I-GFlowNets, ethuthukiswe ngaphansi kokuqondisa kuka-Yoshua Bengio, imelela i-algorithm entsha yokuqeqesha emkhakheni wobuhlakani bokwenziwa (AI), ehlukile kuma-neural network architecture afana ne-Transformers noma i-ResNets. U-Edward Hu, usosayensi wocwaningo oke wasebenza kwa-Microsoft kanye ne-OpenAI, uchaza ukubaluleka kwe-GFlowNets, egqamisa indima yabo ekuxazululeni izinkinga ezivamile ze-AI ezifana nokugcwalisa ngokweqile kanye nemikhawulo yokuhlelwa kwe-hyperparameter.
I-GFlowNets idizayinelwe ukukhiqiza imiphumela ehlukahlukene, enomvuzo kunokuba nje ikhulise umsebenzi womklomelo, onenzuzo ezinhlelweni zokusebenza ezifana nokutholwa kwezidakamizwa, lapho kungcono khona ukuba nezinhlobonhlobo zabantu abathembisayo . I-GFlowNets ingaphinda isetshenziswe kumamodeli asekelwe kumandla, lapho isiza khona ukusampula kokusabalalisa ngokulingana nemodeli.
Uma ucwile kuzwelonke ye-AI, ungase ube nelukuluku lokuthi yini eyenza i-GFlowNets ihluke. Ngokungafani nezindlela ezivamile ezigxile ekwandiseni umsebenzi womklomelo owodwa, i-GFlowNets iphumelela kakhulu ekukhiqizeni imiphumela emihle eyahlukahlukene. Lesi sici sibaluleke kakhulu emisebenzini eyinkimbinkimbi efana nokutholwa kwezidakamizwa, lapho ukuthola izinhlanganisela eziningi ezisebenzayo kubalulekile.
Usosayensi ohamba phambili wocwaningo u-Edward Hu ugqamisa umthelela obalulekile we-GFlowNets. Asebenza njengamanethiwekhi e-neural sampler, okusho ukuthi ayakwazi ukukhiqiza imiphumela ehambisana nomsebenzi onikeziwe womvuzo. Lokhu kuyimpumelelo enkulu, eshintsha indlela amasistimu obuhlakani bokwenziwa afunda ngayo endaweni yawo futhi enze izinqumo.
Yini i-GFlowNets?
Enye indlela i-GFlowNets ethuthukisa ngayo i-AI iwukuthuthukisa amamodeli asekelwe kumandla. Benza lokhu ngokunika amandla ukusampula kokusabalalisa okuhambisana nemodeli, okusiza ukucolisa ikhono lemodeli lokumela amaphethini edatha ayinkimbinkimbi. Lokhu kubaluleke kakhulu enguqulweni ye-GFlowNets eyaziwa ngokuthi i-GFlowNet-EM, ebhekana ne-algorithm yokwandisa okulindelekile, inselele eyisisekelo emkhakheni wokufunda komshini.
Uma kuziwa kumamodeli amakhulu olimi, i-GFlowNets ingenza ukusetshenziswa kwedatha kusebenze kahle kakhulu. Banekhono lokusampula amaketanga okucabanga, okunciphisa inani ledatha elidingekayo ukuqeqesha amasistimu obuhlakani bokwenziwa. Umphumela uba izinqubo zokuqeqesha ezisebenza kahle nezingabizi kakhulu.
Ama-GFlowNets ayaziwa nangokukwazi kwawo ukuhlanganisa ukukala kwamanethiwekhi e-neural nezinkinga zokufunda zomshini zakudala. Lokhu kuhlanganiswa kuthuthukisa ukugcwaliseka kwedatha nokusebenza kahle kwamasistimu obuhlakani bokwenziwa, izici ezimbili ezibalulekile ezithonya ukusebenza nokusebenza kobuchwepheshe bobuhlakani bokwenziwa.
Ithiyori eyisisekelo ye-GFlowNets isuselwe ekufundeni kokuqinisa i-entropy, okufaka phakathi imigomo ehlobene nokuvumelana kwendlela. Lesi sisekelo sethiyori siphakamisa ukuthi i-GFlowNets ingasetshenziswa kuhlelo olubanzi lwezinhlelo zokusebenza ze-AI.
Njengoba i-AI iqhubeka nokuhlanganisa ezimpilweni zethu zansuku zonke, indima ye-GFlowNets ekuthuthukiseni amakhono e-AI iya ngokuya iba sobala. Ngekhono layo eliyingqayizivele lokukhiqiza imiphumela ehlukahlukene, ukuthuthukisa ukusebenza kahle kwedatha, nokubhekana nezinselele zokufunda komshini, i-GFlowNets isethelwe ukudlala indima ebalulekile ekuthuthukisweni kwe-AI.
Funda kabanzi Umhlahlandlela:
- I-ChatGPT nokuthi amanethiwekhi e-neural akufunda kanjani ukukhuluma - uhambo lweminyaka engama-30
- Yini i-multimodal intelligence yokufakelwa?
- Ungawengeza kanjani umbono we-AI ezinhlelweni zakho zokusebenza, i-SaaS, amasayithi namaphrojekthi?
- I-OpenAI Insider Ixoxa Nge-AGI kanye Nemithetho Yokukala Kwenethiwekhi Ye-Neural
- Ungayisebenzisa Kanjani I-PyTorch Yezicelo Zokufunda Okujulile - Umhlahlandlela Wabaqalayo
- Iyini inethiwekhi ye-AI neural futhi isebenza kanjani?
