Okuqukethwe[Fihla][Bonisa]
Abaningi babona ngeso lengqondo amarobhothi afana nalawo asemafilimini ezinganekwane alingisa noma adlula ngisho nokuhlakanipha komuntu lapho ezwa amagama athi ubuhlakani bokwenziwa, ukufunda okujulile, nokufunda ngomshini.
Abanye bacabanga ukuthi le mishini imane ithathela ukwaziswa futhi ifunde kuyo ngokwayo. Awu… Kuyakhohlisa kancane. Ukulebula idatha kuyindlela esetshenziswa ukuqeqesha amakhompuyutha ukuthi “ahlakaniphe,” njengoba enamandla alinganiselwe ngaphandle kokufundiswa umuntu.
Ukuze siqeqeshe ikhompuyutha ukuthi yenze “ngokukhalipha,” sifaka idatha ngezindlela ezihlukahlukene futhi siyifundisa amasu ahlukahlukene ngosizo lokulebula idatha.
Amasethi edatha kufanele achazwe noma alebulwe ngezimvume eziningi zolwazi olufanayo njengengxenye yesayensi eyisisekelo yokulebula idatha.
Umzamo nokuzinikela okubekwe emkhiqizweni wokugcina kuyancomeka, ngisho noma kumangalisa futhi kwenza ukuphila kwethu kwansuku zonke kube lula.
Funda mayelana nokulebula idatha kulesi sihloko ukuze ufunde ukuthi iyini, ukuthi isebenza kanjani, izinhlobo ezihlukene zokulebula idatha, izithiyo, nokunye okuningi.
Ngakho-ke, yini i-Data Labeling?
In ukufunda imishini, izinga kanye nemvelo yedatha yokufaka inquma izinga kanye nemvelo yokuphumayo. Ukunemba kwemodeli yakho ye-AI kuthuthukiswa izinga ledatha esetshenziselwa ukuyiqeqesha.
Ngamanye amagama, ukulebula idatha kuyisenzo sokulebula noma sokuchasisa amasethi edatha angahlelekile noma angahlelekile ukuze kufundiswe ikhompuyutha ukuhlonza umehluko namaphethini phakathi kwawo.
Umfanekiso uzokusiza ukuqonde lokhu. Kudingeka ukumaka zonke izibani ezibomvu ezithombeni ezihlukahlukene ukuze ikhompuyutha ifunde ukuthi ukukhanya okubomvu kuwuphawu lokuma.
Ngokwesisekelo salokhu, i-AI ithuthukisa i-algorithm, kuzo zonke izimo, ezohumusha ukukhanya okubomvu njengenkomba yokumisa. Omunye umfanekiso yikhono lokuhlukanisa amasethi edatha ahlukene ngaphansi kwezihloko ze-jazz, i-pop, i-rock, yakudala, nokuningi ukuze kuhlukaniswe izinhlobo zomculo ezihlukene.
Ukukubeka kalula, ukulebula idatha ekufundeni komshini kubhekisela kunqubo yokuthola idatha engenamalebula (njengezithombe, amafayela ombhalo, amavidiyo, njll.) kanye nokwengeza ilebula eyodwa noma amaningi ahlobene ukuze unikeze umongo ukuze imodeli yokufunda yomshini ifunde kukho. yona.
Amalebula angasho, isibonelo, uma i-x-ray ikhombisa isimila noma cha, yimaphi amagama ashiwo kusiqeshana somsindo, noma isithombe senyoni noma semoto.
Ukulebula idatha kubalulekile ezimweni eziningi zokusetshenziswa, okuhlanganisa ukuqaphela inkulumo, umbono wekhompyutha, kanye nokucutshungulwa kolimi lwemvelo.
Ukulebula idatha: Kungani Kubalulekile?
Okokuqala, inguquko yesine yezimboni igxile ekhonweni lemishini yokuqeqesha. Ngenxa yalokho, ibalwa phakathi kwentuthuko yesoftware ebaluleke kakhulu yamanje.
Isistimu yakho yokufunda yomshini kufanele idalwe, okubandakanya ukulebula idatha. Isungula amakhono esistimu. Alukho uhlelo uma idatha ingekho ilebula.
Amathuba okulebula idatha anqunyelwe kuphela ubuhlakani bakho. Noma yisiphi isenzo ongasenza kumephu ohlelweni sizophinda ngolwazi olusha.
Okusho ukuthi uhlobo, ubuningi, kanye nokuhlukahluka kwedatha ongayifundisa isistimu kuzonquma ubuhlakani nekhono layo.
Okwesibili ukuthi umsebenzi wokulebula idatha uza ngaphambi komsebenzi wesayensi yedatha. Ngakho-ke, ukulebula kwedatha kuyadingeka kwisayensi yedatha. Ukwehluleka namaphutha ekulebulani idatha kuthinta isayensi yedatha. Ngaphandle kwalokho, ukusebenzisa igama elithi cruder cliché, “faka udoti, khipha udoti.”
Okwesithathu, I-Art of Data Labeling isho ushintsho endleleni abantu ababhekana ngayo nokuthuthukiswa kwezinhlelo ze-AI. Ngesikhathi esifanayo sicwenga ukwakheka kokulebula kwedatha ukuze sihlangabezane kangcono nemigomo yethu kunokuzama nje ukuthuthukisa amasu ezibalo.
I-automation yesimanje isuselwe kulokhu, futhi iyisikhungo se-AI Transformation eqhubekayo njengamanje. Manje kunanini ngaphambili, umsebenzi wolwazi wenziwa ngemishini.
Isebenza kanjani ilebula yedatha?
Uhlelo olulandelayo lokulandelana kwezikhathi luyalandelwa phakathi nenqubo yokulebula idatha.
Ukuqoqwa kwedatha
Idatha iyisisekelo sanoma yimuphi umzamo wokufunda umshini. Isigaba sokuqala sokulebula idatha sihlanganisa ukuqoqa inani elifanele ledatha eluhlaza ngamafomu ahlukene.
Ukuqoqwa kwedatha kungase kuthathe uhlobo olulodwa kwezimbili: kungenzeka ukuthi kuvela emithonjeni yangaphakathi ibhizinisi ekade liyisebenzisa, noma kuvela emithonjeni yangaphandle efinyeleleka esidlangalaleni.
Njengoba ikwifomu eluhlaza, le datha idinga ukuhlanzwa futhi icutshungulwe ngaphambi kokuthi kwenziwe amalebula edathasethi. Imodeli ibe isiqeqeshwa kusetshenziswa le datha ehlanziwe yaphinde yacutshungulwa. Okutholakele kuzonemba kakhulu uma isethi yedatha inkulu futhi ihluke kakhulu.
Isichasiselo sedatha
Ngemva kokuhlanzwa kwedatha, ochwepheshe besizinda bahlola idatha futhi basebenzise amalebula kusetshenziswa izindlela ezimbalwa zokulebula idatha. Imodeli inomongo onengqondo ongasetshenziswa njengeqiniso eliyisisekelo.
Lezi yizinto eziguquguqukayo ofuna ukuthi imodeli ibikezele, njengezithombe.
Ukuqinisekiswa kwekhwalithi
Ikhwalithi yedatha, okufanele ithembeke, inembile, futhi ingaguquguquki, ibalulekile empumelelweni yokuqeqeshwa kwemodeli ye-ML. Ukuhlola okujwayelekile kwe-QA kufanele kwenziwe ukuze kuqinisekiswe lokhu kufakwa kwamalebula kwedatha okunembile nokulungile.
Kungenzeka ukuhlola ukunemba kwalezi zichasiselo ngokusebenzisa amasu e-QA afana ne-Consensus kanye nokuhlolwa kwe-alpha kwe-Cronbach. Ukunemba kwemiphumela kuthuthukiswa kakhulu ngokuhlolwa kwe-QA okujwayelekile.
Amamodeli okuqeqesha nokuhlola
Izinqubo ezishiwo ngenhla zinengqondo kuphela uma idatha ihlolelwa ukulunga. Uhlelo luzovivinywa ngokufaka idathasethi engahlelekile ukuze kubhekwe ukuthi iyakhiqiza yini imiphumela oyifunayo.
Amasu okulebula idatha
Ukulebula idatha kuyinqubo ekhandlayo edinga ukunakwa ngemininingwane. Indlela esetshenziselwa ukuchaza idatha izohluka kuye ngesitatimende senkinga, ukuthi ingakanani idatha okufanele imakwe, ukuthi idatha iyinkimbinkimbi kangakanani, kanye nesitayela.
Ake sidlule kwezinye zezinketho ibhizinisi lakho elinalo, kuye ngezisetshenziswa elinazo nesikhathi elinakho.
Ukulebula idatha endlini
Njengoba igama lisho, ukulebula kwedatha yangaphakathi kwenziwa ngochwepheshe ngaphakathi kwenkampani. Uma unesikhathi esanele, abasebenzi, nezinsiza zezezimali, inketho engcono kakhulu njengoba iqinisekisa ukulebula okunembe kakhulu. Nokho, ihamba kancane.
Outsourcing
Enye inketho yokwenza izinto ukuqasha ama-freelancers emisebenzini yokulebula idatha engatholwa ezindaweni ezihlukene zokufuna umsebenzi nezimakethe ezizimele njenge-Upwork.
Ukukhishwa kwemisebenzi kuyindlela esheshayo yokuthola izinsiza zokulebula idatha, nokho, ikhwalithi ingase iphazamiseke, ngokufana nendlela yangaphambili.
Ukubambisana
Ungangena njengomceli futhi usabalalise imisebenzi ehlukahlukene yokulebula kosonkontileka abatholakalayo kumapulatifomu akhethekile e-crowdsourcing afana I-Amazon Mechanical Turk (MTurk).
Indlela, nakuba ishesha ngandlela thile futhi ingabizi, ayikwazi ukuhlinzeka ngedatha yezichasiselo zekhwalithi enhle.
Ukulebula idatha ngokuzenzakalela.
Inqubo ingase isizwa isofthiwe ngaphezu kokuthi yenziwa mathupha. Ngokusebenzisa indlela yokufunda esebenzayo, omaka bangatholwa ngokuzenzakalelayo futhi bangezwe kudathasethi yokuqeqeshwa.
Empeleni, ochwepheshe babantu bakha imodeli yelebula ye-AI Auto ukuze imake idatha engafakiwe ilebula, eluhlaza. Bese benquma ukuthi imodeli isebenzise ngokufanelekile ukulebula. Abantu balungisa amaphutha ngemuva kokwehluleka futhi baqeqeshe kabusha i-algorithm.
Ukuthuthukiswa kwedatha yokwenziwa.
Esikhundleni sedatha yomhlaba wangempela, idatha yokwenziwa iyidathasethi enelebula eyakhiwe ngokwenziwa. Ikhiqizwa ngama-algorithms noma izilinganiso zekhompyutha futhi ivame ukusetshenziselwa amamodeli wokufunda womshini wokuqeqesha.
Idatha yokwenziwa yimpendulo enhle kakhulu ezindabeni zokushoda kwedatha nokuhlukahluka kumongo wezinqubo zokulebula. Ukudalwa kwe idatha yokwenziwa kusukela ekuqaleni inikeza isixazululo.
Ukudalwa kwezilungiselelo ze-3D ngezinto nokuzungezile imodeli kufanele kukwazi ukubonwa onjiniyela bedathasethi. Idatha yokwenziwa eningi edingekayo kuphrojekthi inganikezwa.
Izinselelo Zokulebula Idatha
Kudinga isikhathi nomzamo owengeziwe
Ngaphezu kokuba inselele ukuthola idatha enkulu (ikakhulukazi ezimbonini ezikhetheke kakhulu njengokunakekelwa kwezempilo), ukulebula ucezu ngalunye lwedatha ngesandla kokubili kuwumsebenzi onzima futhi onzima, odinga usizo lwamalebula angabantu.
Cishe u-80% wesikhathi esichithwa kuphrojekthi kuwo wonke umjikelezo wokuthuthukiswa kwe-ML sichithwa ekulungiseni idatha, okuhlanganisa ukulebula.
Amathuba okungahambisani
Isikhathi esiningi, ukulebula okuphambene, okwenzeka lapho abantu abaningi balebula amasethi afanayo edatha, kubangela ukunemba okukhulu.
Kodwa-ke, ngenxa yokuthi abantu ngezinye izikhathi banamazinga ahlukahlukene okusebenza, izindinganiso zokulebula namalebula ngokwawo angahle angahambisani, okungenye inkinga, Kungenzeka ukuthi izichasiselo ezimbili noma ngaphezulu zingavumelani kwabanye omaka.
Ngokwesibonelo, uchwepheshe othile angase alinganisele isibuyekezo sehhotela njengesivuzayo kanti omunye angakubheka njengokubhuqa futhi akunikeze isilinganiso esiphansi.
Ulwazi lwesizinda
Uzozwa isidingo sokuqasha amalebula anolwazi olukhethekile lwemboni kweminye imikhakha.
Izichasiselo ezingenalo ulwazi lwesizinda oludingekayo, isibonelo, zizoba nesikhathi esinzima kakhulu sokumaka izinto ngendlela efanele ngenkathi kwakhiwa uhlelo lokusebenza lwe-ML lomkhakha wezempilo.
Ukuthambekela emaphutheni
Ukulebula okwenziwa ngesandla kungaphansi kwamaphutha abantu, kungakhathaliseki ukuthi abalebula bakho banolwazi futhi bacophelele kangakanani. Ngenxa yokuthi izichasiselo zivame ukusebenza namasethi edatha eluhlaza, lokhu akunakugwemeka.
Cabanga ngomuntu echaza izithombe ezingu-100,000 ngezinto ezifika kweziyi-10 ezahlukene.
Izinhlobo ezijwayelekile zokulebula idatha
Umbono wekhompyutha
Ukuze uthuthukise idathasethi yakho yokuqeqeshwa, kufanele uqale ulebule izithombe, amaphikseli, noma izindawo ezibalulekile, noma umise umngcele ovala ngokuphelele isithombe sedijithali, okwaziwa ngebhokisi elihlanganisayo, lapho wakha uhlelo lokubona lwekhompyutha.
Izithombe zingahlukaniswa ngezindlela ezihlukahlukene, okuhlanganisa nokuqukethwe (lokho okusesithombeni ngokwaso) kanye nekhwalithi (okufana nomkhiqizo uma uqhathaniswa nezithombe zendlela yokuphila).
Izithombe zingabuye zihlukaniswe ngamasegimenti ezingeni lephikseli. Imodeli yombono wekhompuyutha ethuthukiswe kusetshenziswa le datha yokuqeqeshwa ingase kamuva isetshenziselwe ukuhlukanisa izithombe ngokuzenzakalelayo, ukucacisa indawo yezinto, ukugqamisa izindawo ezibalulekile esithombeni, kanye nesegimenti yezithombe.
Ukucutshungulwa kolimi ngokwemvelo
Ngaphambi kokukhiqiza idathasethi yakho yokuqeqeshwa kokucubungula ulimi lwemvelo, kufanele ukhethe mathupha izingcezu zombhalo ezifanele noma uhlukanise okokusebenza ngamalebula acacisiwe.
Isibonelo, ungase ufune ukubona amaphethini enkulumo, uhlukanise amagama afanele njengezindawo nabantu, futhi uhlonze umbhalo ezithombeni, kuma-PDF, noma kwezinye imidiya. Ungase futhi ufune ukunquma imizwa noma inhloso yokufiphala kombhalo.
Dala amabhokisi ahlanganisayo azungeze umbhalo kudathasethi yakho yokuqeqeshwa ukuze ufeze lokhu, bese ukuloba mathupha.
Ukubonwa kwezinhlamvu ezibonakalayo, ukuhlonza igama lebhizinisi, nokuhlaziywa kwemizwelo konke kwenziwa kusetshenziswa amamodeli okucubungula ulimi lwendalo.
Ukucutshungulwa komsindo
Ukucutshungulwa komsindo kuguqula zonke izinhlobo zemisindo zibe ifomethi ehlelekile ukuze isetshenziswe ekufundeni komshini, okuhlanganisa inkulumo, imisindo yezilwane (amagxolo, amakhwela, noma ukutshiyoza), nomsindo wokwakha (ingilazi ephukile, ukuskena, noma izinhlabamkhosi).
Ngokuvamile, ngaphambi kokuthi ukwazi ukuphatha umsindo, kufanele uwuguqulele ube umbhalo. Ngemva kwalokho, ngokuhlukanisa ngokwezigaba nokwengeza omaka emsindweni, ungafunda ulwazi oluningiliziwe ngakho. Eyakho isethi yedatha yokuqeqeshwa yilo msindo ohlukanisiwe.
Isiphetho
Sengiphetha, ukuhlonza idatha yakho kuyingxenye ebalulekile yokuqeqesha noma iyiphi imodeli ye-AI. Inhlangano ematasatasa, nokho, ayinakukwazi ukuchitha isikhathi iyenza mathupha ngoba idla isikhathi futhi idla amandla.
Ukwengeza, kuyinqubo ethambekele ekungangeni kahle futhi ayithembisi ukunemba okukhulu. Akumele kube nzima kangako, okuyizindaba ezinhle kakhulu.
Ubuchwepheshe banamuhla bokulebula idatha buvumela ukusebenzisana phakathi kwabantu nemishini ukuze kunikeze idatha enembile newusizo yezinhlelo zokusebenza ezihlukahlukene zokufunda umshini.
shiya impendulo