Umklomelo oqanjwe ngo-Ilya Segalovich. Indaba ngesayensi yekhompiyutha kanye nokwethulwa kokushicilelwa

Umklomelo oqanjwe ngo-Ilya Segalovich. Indaba ngesayensi yekhompiyutha kanye nokwethulwa kokushicilelwa

Namuhla sethula umklomelo wesayensi oqanjwe ngo-Ilya Segalovich isig. Izoklonyeliswa ngezimpumelelo emkhakheni wesayensi yamakhompiyutha. Abafundi be-Undergraduate kanye ne-postgraduate bangafaka esabo isicelo somklomelo noma ukhethe abaphathi besayensi. Abaphumelele bazokhethwa ngabamele umphakathi wezemfundo kanye ne-Yandex. Indlela yokukhetha eyinhloko: ukushicilelwa kanye nezethulo ezingqungqutheleni, kanye negalelo ekuthuthukisweni komphakathi.

Umcimbi wokuqala wezindondo uzoba ngo-Ephreli. Njengengxenye yomklomelo, ososayensi abasebasha bazothola ama-ruble ayizinkulungwane ezingama-350, futhi ngaphezu kwalokho, bazokwazi ukuya engqungqutheleni yamazwe omhlaba, basebenze nomeluleki futhi bathole i-internship emnyangweni wocwaningo waseYandex. Abaphathi bezesayensi bazothola ama-ruble ayizinkulungwane ezingama-700.

Ngesikhathi sokwethulwa komklomelo, sinqume ukukhuluma lapha ku-HabrΓ© mayelana nenqubo yokuphumelela emhlabeni wesayensi yamakhompiyutha. Abanye abafundi be-Habr sebejwayelene nalezi zindlela zokunquma, kanti abanye bangase babe nombono oyiphutha ngazo. Namuhla sizovala leli gebe - sizothinta zonke izihloko eziyinhloko, okuhlanganisa izihloko, izingqungquthela, amasethi wedatha kanye nokudluliselwa kwemibono yesayensi kumasevisi.

Kososayensi emkhakheni wesayensi yamakhompiyutha, umbandela oyinhloko wempumelelo ukushicilelwa komsebenzi wabo wesayensi kwenye yezingqungquthela zamazwe ngamazwe eziphezulu. Lena "indawo yokuhlola" yokuqala yokubona umsebenzi womcwaningi. Isibonelo, emkhakheni wokufunda komshini ngokuvamile, iNkomfa Yamazwe Ngamazwe Yokufunda Ngomshini (ICML) kanye neNkomfa Yezinhlelo Zokucubungula Ulwazi Lwe-Neural (NeurIPS, ngaphambilini eyayiyi-NIPS) ziyahlukaniswa. Kunezingqungquthela eziningi ezindaweni ezithile ze-ML, njengokubona ngekhompyutha, ukubuyisa ulwazi, ubuchwepheshe benkulumo, ukuhumusha ngomshini, njll.

Kungani ushicilela imibono yakho

Abantu abakude nesayensi yekhompiyutha bangase babe nombono oyiphutha wokuthi kungcono ukugcina imibono ebaluleke kakhulu iyimfihlo futhi balwele ukuzuza ngokuhluka kwabo. Nokho, isimo sangempela ensimini yethu sihluke ngokuphelele. Igunya likasosayensi lihlulelwa ngokubaluleka kwemisebenzi yakhe, ngokuthi izihloko zakhe zicashunwa kaningi kangakanani abanye ososayensi (inkomba yokucaphuna). Lesi yisici esibalulekile somsebenzi wakhe. Umcwaningi ukhuphula ileli lochwepheshe, ehlonishwa kakhulu emphakathini wakubo, kuphela uma eqhubeka ekhiqiza incwadi eqinile enyatheliswayo, edumile, futhi eyakha isisekelo somsebenzi wabanye ososayensi.

Izindatshana eziningi eziphezulu (mhlawumbe eziningi) ziwumphumela wokubambisana phakathi kwabacwaningi bamanyuvesi ahlukene nezinkampani emhlabeni jikelele. Isikhathi esibalulekile nesibaluleke kakhulu emsebenzini womcwaningi yilapho ethola ithuba lokuthola nokuhlunga imibono eyedwa ngokusekelwe kokuhlangenwe nakho kwakhe - kodwa nangemva kwalokhu, ozakwabo bayaqhubeka nokumnikeza usizo olungenakuqhathaniswa. Ososayensi bayasizana ukuthuthukisa imibono, ukubhala izindatshana ngokubambisana - futhi lapho usosayensi ebamba iqhaza elikhulu kwezesayensi, kuba lula kuye ukuthola abantu abanomqondo ofanayo.

Okokugcina, ukuminyana nokutholakala kolwazi manje kukhulu kangangokuthi abacwaningi abahlukene ngesikhathi esisodwa beza nemibono yesayensi efanayo (futhi eyigugu ngempela). Uma ungawushicileli umbono wakho, othile uzocishe awushicilele wona. β€œOwinile” kaningi akusiyena oqhamuke nokusha kusenesikhathi, kodwa yilowo owashicilela ngaphambi kwesikhathi. Noma - lowo okwazile ukwembula umqondo ngokugcwele, ngokucacile nangokukholisayo ngangokunokwenzeka.

Umklomelo oqanjwe ngo-Ilya Segalovich. Indaba ngesayensi yekhompiyutha kanye nokwethulwa kokushicilelwa

Ama-athikili namasethi edatha

Ngakho-ke, i-athikili yesayensi yakhiwe ngokuzungeza umbono oyinhloko umcwaningi awuhlongoza. Lo mbono uwumnikelo wakhe kwisayensi yekhompyutha. I-athikili iqala ngencazelo yombono, owenziwe ngemisho embalwa. Lokhu kulandelwa isingeniso esichaza ububanzi bezinkinga ezixazululwe ngosizo lokuqamba okusha okuhlongozwayo. Incazelo nesingeniso kuvame ukubhalwa ngolimi olulula oluqondakalayo ezilalelini eziningi. Ngemuva kwesingeniso, kuyadingeka ukwenza ngokusemthethweni izinkinga ezivezwa ngolimi lwezibalo futhi wethule ukuphawula okuqinile. Khona-ke, usebenzisa amanothi ethuliwe, udinga ukwakha isitatimende esicacile nesiphelele somnyombo wokuqamba okusha okuhlongozwayo, futhi ukhombe umehluko ezindleleni zangaphambilini ezifanayo. Zonke izitatimende zetiyori kufanele zisekelwe yizinkomba zobufakazi obahlanganiswa ngaphambilini, noma zifakazelwe ngokuzimela. Lokhu kungenziwa ngokuqagela okuthile. Isibonelo, unganikeza ubufakazi becala uma kunenani elingapheli ledatha yokuqeqeshwa (isimo ngokusobala esingafinyeleleki) noma zizimele ngokuphelele komunye nomunye. Ngasekupheleni kwesihloko, usosayensi ukhuluma ngemiphumela yokuhlola akwazi ukuyithola.

Umklomelo oqanjwe ngo-Ilya Segalovich. Indaba ngesayensi yekhompiyutha kanye nokwethulwa kokushicilelwa

Ukuze ababuyekezi abaqashwe abahleli bengqungquthela babe maningi amathuba okuba bagunyaze iphepha, kufanele libe nesibaluli esisodwa noma ngaphezulu. Isici esiyinhloko esandisa amathuba okuvunywa ubusha besayensi bombono ohlongozwayo. Ngokuvamile, ubusha buhlolwa ngokuphathelene nemibono ekhona kakade - futhi umsebenzi wokuhlola awenziwa umbuyekezi, kodwa ngumlobi we-athikili ngokwakhe. Ngokufanelekile, umbhali kufanele atshele ngokuningiliziwe esihlokweni mayelana nezindlela ezikhona futhi, uma kungenzeka, aziveze njengamacala akhethekile wendlela yakhe. Ngakho-ke, usosayensi ubonisa ukuthi izindlela ezamukelekayo azisebenzi ngaso sonke isikhathi, ukuthi wazenza ngokujwayelekile futhi wahlongoza ukwakheka okubanzi, okuguquguqukayo futhi ngakho-ke okusebenzayo kwethiyori. Uma into entsha ingenakuphikwa, ngakho-ke ababuyekezi abahloli i-athikili ngendlela engafanele - ngokwesibonelo, bangase bashaye indiva isiNgisi esimpofu.

Ukuze uqinise ubusha, kuyasiza ukufaka ukuqhathanisa nezindlela ezikhona kudathasethi eyodwa noma ngaphezulu. Ngamunye wabo kufanele avuleke futhi amukelwe endaweni yezemfundo. Isibonelo, kukhona inqolobane yesithombe se-ImageNet kanye nesizindalwazi sezikhungo ezifana ne-Modified National Institute of Standards and Technology (MNIST) kanye ne-CIFAR (i-Canadian Institute For Advanced Research). Ubunzima ukuthi isethi yedatha β€œyezemfundo” enjalo ivamise ukuhluka ngesakhiwo sokuqukethwe kudatha yangempela imboni ebhekene nayo. Idatha ehlukile isho imiphumela ehlukene yendlela ehlongozwayo. Ososayensi abasebenza kancane embonini bazama ukukunakisisa lokhu futhi kwesinye isikhathi bafake izinkulumo zokuziqhathulula njengokuthi β€œkudatha yethu umphumela uba ngokuthi nalokhu, kodwa kudathasethi yomphakathi – okuthi nalokhu.”

Kwenzeka ukuthi indlela ehlongozwayo "ilungiselelwe" ngokuphelele ku-database evulekile futhi ayisebenzi kudatha yangempela. Ungakwazi ukulwa nale nkinga evamile ngokuvula amasethi edatha amasha, amele kakhudlwana, kodwa ngokuvamile sikhuluma ngokuqukethwe okuyimfihlo izinkampani ezingenalo ilungelo lokukuvula. Kwezinye izimo, benza (ngezinye izikhathi okuyinkimbinkimbi futhi okubuhlungu) ukwenza idatha ingaziwa - basusa noma yiziphi izingcezu ezikhomba umuntu othile. Isibonelo, ubuso nezinombolo ezithombeni ziyasulwa noma zenziwa zingafundeki. Ngaphezu kwalokho, ukuze idathasethi ingatholakali kuwo wonke umuntu kuphela, kodwa ibe yindinganiso phakathi kososayensi lapho kulula ukuqhathanisa imibono, akudingekile ukuyishicilela kuphela, kodwa futhi nokubhala isihloko esicashuniwe esihlukile mayelana. yona kanye nezinzuzo zayo.

Kubi kakhulu uma kungekho dathasethi evuliwe esihlokweni esifundwayo. Khona-ke umbuyekezi angamukela kuphela imiphumela eyethulwa ngumbhali ngokukholwa. Ngokwethiyori, umbhali angakwazi ngisho nokuzicabangela ngokweqile futhi ahlale engabonakali, kodwa endaweni yezemfundo akunakwenzeka, ngoba kuphambene nesifiso sososayensi abaningi sokuthuthukisa isayensi.

Ezindaweni eziningi ze-ML, okuhlanganisa ukubona kwekhompuyutha, kuvamile futhi ukunamathisela izixhumanisi kukhodi (imvamisa ku-GitHub) ngezindatshana. Izindatshana ngokwazo ziqukethe ikhodi encane kakhulu noma ziyi-pseudocode. Futhi lapha, futhi, kuphakama ubunzima uma isihloko sibhalwa umcwaningi wenkampani, hhayi eyunivesithi. Ngokuzenzakalela, ikhodi ebhalwe enkampanini noma eqalayo ibhalwe ukuthi NDA. Abacwaningi nozakwabo kufanele basebenze kanzima ukuze bahlukanise ikhodi ehlobene nombono ochazwayo kumakhosombe angaphakathi futhi avaliwe ngokuqinisekile.

Ithuba lokushicilela nalo lincike ekuhlobaneni kwesihloko esikhethiwe. Ukufaneleka kunqunywa kakhulu imikhiqizo namasevisi: uma inkampani noma isiqalisi sinentshisekelo yokwakha isevisi entsha noma ukuthuthukisa ekhona ngokususelwe embonweni ophuma kusihloko, lokho kuwukuhlanganisa.

Umklomelo oqanjwe ngo-Ilya Segalovich. Indaba ngesayensi yekhompiyutha kanye nokwethulwa kokushicilelwa

Njengoba sekushiwo, amaphepha esayensi yekhompyutha awavamile ukubhalwa ewodwa. Kodwa njengomthetho, omunye wababhali uchitha isikhathi esiningi nomzamo kunabanye. Igalelo lakhe entweni entsha yesayensi likhulu kakhulu. Ohlwini lwababhali, umuntu onjalo uboniswa kuqala - futhi esikhathini esizayo, lapho ebhekisela esihlokweni, bangakhuluma ngaye kuphela (ngokwesibonelo, "Ivanov et al" - "Ivanov nabanye" elihunyushwe lisuka kwisiLatini). Kodwa-ke, iminikelo yabanye nayo ibaluleke kakhulu - ngaphandle kwalokho akunakwenzeka ukuba sohlwini lwababhali.

Inqubo yokubuyekeza

Amaphepha ngokuvamile ayayeka ukwamukelwa ezinyangeni ezimbalwa ngaphambi kwengqungquthela. Ngemva kokuthunyelwa kwe-athikili, ababuyekezi banamaviki angu-3–5 okufanele bayifunde, bayihlole, futhi baphawule ngayo. Lokhu kwenzeka ngokulandela uhlelo olulodwa oluyimpumputhe, lapho abalobi bengaboni amagama ababuyekezi, noma abangaboni kabili, lapho ababuyekezi ngokwabo bengaboni amagama abalobi. Inketho yesibili ibhekwa njengokungakhethi: amaphepha amaningana esayensi abonise ukuthi ukuthandwa kombhali kuthonya isinqumo sombuyekezi. Isibonelo, angase acabange ukuthi usosayensi onenani elikhulu lama-athikili asevele ashicilelwe uyi-priori efaneleka ukulinganisa okuphezulu.

Ngaphezu kwalokho, ngisho nasendabeni yezimpumputhe eziphindwe kabili, umbuyekezi cishe uzoqagela umbhali uma besebenza ensimini efanayo. Ngaphezu kwalokho, ngesikhathi sokubuyekezwa, i-athikili ingase isivele ishicilelwe ku-database ye-arXiv, inqolobane enkulu kunazo zonke yamaphepha esayensi. Abahleli benkomfa abakwenqabeli lokhu, kodwa batusa ukusebenzisa isihloko esihlukile kanye ne-abstract ehlukile ekushicilelweni kwe-arXiv. Kodwa uma isihloko sifakwe lapho, ngeke namanje kube nzima ukuyithola.

Kuhlala kukhona ababuyekezi abambalwa abahlola isihloko. Omunye wabo unikezwe indima yombuyekezi we-meta, okumele abuyekeze kuphela izinqumo zozakwabo futhi enze isinqumo sokugcina. Uma ababuyekezi bengavumelani ne-athikili, umbuyekezi we-meta angaphinde ayifunde ukuze aphelele.

Ngezinye izikhathi, ngemva kokubuyekeza isilinganiso namazwana, umbhali unethuba lokungena engxoxweni nombuyekezi; kukhona ngisho nethuba lokumkholisa ukuthi ashintshe isinqumo sakhe (kodwa-ke, uhlelo olunjalo alusebenzi kuzo zonke izingqungquthela, futhi akunakwenzeka ngisho nokuthonya kakhulu isinqumo). Engxoxweni, awukwazi ukubhekisela kweminye imisebenzi yesayensi, ngaphandle kwaleyo eseyibaluliwe esihlokweni. Ungakwazi kuphela "ukusiza" umbuyekezi ukuthi aqonde kangcono okuqukethwe yi-athikili.

Umklomelo oqanjwe ngo-Ilya Segalovich. Indaba ngesayensi yekhompiyutha kanye nokwethulwa kokushicilelwa

Izinkomfa namajenali

Izihloko zesayensi yekhompyutha zivame ukuhanjiswa ezingqungqutheleni kunamajenali esayensi. Lokhu kungenxa yokuthi ukushicilelwa kwamajenali kunezidingo okunzima kakhulu ukuhlangabezana nazo, futhi inqubo yokubuyekezwa kontanga ingathatha izinyanga noma ngisho neminyaka. Isayensi yamakhompiyutha iwumkhakha osheshayo, ngakho-ke ababhali ngokuvamile abazimisele ukulinda isikhathi eside kangako ukuze bashicilelwe. Nokho, isihloko esesamukelwe kakade engqungqutheleni singase sengezwe (ngokwesibonelo, ngokwethula imiphumela enemininingwane eyengeziwe) futhi sishicilelwe kujenali lapho imingcele yendawo ingaqinile kangako.

Imicimbi kule ngqungquthela

Ifomethi yokuba khona kwababhali bezindatshana ezigunyaziwe engqungqutheleni inqunywa ababuyekezi. Uma i-athikili inikezwe ukukhanya okuluhlaza, khona-ke uvame ukunikezwa isitendi sephosta. Iphosta iyisilayidi esimile esinesifinyezo se-athikili nemifanekiso. Amanye amagumbi enkomfa agcwaliswa ngemigqa emide yezitendi zamaphosta. Umbhali uchitha ingxenye ebalulekile yesikhathi sakhe eduze nephosta yakhe, ekhuluma nososayensi abanentshisekelo kulesi sihloko.

Umklomelo oqanjwe ngo-Ilya Segalovich. Indaba ngesayensi yekhompiyutha kanye nokwethulwa kokushicilelwa

Umklomelo oqanjwe ngo-Ilya Segalovich. Indaba ngesayensi yekhompiyutha kanye nokwethulwa kokushicilelwa

Inketho ehloniphekile yokubamba iqhaza inkulumo yombani. Uma ababuyekezi bebheka i-athikili njengefanele umbiko osheshayo, umbhali unikezwa cishe imizuzu emithathu ukuze akhulume nezithameli eziningi. Ngakolunye uhlangothi, inkulumo yombani iyithuba elihle lokukhuluma ngombono wakho hhayi kuphela kulabo ababe nesithakazelo kuphosta ngokusuka kwabo. Ngakolunye uhlangothi, abavakashi bephosta abamatasa balungiselela kakhulu futhi bagxile kakhulu esihlokweni sakho esithile kunomlaleli ojwayelekile ehholo. Ngakho-ke, embikweni osheshayo, kusadingeka ube nesikhathi sokuletha abantu abasesikhathini samanje.

Umklomelo oqanjwe ngo-Ilya Segalovich. Indaba ngesayensi yekhompiyutha kanye nokwethulwa kokushicilelwa

Ngokuvamile, ekupheleni kwenkulumo yabo ekhanyayo, ababhali basho inombolo yephosta ukuze abalaleli bayithole futhi bayiqonde kangcono i-athikili.

Umklomelo oqanjwe ngo-Ilya Segalovich. Indaba ngesayensi yekhompiyutha kanye nokwethulwa kokushicilelwa

Inketho yokugcina, ehlonishwa kakhulu iphosta kanye nesethulo esigcwele sombono, lapho singasekho isidingo sokujaha ukuxoxa indaba.

Umklomelo oqanjwe ngo-Ilya Segalovich. Indaba ngesayensi yekhompiyutha kanye nokwethulwa kokushicilelwa

Kodwa-ke, ososayensi - kuhlanganise nababhali bezindatshana ezigunyaziwe - beza engqungqutheleni elandelayo hhayi nje ukuzobukisa. Okokuqala, bavame ukuthola amaphosta ahlobene nensimu yabo ngezizathu ezisobala. Futhi okwesibili, kubalulekile ukuthi bandise uhlu lwabo loxhumana nabo ngenjongo yomsebenzi ohlangene wezemfundo esikhathini esizayo. Lokhu akukhona ukuzingela - noma, okungenani, isigaba saso sokuqala, okungenani esilandelwa ukushintshana okuzuzisayo kwemibono, intuthuko kanye nomsebenzi ohlanganyelwe esihlokweni esisodwa noma ngaphezulu.

Ngesikhathi esifanayo, ukuxhumana kwenethiwekhi okukhiqizayo engqungqutheleni ephezulu kunzima ngenxa yokuntuleka okuphelele kwesikhathi samahhala. Uma, ngemva kosuku lonke oluchithwe ezethulweni nasezingxoxweni ezingxoxweni, usosayensi uye wagcina amandla akhe futhi usevele wanqoba i-jet lag, khona-ke uya kwelinye lamaqembu amaningi. Zisingathwa yizinkampani - ngenxa yalokho, amaqembu avame ukuba nomlingiswa wokuzingela. Ngesikhathi esifanayo, izivakashi eziningi azisebenzisi nhlobo ukuthola umsebenzi omusha, kodwa, futhi, ukuze zithole inethiwekhi. Kusihlwa ayisekho imibiko namaphosta - kulula "ukubamba" uchwepheshe omthandayo.

Umklomelo oqanjwe ngo-Ilya Segalovich. Indaba ngesayensi yekhompiyutha kanye nokwethulwa kokushicilelwa

Kusukela kumbono kuya ekukhiqizeni

Isayensi yamakhompiyutha ingenye yezimboni ezimbalwa lapho izithakazelo zezinkampani kanye neziqalo zixhunyaniswe ngokuqinile nendawo yezemfundo. I-NIPS, ICML nezinye izingqungquthela ezifanayo ziheha abantu abaningi abavela embonini, hhayi emanyuvesi kuphela. Lokhu kuvamile emkhakheni wesayensi yamakhompiyutha, kodwa okuphambene nalokho kwamanye amasayensi amaningi.

Ngakolunye uhlangothi, akuyona yonke imibono ethulwa ezihlokweni ngokushesha eya ekudaleni noma ekuthuthukiseni izinsizakalo. Ngisho nangaphakathi kwenkampani eyodwa, umcwaningi angahlongoza ozakwabo enhlanganweni umbono ophunyeleliswa yizindinganiso zesayensi futhi athole ukwenqaba ukuwusebenzisa ngenxa yezizathu eziningi. Enye yazo isivele ishiwo lapha - lona umehluko phakathi kwesethi yedatha "yezemfundo" okwabhalwa kuyo isihloko kanye nesethi yedatha yangempela. Ngaphezu kwalokho, ukusetshenziswa kombono kungase kubambezeleke, kudinge inani elikhulu lezisetshenziswa, noma kuthuthukiswe inkomba eyodwa kuphela ngezindleko zokuwohloka kwamanye amamethrikhi.

Umklomelo oqanjwe ngo-Ilya Segalovich. Indaba ngesayensi yekhompiyutha kanye nokwethulwa kokushicilelwa

Isimo sigcinwa yiqiniso lokuthi abathuthukisi abaningi ngokwabo bangabacwaningi abancane. Bahambela izingqungquthela, bakhulume ulimi olufanayo nezifundiswa, baphakamise imibono, ngezinye izikhathi babambe iqhaza ekwakhiweni kwama-athikili (isibonelo, ukubhala ikhodi), noma benze njengababhali ngokwabo. Uma umthuthukisi egxilile enqubweni yezemfundo, elandela okwenzeka emnyangweni wocwaningo, ngegama - uma ebonisa ukunyakaza okuphikisayo kososayensi, khona-ke umjikelezo wokuguqula imibono yesayensi ibe amandla amasha enkonzo uyafinyezwa.

Sifisela bonke abacwaningi abasebasha inhlanhla nempumelelo enkulu emsebenzini wabo. Uma lokhu okuthunyelwe kungakutshelanga lutho olusha, kungenzeka ukuthi usushicilele kakade engqungqutheleni ephezulu. Bhalisa i i-premium ngokwakho futhi uqoke abaphathi besayensi.

Source: www.habr.com

Engeza amazwana