Slav Petrov

Tarihin Ilimi da Aiki

Hanyar ilimin Petrov ta shafi ƙasashe uku kuma tana nuna yanayin aikinsa na ƙasashen waje. Ya kammala Diplom a fannin Kimiyyar Kwamfuta da Kasuwanci a Freie Universität Berlin kafin ya ci gaba da karatun digiri na uku a fannin Kimiyyar Kwamfuta a Jami'ar California, Berkeley. Daga baya ya yi karatu a Duke University a matsayin wani ɓangare na matakin karatunsa na farko. Daga 2010 zuwa 2016, ya yi aiki a matsayin Adjunct Professor a New York University, inda ya koyar da Statistical Natural Language Processing — wannan matsayi ya kasance tare da aikin masana'antu da yake yi a Google, inda ya shiga a watan Agustan 2009.

Matsayi a Google DeepMind

A Google DeepMind, Petrov na riƙe da matsayin Mataimakin Shugaban Bincike kuma yana haɗin gwiwa wajen jagorantar Post-Training ga Gemini, babban samfurin kamfanin na manyan harsunan da ake amfani da su wajen fahimtar rubutu. Post-training muhimmin mataki ne a ci gaban manyan harsunan harshe, wanda ya ƙunshi dabarun da ke inganta halayen samfurin, iya bin umarni, da daidaituwa bayan pretraining na farko. Ƙungiyarsa tana da rarrabuwar kai a duk duniya, wanda ke nuna girma da rikitarwar aikin da ake yi. Kafin kafa Google DeepMind, Petrov ya jagoranci ƙoƙarin binciken NLP da ya ciyar da kai tsaye cikin samfuran Google kamar Search, Ads, Translate, Assistant, da Chrome.

Takardun da ya wallafa sun wuce shekaru goma, kuma sun ƙunshi fannoni da dama a cikin sarrafa harshe ta dabi'a, ciki har da fassarar injiniya, parsing, da dangantawa a samar da rubutu ta harshe ta dabi'a. Shi mai ba da gudummawa ne ga manyan takardu da ke da alaƙa da ƙoƙarin manyan samfuran Google, ciki har da aiki a kan PaLM, PaLM 2, Gemini, Gemini 1.5, da Gemma 2, da kuma bayanan ma'aunin Natural Questions — wani abu da ake yawan amfani da shi wajen binciken amsa tambayoyi.

Gudummawar Bincike da Gane-Gane

Abubuwan da Petrov ya fitar a bincike sun haɗa da gudummawa a fannonin syntactic parsing, fassarar injiniya, da kuma gina tsarin tantancewa don samar da rubutu. Wani shahararren sabon yanki na aiki ya shafi auna dangantaka (attribution) a cikin samfuran samar da rubutu ta harshe ta dabi'a — binciken ko ana iya gano abin da samfurin ya fitar zuwa ga ingantattun kafofin da za a iya tabbatarwa, lamarin da ya ƙara zama muhimmanci tare da aiwatar da manyan tsarin samar da rubutu. Wannan aiki, da aka wallafa a Computational Linguistics a 2023, yana nuna sha'awar da ta fi faɗi wajen sa fitarwar samfuran harshe ta zama mafi dogaro da kuma mai sauƙin fahimta.

Baya ga kyaututtukan takardunsa a ACL da NAACL, Petrov ya samu kyautar Test-of-Time mai shekaru 10 a ACL 2023, wadda ta gane binciken da aka yi shekaru goma da suka wuce wanda ya nuna ya ci gaba da tasirinsa a fannin. Fannonin bincikensa, kamar yadda aka jera a kundin binciken Google, sun haɗa da Natural Language Processing, Machine Translation, Machine Intelligence, da General Science.

Tushen Bayanai da Sauran Nasarori

Petrov ya rayu a Bulgaria, Jamus, da Amurka a kusan daidai gwargwado, kuma a yanzu yana zaune a Berlin. Nasarorinsa na farko a gasa sun haɗa da lashe Gasar Duniya a RoboCup 2004 — gagarumin gasa ta robotika ta ƙasashen waje — wadda ta gabaci aikinsa na binciken NLP. Ya kammala Diplom dinsa a Freie Universität Berlin, kuma lokacin da ya yi a Jamus ya zama kamar ya rufe aikinsa ta hanyar da ta dace, inda Berlin ta zama tushensa na yanzu ko da yake yana jagorantar bincike ga ɗaya daga cikin manyan ƙungiyoyin AI a duniya. Ya bayyana da kansa cewa yana da aminci ga ƙungiyar ƙwallon ƙafa ta ƙasar Bulgaria, musamman a wasannin da suke yi da Jamus.