Kule post, sabelana nawe ngokukhethiweyo kwemithombo yolwazi oluluncedo malunga neNzululwazi yeDatha evela kumququzeleli kunye ne-CTO ye-DAGsHub, uluntu kunye neqonga lewebhu lolawulo lwenguqu yedatha kunye nentsebenziswano phakathi kweenzululwazi zedatha kunye neenjineli zokufunda ngomatshini. Ukhetho lubandakanya imithombo eyahlukeneyo, ukusuka kwii-akhawunti ze-Twitter ukuya kwiiblogi zobunjineli ezipheleleyo, ezijoliswe kwabo bazi kakuhle into abayifunayo. Iinkcukacha phantsi kokusikwa.
Isuka kumbhali:
Uyiloo nto uyityayo, kwaye njengomsebenzi wolwazi udinga ulwazi olulungileyo lokutya. Ndifuna ukwabelana ngemithombo yolwazi malunga neNzululwazi yeDatha, ubukrelekrele bokwenziwa kunye nobuchwephesha obunxulumeneyo endibufumana buluncedo kakhulu okanye bunomtsalane. Ndiyathemba ukuba oku kuyakunceda nawe!
Ijelo likaYouTube elikufaneleyo ukuhambelana neziganeko zamva nje. Ijelo lihlaziywa rhoqo kwaye umamkeli unenzondelelo eyosulelayo kunye nethemba kuzo zonke izihloko ezigutyungelweyo. Lindela ukugubungela umsebenzi onomdla kungekuphela nje kwi-AI, kodwa nakwimizobo yekhompyuter kunye nezinye izihloko ezibukekayo.
Kwitshaneli yakhe yeYouTube, uYannick ucacisa uphando olubalulekileyo lokufunda ngokunzulu kwiinkcukacha zobugcisa. Kunokuba uzifundele ngokwakho isifundo, kudla ngokukhawuleza kwaye kulula ukubukela enye yeevidiyo zaso ukuze uqonde ngakumbi amanqaku abalulekileyo. Iingcaciso zidlulisela undoqo wamanqaku ngaphandle kokutyeshela imathematika okanye ukulahleka kwiipayini ezintathu. UYannick ukwabelana ngezimvo zakhe malunga nendlela izifundo ezidibana ngayo, ukuba nzulu kangakanani ukuthatha iziphumo, ukutolika okubanzi, kunye nokunye. Kunzima ngakumbi kwabaqalayo (okanye abo bangezongcali kwimfundo ephakamileyo) ukuba bazizele ngokwabo ezi zinto zifunyenweyo.
Ngamazwi abo:
Uphando lokufunda ngoomatshini kufuneka lucace, luguquke kwaye ludlamke. Kwaye i-Distill yadalwa ukunceda ngophando.
I-Distill yimpapasho ekhethekileyo kunye nophando kwinkalo yokufunda koomatshini. Amanqaku anemibono emangalisayo ayakhuthazwa ukunika umfundi ukuqonda okucacileyo kwezihloko. Ingcinga yendawo kunye nentelekelelo idla ngokusebenza kakuhle kakhulu ekuncedeni ukuqonda iSifundo soomatshini kunye nezihloko zeNzululwazi. Iifomathi zopapasho lwesintu, kwelinye icala, zithande ukuba ngqongqo kulwakhiwo lwazo, zimile kwaye zomile, kwaye ngamanye amaxesha. . UChris Olah, omnye wabadali beDistill, naye unebhlog yobuqu emangalisayo . Ayikahlaziywa kangangethuba elithile, kodwa iseyingqokelela yezona nkcazo zibalaseleyo ngesihloko sokufunda nzulu kwakha kwabhalwa. Ngokukodwa, yandinceda kakhulu I-LSM!
USebastian Ruder ubhala iblogi enengqiqo kakhulu kunye nencwadana yeendaba, ngokuyinhloko malunga nokudibana kwenethiwekhi ye-neural kunye nokumbiwa kwesicatshulwa solwimi lwendalo. Ukwanalo neengcebiso ezininzi kubaphandi kunye nezithethi zenkomfa, ezinokuba luncedo kakhulu ukuba ukwi-academia. Amanqaku kaSebastian athatha imo yophononongo, eshwankathela kwaye echaza imeko yophando lwangoku kunye neendlela kwindawo ethile. Oku kuthetha ukuba amanqaku aluncedo kakhulu kwiingcali ezifuna ukufumana ngokukhawuleza iibheringi zabo. USebastian naye uyabhala .
UAndrei Karpathy akafuni kwaziswa. Ukongezelela ekubeni ngomnye wabaphandi abadumileyo bokufunda okunzulu emhlabeni, udala izixhobo ezisetyenziswa ngokubanzi, umz. njengeeprojekthi ezisecaleni. Abantu abaninzi bangene kweli candelo ngekhosi yakhe yaseStanford , kwaye kuya kuba luncedo kuwe ukuyazi uqeqesho lwenethiwekhi ye-neural. Ndikwacebisa ukuyibukela malunga nemingeni yehlabathi yokwenyani uTesla kufuneka ayoyise xa ezama ukusebenzisa umatshini wokufunda kwisikali sobunzima kwihlabathi lokwenyani. Le ntetho ifundisa, iyachukumisa kwaye ixhokonxa ingqondo. Ukongeza kumanqaku malunga neML ngokwayo, u-Andrei Karpathy unika kuba . Funda uAndrey kwi kwaye kuqhubeke .
Ibhlog yobunjineli be-Uber inomtsalane ngokwenyani kwisikali sayo kunye nobubanzi bokugubungela, igubungela izihloko ezininzi, ngakumbi. . Eyona nto ndiyithanda kakhulu malunga nenkcubeko yobunjineli bakwa-Uber lutyekelo lwabo lokuvelisa umdla kakhulu kwaye uxabisekile umthombo ovulekileyo ngesantya esiqhekezayo. Nantsi eminye imizekelo:
- Kwaye uluhlu luyaqhubeka kwaye luqhubeke ... Hats off, Uber
Iimpikiswano ecaleni, ibhlog ye-OpenAI iyamangalisa ngokungathandabuzekiyo. Ngamaxesha ngamaxesha, ibhlog ithumela umxholo kunye nezimvo malunga nokufunda okunzulu ezinokuza kuphela kwisikali se-OpenAI: I-hypothetical. ukuhla kabini okunzulu. Iqela le-OpenAI lithanda ukuthumela ngokungaqhelekanga, kodwa oku zizinto ezibalulekileyo.
Iblogi yeTaboola ayaziwa njengeminye imithombo kule post, kodwa ndicinga ukuba yodwa - ababhali babhala malunga neengxaki eziphantsi kakhulu, zobomi bokwenyani xa bezama ukusebenzisa iML kwimveliso "yesiqhelo". " amashishini: ngaphantsi malunga neemoto eziziqhubayo kunye neearhente zeRL eziphumelele iintshatsheli zehlabathi, ngakumbi "ndazi njani ukuba imodeli yam ngoku iqikelela izinto ngokuzithemba kobuxoki?" Ezi ngxaki zifanelekile phantse wonke umntu osebenza kwintsimi, kwaye bafumana ukushicilelwa okuncinci kunezihloko eziqhelekileyo ze-AI, kodwa kusathatha italente yehlabathi ukusombulula ezi ngxaki ngokuchanekileyo. Ngethamsanqa, uTaboola unazo zombini ezi talente kunye nokuzimisela kunye nokukwazi ukubhala ngayo ukuze nabanye abantu bafunde.
Kunye ne-Twitter, akukho nto ingcono kwi-Reddit kunokuba ubanjwe kuphando, izixhobo, okanye ubulumko besihlwele.
Izithuba zipapashwa kuphela ngonyaka, kodwa zizaliswe ngolwazi oluxinana kakhulu. Xa kuthelekiswa neminye imithombo kolu luhlu, olu lufikeleleke ngakumbi kubantu abangengobuchwephesha beshishini. Into endiyithandayo malunga neentetho kukuba izama ukubonelela ngombono obanzi ngakumbi wokuba ishishini kunye nophando lusingise phi, ludibanisa inkqubela phambili kwi-hardware, uphando, ishishini, kunye ne-geopolitics kwimboniselo yeso lentaka. Qiniseka ukuba uqala ekugqibeleni ukufunda malunga nokungqubana kwemidla.
Iipodcast
Ngeliphandle, ndicinga ukuba iipodcasts azikufanelanga ukuphonononga izihloko zobugcisa. Emva kwayo yonke loo nto, basebenzisa kuphela iaudio ukuchaza izihloko, kwaye isayensi yedatha yinkalo ebonakalayo kakhulu. Iipodcasts zidla ngokukunika isizathu sokwenza uphando olunzulu ngakumbi kamva okanye ube neengxoxo zefilosofi ezinomdla. Nangona kunjalo, nazi ezinye iingcebiso:
- , njengoko ethetha nabaphandi abadumileyo kwinkalo yobukrelekrele bokwenziwa. Iziqendu kunye noFrancois Chollet zilungile ngakumbi!
- . Kuhle ukuva malunga nezixhobo ezintsha zeziseko zedatha.
Izintlu ezoyikekayo
Kuncinci ukugcina umkhondo apha, kodwa izixhobo ezininzi eziluncedo xa usazi ukuba ukhangela ntoni:
UMatty ufumana iindlela ezintle, ezinobuchule zokusebenzisa uthungelwano lwe-neural, kwaye kuyonwabisa nje ukubona iziphumo zakhe kwifidi yakho ye-Twitter. Ubuncinci jonga ngokukhawuleza.
U-Ori ngumatshini wokuqhuba nje . Ubhala ngokubanzi malunga neengxaki kunye nezisombululo zezazinzulu zedatha. Qiniseka ukuba urhuma ukuze waziswe xa inqaku lipapashwa. Yakhe , ngokukodwa, ichukumisa ngokwenene.
Umseki we-fast.ai, umthombo obanzi wokudala kunye nemveliso.
Injineli ye-ML yabasebenzi e-Github, u-Hamel Hussain uxakeke emsebenzini wokudala kunye nokunika ingxelo ngezixhobo ezininzi zeekhowudi zedatha.
Umdali weKeras, ngoku hlaziya ukuqonda kwethu ukuba yintoni na ubukrelekrele kunye nendlela yokubuvavanya.
Uphando lweNzululwazi kuGoogle Brain.
isiphelo
Isithuba sokuqala sinokuhlaziywa njengoko umbhali efumana imithombo emikhulu yomxholo ukuba kuya kuba lihlazo ukungabandakanyi kuluhlu. Zive ukhululekile ukuqhagamshelana naye ku , ukuba ufuna ukucebisa umthombo omtsha! Kwaye kwakhona DAGsHub Gqwetha [malunga. inguqulelo umsebenzi woluntu] kwiNzululwazi yeDatha, ngoko ukuba udala umxholo wakho weNzululwazi yeDatha, zikhululeke ukubhala kumbhali weposi.
Ziphuhlise ngokufunda imithombo ecetyiswayo, kunye nokusebenzisa ikhowudi yokuthengisa IHABR, ungafumana i-10% eyongezelelweyo kwisaphulelo esibonakaliswe kwibhena.
Izifundo ezingakumbi
Amanqaku akhoyo
umthombo: www.habr.com