Kwenzekeni
Abacwaningi basesikoleni i-Wharton bashicilele ucwaningo ku-Nature Computational Science olubonisa ukuthi ukuzibeka kwababhali ngokwamaphepha abo e-AI kuwukubikezela okunamandla kokucashunwa kwezemfundo kwesikhathi esizayo kunamaphuzu ajwayelekile okubuyekezwa kontanga. Ucwaningo luhlaziye idatha evela ku-2023 International Conference on Machine Learning (ICML) futhi lwathola ukuthi i-ICML selokhu yahlanganisa le ndlela yokwehluka okuzicabangela yona enqubweni yayo yokubuyekeza yango-2026 ukuze ihlabe umkhosi amaphepha ukuze ihlolwe eduze nezihlalo zendawo.
Abacwaningi beSikole SaseWharton, okuhlanganisa nosolwazi u-Weijie Su no-Bingxin Zhao kanye nomfundi ofundela ubudokotela u-Buxin Su, bashicilele ucwaningo ku-Nature Computational Science ngo-Agasti 24. Lolu cwaningo luphenye ukuthi ingabe izinga lababhali bamaphepha abo lingasebenza yini njengenkomba enokwethenjelwa yomthelela wesayensi emkhakheni wobuhlakani bokwenziwa.
Ithimba labacwaningi lihlaziye idatha evela ku-2023 International Conference on Machine Learning (ICML). Baqoqe izikhundla zabo kubabhali abangu-1,342 abahlanganisa izethulo eziyi-2,592. Ababhali bacelwe ukuthi bahlele amaphepha abo ngekhwalithi yesayensi ebonwayo ngaphambi kokukhishwa kokubuyekezwa kontanga. Ukuhlaziywa kokugcina kufaka ababhali abangama-797 kanye namaphepha ahlukile ayi-1,527 ngemuva kokumataniswa nedatha ecashuniwe.
Ucwaningo luthole ukuthi amaphepha abekwe phezulu kakhulu ngababhali bawo athole isilinganiso sokucashunwa okuphindwe kabili ezinyangeni eziyi-16 ezilandelayo njengalawo akleliswe phansi kakhulu. Le phethini ibambele kokubili amaphepha amukelwayo nanqatshiwe. Ngaphezu kwalokho, amazinga ombhali abikezele ukubala okucashuniwe okuzayo ngokunembe kakhulu kunezikolo zokubuyekezwa kontanga. Emaphepheni angama-22 kusampula athole izingcaphuno ezingaphezu kwe-150, ayi-17 abekwe kuqala okungenani ngumbhali oyedwa.
Ngokusekelwe kulokhu okutholakele, abacwaningi bahlongoze uhlelo lapho umehluko phakathi kwamazinga ababhali kanye nezikolo zababuyekezi zisetshenziselwa ukumaka amaphepha ukuze ahlolwe eduze. I-ICML ifake le ndlela kunqubo yayo yokubuyekeza yango-2026. Esivivinyweni esingahleliwe, izihlalo zendawo ezikwazi ukubona izigaba zokungafani zibhale cishe umbhalo wamazwana ongu-71% ngaphezulu ephepheni ngalinye futhi baxhumana kakhulu nababuyekezi uma kuqhathaniswa nalabo abangakwazi ukubona izigaba.
Imininingwane yomthombo: thedp.com ↗
Kungani kubalulekile
Inkambu yocwaningo ye-AI ikhula ngokushesha kunechibi lababuyekezi ontanga abanolwazi, idala ibhodlela lokuhlonza umsebenzi onomthelela omkhulu. Lolu cwaningo luhlinzeka ngendlela esebenzayo, eqhutshwa idatha ukuze kwabiwe izinsiza zokubuyekeza okunomkhawulo ngempumelelo kakhudlwana. Ngokusebenzisa izikhundla ukuze uhlabe umkhosi ngokungafani nezikolo zababuyekezi, izinkomfa zingaqondisa ukunaka emaphepheni okungenzeka anganakwa noma angahlulelwanga kahle. Le ndlela ibhekana nokungasebenzi kahle kwesakhiwo ekushicileleni kwezemfundo ngaphandle kokuthatha indawo yokwahlulela komuntu, okungase kube ngcono ikhwalithi iyonke kanye nokuhambisana kocwaningo lwe-AI olwamukelwe.
Ukukhula okusheshayo kocwaningo lwe-AI kuye kwadlula ukutholakala kwababuyekezi ontanga abanolwazi, okwenza kube nzima ukuhlonza umsebenzi onomthelela omkhulu ngokusebenzisa izindlela zendabuko kuphela. Lolu cwaningo lunikeza isixazululo esinokwethenjelwa ngokusebenzisa ukuhlola kwababhali ukuze kubekwe phambili amaphepha adinga ukucutshungulwa okwengeziwe.
Indlela ayithathi indawo yokubuyekezwa kontanga kodwa iyakwengeza ngokugqamisa ukungezwani okungase kubonise ukuthi iphepha lenziwa inani eliphansi noma leqiwa ababuyekezi. Lokhu kuvumela izihlalo zendawo ukuthi zihlukanise isikhathi sazo esilinganiselwe ngokuphumelelayo, ngokugxila ezimweni lapho inqubo yokubuyekeza ingase ishode khona ama-nuances abalulekile.
Ucwaningo luvuma ukuthi ukubalwa kwengcaphuno kungummeleli ongaphelele wekhwalithi, njengoba kungathonywa ukuduma kwesihloko, isikhathi, nokubonakala kombhali. Kodwa-ke, ukuhlobana okuqinile phakathi kwezikhundla nokucashuniwe kuphakamisa ukuthi ababhali banomuzwa ozwakalayo womthelela ongaba khona womsebenzi wabo, ongasetshenziselwa ukuthuthukisa ukusebenza kahle kwenqubo yokubuyekeza.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
Which component of an AI application is the machine-learning model itself?
Ongakubuka ngokulandelayo
Qaphela ukuthi ingabe ezinye izingqungquthela ezinkulu ze-AI nezingqungquthela zokufunda ngomshini zisebenzisa izindlela ezifanayo zokuziqhathanisa ezilinganisweni ezinqubweni zazo zokubuyekeza. Buka izifundo zokulandelela ezihlola ukuthi ingabe le ndlela iyalithuthukisa ngempela ikhwalithi yamaphepha amukelwayo noma uma imane yandisa umsebenzi wombuyekezi. Ukwengeza, bheka ukuthi ngabe umphakathi wezemfundo wenza izimiso ezijwayelekile zokuziphakamisa ukuze uqinisekise ukungaguquguquki futhi uvimbele ukukhohlisa ezindaweni ezahlukene.
Akukacaci ukuthi le ndlela izosetshenziswa yini ezinye izingqungquthela ezinkulu ze-AI ezifana ne-NeurIPS noma i-ICLR. Impumelelo yokusetshenziswa kwe-ICML ingase isebenze njengobufakazi bomqondo womphakathi wezemfundo obanzi.
Ucwaningo lwesikhathi esizayo luzodinga ukunquma ukuthi ingabe ukucutshungulwa okwengeziwe kwamaphepha ahlatshwe umkhosi kuholela ezinqumweni ezingcono zokwamukelwa noma uma kumane kwengeze umthwalo ezihlalweni zendawo nababuyekezi. Ucwaningo lwamanje lulinganise ukuhlanganyela enqubweni yokubuyekeza kodwa aluzange lusungule ukuthuthukiswa kwezinga lezinqumo zokugcina.
Kukhona ingozi engaba khona yokukhohlisa, nakuba abacwaningi baklame isistimu ukuze inciphise izisusa ezinjalo ngokufihla isiqondiso sokungafani ezihlalweni zendawo. Ukuqapha ukuthi ababhali nababuyekezi bazijwayeza kanjani le nqubo entsha kuzobaluleka eminyakeni ezayo.