulwa

Namhlanje siphehlelela imbasa yezenzululwazi ebizwa ngokuba ngu-Ilya Segalovich umzekelo. Iza kuwongwa ngempumelelo kwinkalo yesayensi yekhompyuter. Abafundi abangekabinazidanga kunye nabasele benezidanga banokungenisa ezabo izicelo zokufumana ibhaso okanye batyumbe iisuphavayiza zenzululwazi. Abaphumeleleyo baya kukhethwa ngabameli boluntu lwezemfundo kunye neYandex. Iikhrayitheriya zokukhetha eziphambili: iimpapasho kunye neentetho kwiinkomfa, kunye negalelo kuphuhliso loluntu.

Umsitho wokuqala wokuwonga uya kuba ngo-Epreli. Njengenxalenye yebhaso, oososayensi abancinci baya kufumana i-ruble eyi-350 lamawaka, kwaye ngaphezu koko, baya kukwazi ukuya kwinkomfa yamazwe ngamazwe, basebenze nomcebisi kwaye baqhube i-internship kwisebe lophando laseYandex. Abaphathi bezenzululwazi baya kufumana i-ruble eyi-700 lamawaka.

Ngomsitho wokuphehlelelwa kwebhaso, sigqibe kwelokuba sithethe apha kuHabrΓ© malunga neendlela zokuphumelela kwihlabathi lesayensi yekhompyuter. Abanye abafundi bakaHabr sele beqhelene nezi ndlela, ngelixa abanye banokuba noluvo olungeyonyani ngabo. Namhlanje siza kubamba le gap - siya kuchukumisa kuzo zonke izihloko eziphambili, kubandakanywa amanqaku, iinkomfa, iiseti zedatha kunye nokudluliselwa kweengcamango zesayensi kwiinkonzo.

Kwizazinzulu kwinkalo yesayensi yekhompyuter, eyona nqobo iphambili yempumelelo kukupapashwa komsebenzi wabo wenzululwazi kwenye yeenkomfa eziphezulu zamazwe ngamazwe. Le yinto yokuqala "yokukhangela" yokuqaphela umsebenzi womphandi. Ngokomzekelo, kwintsimi yokufunda ngomatshini ngokubanzi, iNkomfa yeZizwe ngezizwe malunga nokuFunda koMatshini (ICML) kunye neNkomfa kwiiNkqubo zeNeural Information Processing Systems (NeurIPS, eyayiyi-NIPS) yahlula. Kukho iinkomfa ezininzi kwiindawo ezithile ze-ML, ezifana nombono wekhompyutheni, ukufumana ulwazi, iteknoloji yokuthetha, ukuguqulelwa komatshini, njl.

Kutheni upapasha izimvo zakho

Abantu abakude nesayensi yekhompyuter banokuba nembono engalunganga yokuba kungcono ukugcina ezona zimvo zixabisekileyo ziyimfihlo kwaye sizame ukufumana inzuzo ngokungafaniyo. Nangona kunjalo, imeko yokwenyani kwintsimi yethu yahluke ngokupheleleyo. Igunya lenzululwazi ligwetywa ngokubaluleka kwemisebenzi yakhe, ngokuphindaphindiweyo amanqaku akhe acatshulwa ngezinye izazinzulu (i-citation index). Olu luphawu olubalulekileyo lomsebenzi wakhe. Umphengululi unyusa ileli yobugcisa, esiba nembeko ngakumbi kwindawo ahlala kuyo, kuphela ukuba usoloko evelisa incwadi eyomeleleyo epapashwayo, edume, neyila isiseko somsebenzi wezinye izazinzulu.

Amanqaku amaninzi aphezulu (mhlawumbi amaninzi) sisiphumo sentsebenziswano phakathi kwabaphandi kwiiyunivesithi ezahlukeneyo kunye neenkampani emhlabeni jikelele. Umzuzu obalulekileyo noxabiseke kakhulu kumsebenzi womphandi kuxa efumana ithuba lokufumana nokuhluza izimvo ngokwakhe ngokusekelwe kumava akhe – kodwa nasemva koku, oogxa bakhe bayaqhubeka nokumnika uncedo oluxabisekileyo. Izazinzulu ziyancedana ukuphuhlisa izimvo, ukubhala amanqaku ngentsebenziswano-kwaye ngokukhona igalelo lenzululwazi kwisayensi, kokukhona kulula kuye ukufumana abantu abanengqondo efanayo.

Okokugqibela, ukuxinana kunye nokufumaneka kolwazi ngoku kukhulu kangangokuba abaphandi abahlukeneyo ngaxeshanye beza neengcamango zenzululwazi ezifanayo (kwaye zixabiseke ngokwenene). Ukuba awuwushicileli umbono wakho, omnye umntu ngokuqinisekileyo uya kukupapasha. "Umnqobi" kaninzi akayena lowo weza ne-innovation encinci ngaphambili, kodwa lowo wayipapasha kancinci ngaphambili. Okanye - lowo ukwazile ukuveza umbono ngokupheleleyo, ngokucacileyo nangokukholisayo kangangoko kunokwenzeka.

Ibhaso elibizwa ngokuba ngu-Ilya Segalovich. Ibali malunga nesayensi yekhompyuter kunye noshicilelo ngexesha lokusungulwa

Amanqaku kunye neeseti zedatha

Ngoko ke, inqaku lenzululwazi lakhiwe malunga nombono ophambili ophakanyiswa ngumphandi. Lo mbono ligalelo lakhe kwisayensi yekhompyuter. Inqaku liqala ngenkcazo yengcamango, eyenziwe kwizivakalisi ezimbalwa. Oku kulandelwa yintshayelelo echaza uluhlu lweengxaki ezisonjululwe ngoncedo lwenguqulelo ecetywayo. Inkcazo nentshayelelo zidla ngokubhalwa ngolwimi olulula oluqondakalayo kubaphulaphuli abaninzi. Emva kwentshayelelo, kuyimfuneko ukwenza ngokusesikweni iingxaki ezivezwa ngolwimi lwemathematika kunye nokwazisa ngokubhaliweyo okungqongqo. Emva koko, usebenzisa i-notations eyaziswayo, kufuneka udale inkcazo ecacileyo kunye nebanzi yento ecetywayo yokutsha, kwaye uchonge ukungafani kweendlela zangaphambili, ezifanayo. Zonke iinkcazo zethiyori kufuneka zixhaswe ngokubhekisele kubungqina obuqulunqwe ngaphambili, okanye zingqinwe ngokuzimeleyo. Oku kunokwenziwa ngentelekelelo ethile. Ngokomzekelo, unokunika ubungqina betyala xa kukho inani elingenasiphelo ledatha yoqeqesho (imeko ecacileyo engenakufikeleleka) okanye bazimele ngokupheleleyo omnye komnye. Ngasekupheleni kwenqaku, isazinzulu sithetha ngeziphumo zovavanyo ekwazileyo ukuzifumana.

Ibhaso elibizwa ngokuba ngu-Ilya Segalovich. Ibali malunga nesayensi yekhompyuter kunye noshicilelo ngexesha lokusungulwa

Ukuze ababuyekezi abaqeshwe ngabaququzeleli beenkomfa babenakho ukuvuma iphepha, kufuneka libe nempawu enye okanye ngaphezulu. Umba ophambili owandisa amathuba okuvunywa bubunzululwazi obutsha bembono ecetywayo. Ngokuqhelekileyo, ubutsha buvavanywa ngokumalunga neengcamango esele zikhona - kwaye umsebenzi wokuwuvavanya awuqhutywa ngumhloli, kodwa ngumbhali wenqaku ngokwakhe. Ngokufanelekileyo, umbhali kufuneka axelele ngokubanzi kwinqaku malunga neendlela ezikhoyo kwaye, ukuba kunokwenzeka, zibonise njengeemeko ezikhethekileyo zendlela yakhe. Ngaloo ndlela, isazinzulu sibonisa ukuba iindlela ezamkelweyo azisoloko zisebenza, ukuba waziqhelanisa kwaye wacebisa uqulunqo olubanzi, oluguquguqukayo kwaye ngoko ke lusebenza ngakumbi. Ukuba into entsha ayinakuphikiswa, kungenjalo ababuyekezi bavavanya inqaku ngokungakhethi-umzekelo, banokungasihoyi isiNgesi.

Ukuqinisa ubutsha, kuluncedo ukubandakanya uthelekiso kunye neendlela ezikhoyo kwiseti yedatha enye okanye ngaphezulu. Ngamnye kubo kufuneka avuleleke kwaye amkelwe kwimeko yemfundo. Ngokomzekelo, kukho indawo yokugcina umfanekiso we-ImageNet kunye neenkcukacha zamaziko afana ne-Modified National Institute of Standards and Technology (MNIST) kunye ne-CIFAR (i-Canadian Institute for Advanced Research). Ubunzima kukuba i-dataset "yemfundo" enjalo ihlala ihluke kwisakhiwo somxholo ukusuka kwidatha yangempela ejongene nayo ishishini. Idatha eyahlukileyo ithetha iziphumo ezahlukeneyo zendlela ecetywayo. Izazinzulu ezisebenza ngokuyinxenye kushishino zizama ukuthathela ingqalelo oku kwaye ngamanye amaxesha zifake izilandulo ezinje β€œkwidatha yethu isiphumo sinje nanje, kodwa kwidathasethi yoluntu-enje nanje.”

Kwenzeka ukuba indlela ecetywayo "ilungiselelwe" ngokupheleleyo kwi-database evulekile kwaye ayisebenzi kwidatha yangempela. Ungayilwa le ngxaki ixhaphakileyo ngokuvula iiseti zedatha ezintsha, ezimele ngakumbi, kodwa sihlala sithetha ngomxholo wabucala apho iinkampani zingenalo ilungelo lokuwuvula. Kwezinye iimeko, benza (ngamanye amaxesha banzima kwaye babuhlungu) ukungaziwa kwedatha - basusa naziphi na iziqwenga ezikhomba umntu othile. Umzekelo, ubuso kunye namanani kwiifoto ziyacinywa okanye zenziwe ukuba zingafundeki. Ukongeza, ukuze isethi yedatha ingafumaneki kuphela kuye wonke umntu, kodwa ibe ngumgangatho phakathi kwezazinzulu apho kukulungele ukuthelekisa izimvo, akufuneki nje ukupapasha, kodwa nokubhala inqaku elahlukileyo elicatshuliweyo malunga nalo. yona kunye neengenelo zayo.

Kubi ngakumbi xa kungekho datha evulekileyo kwisihloko esifundwayo. Emva koko umhlalutyi unokwamkela kuphela iziphumo ezivezwe ngumbhali ngokholo. Ngokwethiyori, umbhali unokude azicingele kakhulu kwaye angabonwa, kodwa kwimeko yemfundo ayinakwenzeka, kuba ichasene nomnqweno woninzi lwezazinzulu ukuphuhlisa isayensi.

Kwiindawo ezininzi ze-ML, kubandakanywa nombono wekhompyutheni, kuqhelekile ukuba unamathele amakhonkco kwikhowudi (ngokuqhelekileyo kwiGitHub) kunye namanqaku. Amanqaku ngokwawo aqulathe ikhowudi encinci kakhulu okanye yipseudocode. Kwaye apha, kwakhona, kuvela ubunzima ukuba inqaku libhalwe ngumphandi ovela kwinkampani, kungekhona eyunivesithi. Ngokungagqibekanga, ikhowudi ebhalwe kwiqumrhu okanye isiqalo ibhalwe NDA. Abaphandi kunye noogxa babo kufuneka basebenze nzima ukuze bahlule ikhowudi ehambelana nengcamango echazwe kwi-repositories yangaphakathi kwaye ngokuqinisekileyo ivaliwe.

Ithuba lokupapashwa likwaxhomekeke kukufaneleka kwesihloko esikhethiweyo. Ukufaneleka kuchazwe kakhulu kwiimveliso kunye neenkonzo: ukuba i-corporation okanye i-startup inomdla wokwakha inkonzo entsha okanye ukuphucula esele ikhona ngokusekelwe kwingcamango evela kwinqaku, oko kukudibanisa.

Ibhaso elibizwa ngokuba ngu-Ilya Segalovich. Ibali malunga nesayensi yekhompyuter kunye noshicilelo ngexesha lokusungulwa

Njengoko sele kukhankanyiwe, amaphepha esayensi yekhompyuter awafane abhalwe yedwa. Kodwa njengomthetho, omnye wababhali uchitha ixesha elininzi kunye nomzamo kunabanye. Igalelo lakhe kwizinto ezintsha zenzululwazi lelona likhulu. Kuluhlu lwababhali, umntu onjalo uboniswa kuqala - kwaye kwixesha elizayo, xa ebhekisela kwinqaku, banokukhankanya kuphela (umzekelo, "Ivanov et al" - "Ivanov nabanye" iguqulelwe kwisiLatini). Nangona kunjalo, igalelo labanye libaluleke kakhulu - kungenjalo akunakwenzeka ukuba ube kuluhlu lwababhali.

Inkqubo yokuphonononga

Amaphepha adla ngokuyeka ukwamkelwa kwiinyanga eziliqela ngaphambi kwenkomfa. Emva kokuba inqaku lingenisiwe, abahlaziyi baneveki ezi-3-5 zokufunda, ukuvavanya, kunye nokuphawula kuyo. Oku kwenzeka ngokwenkqubo eyodwa eyimfama, xa ababhali bengaboni amagama abahlaziyi, okanye iimfama eziphindwe kabini, xa abahlaziyi ngokwabo bengaboni amagama ababhali. Inketho yesibini ithathwa njengokungakhethi cala: amaphepha amaninzi enzululwazi abonise ukuba ukuthandwa kombhali kunempembelelo kwisigqibo somhlalutyi. Ngokomzekelo, unokucinga ukuba isazinzulu esinenani elikhulu lamanqaku asele apapashiwe yinto ephambili efanelwe ukulinganisa okuphezulu.

Ngaphezu koko, nakwimeko yokungaboni kakuhle kabini, umhloli uya kuqikelela umbhali ukuba basebenza kwintsimi efanayo. Ukongeza, ngexesha lophononongo, inqaku linokuthi sele lipapashwe kwi-database ye-arXiv, eyona ndawo inkulu yamaphepha enzululwazi. Abaququzeleli benkomfa abakuthinteli oku, kodwa bacebisa ukuba kusetyenziswe isihloko esahlukileyo kunye nesicatshulwa esahlukileyo kupapasho lwe-arXiv. Kodwa ukuba inqaku lithunyelwe apho, akusayi kuba nzima ukulifumana.

Kuhlala kukho abahlaziyi abaninzi abaphonononga inqaku. Omnye wabo unikwe indima ye-meta-reviewer, omele ahlaziye kuphela izigwebo zabalingane bakhe kwaye enze isigqibo sokugqibela. Ukuba abaphononongi abavumelani nenqaku, uphononongo lwe-meta unokulufunda ukuze luphelele.

Ngamanye amaxesha, emva kokuphonononga ukulinganisa kunye namazwana, umbhali unethuba lokungena kwingxoxo kunye nombuyekezi; kukho nethuba lokumqinisekisa ukuba atshintshe isigqibo sakhe (nangona kunjalo, inkqubo enjalo ayisebenzi kuzo zonke iinkomfa, kwaye akunakwenzeka ukuba nefuthe elibi kwisigwebo). Kwingxoxo, awukwazi ukubhekisa kweminye imisebenzi yesayensi, ngaphandle kwaleyo sele ikhankanyiwe kwinqaku. Unako kuphela "ukunceda" umhlalutyi aqonde ngcono umxholo wenqaku.

Ibhaso elibizwa ngokuba ngu-Ilya Segalovich. Ibali malunga nesayensi yekhompyuter kunye noshicilelo ngexesha lokusungulwa

Iinkomfa kunye neejenali

Amanqaku enzululwazi yekhompyutha asoloko ethunyelwa kwiinkomfa kunakwiijenali zenzululwazi. Oku kungenxa yokuba upapasho lwejenali luneemfuno ekunzima ukuhlangabezana nazo, kwaye inkqubo yokuphononongwa koontanga inokuthatha iinyanga okanye iminyaka. Inzululwazi yeKhompyutha yintsimi ehamba ngokukhawuleza kakhulu, ngoko ke ababhali abaqhelekanga ukulinda ixesha elide ukupapashwa. Nangona kunjalo, inqaku esele lamkelwe kwinkomfa linokongeza emva koko (umzekelo, ngokubonisa iziphumo ezineenkcukacha) kwaye lipapashwe kwijenali apho izithintelo zendawo zingekho ngqongqo kangako.

Iziganeko kwinkomfa

Ifomathi yobukho bababhali bamanqaku avunyiweyo kwinkomfa igqitywa ngabahlalutyi. Ukuba inqaku linikwe ukukhanya okuluhlaza, ngoko udla ngokunikwa indawo yokumisa ipowusta. Ipowusta sisilayidi esingatshintshiyo esinesishwankathelo senqaku kunye nemifanekiso. Amanye amagumbi enkomfa azaliswe yimigca emide yeepowusta. Umbhali uchitha ixesha elininzi kufutshane nepowusta yakhe, enxibelelana noososayensi abanomdla kweli nqaku.

Ibhaso elibizwa ngokuba ngu-Ilya Segalovich. Ibali malunga nesayensi yekhompyuter kunye noshicilelo ngexesha lokusungulwa

Ibhaso elibizwa ngokuba ngu-Ilya Segalovich. Ibali malunga nesayensi yekhompyuter kunye noshicilelo ngexesha lokusungulwa

Olona khetho luhloniphekileyo ngakumbi lokuthatha inxaxheba yintetho yombane. Ukuba abahlalutyi baqwalasela inqaku elifanelwe yingxelo ekhawulezayo, umbhali unikezwa malunga nemizuzu emithathu ukuthetha nabaphulaphuli abaninzi. Kwelinye icala, intetho yombane lithuba elihle lokuxelela ngombono wakho kungekuphela nje kwabo baye banomdla kwipowusta ngokuthanda kwabo. Kwelinye icala, abatyeleli bepowusta esebenzayo balungiselelwe ngakumbi kwaye bantywiliselwe ngakumbi kwisihloko sakho esikhethekileyo kunomphulaphuli ophakathi eholweni. Ke ngoko, kwingxelo ekhawulezayo, kusafuneka ube nexesha lokwazisa abantu ngoku.

Ibhaso elibizwa ngokuba ngu-Ilya Segalovich. Ibali malunga nesayensi yekhompyuter kunye noshicilelo ngexesha lokusungulwa

Ngokuqhelekileyo, ekupheleni kwentetho yabo yombane, ababhali babiza inombolo yepowusta ukuze abaphulaphuli banokuyifumana kwaye baqonde ngcono inqaku.

Ibhaso elibizwa ngokuba ngu-Ilya Segalovich. Ibali malunga nesayensi yekhompyuter kunye noshicilelo ngexesha lokusungulwa

Olona khetho lokugqibela, luhloniphekileyo yipowusta kunye nenkcazo-ntetho epheleleyo yoluvo, xa kungekho mfuneko yokungxama ukubalisa ibali.

Ibhaso elibizwa ngokuba ngu-Ilya Segalovich. Ibali malunga nesayensi yekhompyuter kunye noshicilelo ngexesha lokusungulwa

Kodwa ke, izazinzulu-kubandakanywa nababhali bamanqaku avunyiweyo-beza kwinkomfa elandelayo kungekhona nje ukubonisa. Okokuqala, badla ngokufumana iipowusta ezinxulumene nentsimi yabo ngezizathu ezicacileyo. Kwaye okwesibini, kubalulekile ukuba bandise uluhlu lwabo loqhagamshelwano ngenjongo yomsebenzi odibeneyo wezemfundo kwixesha elizayo. Oku asikuko ukuzingela - okanye, ubuncinane, isigaba sayo sokuqala, esilandelwa ubuncinane ngokutshintshiselana ngeengcamango, uphuhliso kunye nomsebenzi odibeneyo kwinqaku elinye okanye ngaphezulu.

Kwangaxeshanye, unxibelelwano olunemveliso kwinkomfa ephezulu kunzima ngenxa yokunqongophala kwexesha elipheleleyo. Ukuba, emva kosuku lonke oluchithwe kwiintetho kunye neengxoxo kwiipowusta, isazinzulu sigcine amandla akhe kwaye sele soyise i-jet lag, ngoko uya kwelinye lamaqela amaninzi. Zisingathwa ngamaqumrhu - ngenxa yoko, amaqela ahlala enobuntu bokuzingela ngakumbi. Ngelo xesha, iindwendwe ezininzi azisebenzisi konke konke ukufumana umsebenzi omtsha, kodwa, kwakhona, kwinethiwekhi. Ngokuhlwa azikho iingxelo kunye neepowusta - kulula "ukubamba" ingcali onomdla kuyo.

Ibhaso elibizwa ngokuba ngu-Ilya Segalovich. Ibali malunga nesayensi yekhompyuter kunye noshicilelo ngexesha lokusungulwa

Ukusuka kwingcamango ukuya kwimveliso

Inzululwazi yeKhompyutha lelinye lamashishini ambalwa apho umdla weenkampani kunye nokuqalisa unxulunyaniswa ngamandla nemeko yemfundo. I-NIPS, i-ICML kunye nezinye iinkomfa ezifanayo zitsala abantu abaninzi abavela kumashishini, kungekhona nje kwiiyunivesithi. Oku kuqhelekile kwicandelo lesayensi yekhompyuter, kodwa ngokuphambeneyo kwezinye iisayensi ezininzi.

Ngakolunye uhlangothi, ayizizo zonke iingcamango ezichazwe kumanqaku ngokukhawuleza eziya ekudaleni okanye ekuphuculeni iinkonzo. Nangaphakathi kwenkampani enye, umphandi unokucebisa oogxa bakhe kwinkonzo ingcamango ephunyezwe yimigangatho yenzululwazi kwaye afumane ukwala ukuyiphumeza ngenxa yezizathu ezininzi. Enye yazo sele ikhankanywe apha - lo ngumahluko phakathi kwedatha "yezemfundo" apho inqaku libhalwe khona kunye nedatha yangempela. Ukongezelela, ukuphunyezwa kombono kunokulibaziseka, kufuna isixa esikhulu sezibonelelo, okanye kuphuculwe isalathisi esinye kuphela kwiindleko zokuwohloka kwezinye iimethrikhi.

Ibhaso elibizwa ngokuba ngu-Ilya Segalovich. Ibali malunga nesayensi yekhompyuter kunye noshicilelo ngexesha lokusungulwa

Imeko igcinwa yinto yokuba abaninzi abaphuhlisi ngokwabo bancinci yabaphandi. Baya kwiinkomfa, bathetha ulwimi olunye kunye nezifundiswa, bacebise ngezimvo, ngamanye amaxesha bathathe inxaxheba ekwenziweni kwamanqaku (umzekelo, ikhowudi yokubhala), okanye bade benze njengababhali ngokwabo. Ukuba umphuhlisi ugxininiswe kwinkqubo yezemfundo, ulandela oko kwenzekayo kwisebe lophando, ngelizwi - ukuba ubonisa i-counter-movement ukuya kwizazinzulu, ngoko umjikelo wokuguqula iingcamango zesayensi kwizakhono ezintsha zenkonzo zifutshane.

Sinqwenelela bonke abaphandi abancinci amathamsanqa kunye nempumelelo enkulu kumsebenzi wabo. Ukuba esi sithuba asizange sikuxelele nantoni na entsha, ngoko usenokuba sele upapashiwe kwinkomfa ephezulu. Bhalisa kwi iprimiyamu ngokwakho kwaye wonyule abaphathi bezenzululwazi.

umthombo: www.habr.com

Yongeza izimvo