Slav Petrov
Vice President, Research at Google DeepMind; co-leads Post-Training for Gemini
Almaañ Themes
Slav Petrov waa cilmi-baare reer Bilgariya ah, isla markaana ah Madaxweyne Ku-xigeenka Cilmi-baarista ee Google DeepMind, halkaas oo uu wada-hoggaamiyo Post-Training-ka Gemini, dadaalka Google ee moodelka luqadda weyn. Wuxuu haystaa PhD-da Jaamacadda California, Berkeley, wuxuuna ka soo shaqeeyay baridda Statistical Natural Language Processing ee Jaamacadda New York.
Cilmi-baaristiisu waxay diiradda saartaa habaynta luqadda dabiiciga ah iyo moodellada luqadda waaweyn, shaqadiisuna waxay gacan ka geysatay awoodaha luqadeed ee waxyaabaha Google ka mid ah sida Search, Assistant, Translate, iyo Cloud. Wuxuu helay abaalmarinno badan oo Best Paper ah shirar waaweyn oo NLP ah (ACL 2011, NAACL 2012, ACL 2016), iyo sidoo kale abaalmarin 10-sano ah oo Test-of-Time ah ACL 2023, waxaana la guddoonsiiyay Abaalmarinta John Atanasoff sanadkii 2014 Madaxweynaha Bilgariya.
Asali ak Yënguñ ak Xelal
Safar ak waxaan Petrov xamleewul ci jëfandikoo jëm ci sedd bi, waaye li am solo ci biir 3 yoon wi, te am nañu ak dencukaay bi ci jëfandikoo ci kariyer bi. Li mu daldi wàcc ci Diplom ci Computer Science ak Business Administration ci Freie Universität Berlin, bu amul man ci wàccal doctoral studies ci Computer Science ci University of California, Berkeley. Dañu amul ci Duke University ci xelal bu mu wàccal akademi bi awal. Ci 2010 ak 2016, mu waxul ci Adjunct Professor ci New York University, bu mu waxe Statistical Natural Language Processing — waxtaan bi, li am nañu ci jëfandikoo ci industry bi ci Google, bu mu jëfandikoo ci August 2009.
Wàllu ci Google DeepMind
Ci Google DeepMind, Petrov am na wàllu bu Vice President of Research, te mu co-lead Post-Training ci Gemini, modèl bu am solo ci gwoñi language model bi. Post-training waa faz bi mu am nañu ci wàll bi ci jëfandikoo large language models, te li am nañu ci mbind mi mu jëfandikoo ci xam-xam bu rafet dencukaay bi modèl bi, xam-xam bu am ci instruction-following, ak alignment bu amul ci pretraining bu awal. Ekip bi, ñu rëy ci sa bopp ci dund bi, waaye li am nañu ak baaxal ak wàllu bu baax. Bu amul Google DeepMind, Petrov wàccal NLP research bu mu yàgg ci Google products bi, ngir Search, Ads, Translate, Assistant, ak Chrome.
Li mu dafa daan ci publication, am nañu sama ci wàllu bu kawe ci yàgg, te li am nañu ci jëfandikoo bu baax ci natural language processing, ay mbind ci machine translation, parsing, ak attribution ci natural language generation. Mu am nañu ci jëfandikoo ci papers bu am nañu ci landmark, bu wàcc ci wàllu large model bi ci Google, ay work bu am ci PaLM, PaLM 2, Gemini, Gemini 1.5, ak Gemma 2, ak biir dataset Natural Questions benchmark — resource bu ñu jëfandikoo ñu baax ci research ci question answering.
Kontribusyon ci Research ak Jëfandikoo bu ñu am
Output bi ci research bi Petrov am na kontribusyon ci syntactic parsing, machine translation, ak wàll bi ci develop evaluation frameworks ci language generation. Wone area bu mu defar ci yeneen, am na ci measuring attribution ci natural language generation models — bu ñu jëfandikoo ngir xam se outputs bi modèl bi, ñu man nañu ci sources bu ñu wone, te wala li output bi wàcc ci li ñu wone. Concern bi, dafa wax ci wàllu bu amul ci deployment ci large generative systems. Work bi, bu ñu publie ci Computational Linguistics ci 2023, mu wax na ci interest bu baax ci rafet se outputs ci language model bi am nañu ci lëkkal ak interpretability.
Bu kawe ci awards ci paper bu mu wone ci ACL ak NAACL, Petrov am na 10-year Test-of-Time Award ci ACL 2023, bu ñu wàccal ngir research bu mu wàccal ci decade bu amul, te li mu daldi wàcc ci sa influence ci biir jëfandikoo bi. Areas ci research bi, bu ñu taxaw ci Google research directory, span Natural Language Processing, Machine Translation, Machine Intelligence, ak General Science.
Asali ak Yeneen Achievements
Petrov dafa jëfandikoo ci Bulgaria, Germany, ak United States ci sa bopp, ci roughly equal measure, te am nañu ci Berlin. Achievements bu mu wone ci awal ci compétition, am na World Championship ci RoboCup 2004 — compétition ci robotics bu am nañu ci international — bu amul ci kariyer bi ci NLP research. Mu daldi wàccal Diplom bi ci Freie Universität Berlin, te waxtaan bi mu jëfandikoo ci Germany, dafa am ci bookended kariyer bi, bu Berlin am nañu ci base bi mu am ci amul, xam-xam bu mu leads research ci yeneen ci biir yeneen organisasi AI bu am nañu ci baax. Mu dafa wax ci loyalty bu mu am ci ekip bu jàmm ci Bulgaria, gaaru ci matches bu am ci Germany.