Slav Petrov
Vice President, Research at Google DeepMind; co-leads Post-Training for Gemini
Themes
Slav Petrov na en researcher for Bulgaria, e get position Vice President for Research for Google DeepMind. For there, e dey co-lead Post-Training wey dey come after training for Gemini, the big language model project for Google. E get PhD for University of California, Berkeley, and e don teach Statistical Natural Language Processing for New York University.
Research wey e dey do focus on natural language processing and large language models, and work for e don help add language capabilities to Google products like Search, Assistant, Translate, and Cloud. E don win multiple Best Paper Awards for big NLP conferences (ACL 2011, NAACL 2012, ACL 2016), plus 10-year Test-of-Time Award for ACL 2023. Also, for 2014, the President of Bulgaria award am John Atanasoff Award.
Akademik na Profeshinal Background
Di pikin pikin irin-ajo akademik Petrov ràn nínú tólú nàìjírí (country) na ó hàn say karíyà-him na international. Ó pari Diplom for Computer Science and Business Administration for Freie Universität Berlin, bifo e go start doctoral studies for Computer Science for University of California, Berkeley. Later, ó kẹ́kọ̀ọ́ for Duke University gẹgẹ bi apá nínú di pátápátá akademik him. Tí ó wà láti 2010 títí di 2016, ó ṣiṣẹ́ gẹgẹ bi Adjunct Professor for New York University, níbi tí ó kọ́ni Statistical Natural Language Processing — rolù náà ń lọ pẹ̀lú iṣẹ́ di shíshẹ̀ (industry) him for Google, níbi tí ó darapọ̀ ní August 2009.
Rolù Petrov ní Google DeepMind
For Google DeepMind, Petrov ní ipo Vice President of Research, ó sì ń co-lead Post-Training for Gemini, di flagship large language model di konpani. Post-training na apá pàtàkì nínú bí wọ́n ṣe ń dá large language models sílẹ̀, ó ní àwọn technique tí ń mú kí model náà yí padà dáadáa—bí e ṣe ń hù, bí e ṣe ń tẹ̀lé instruction, àti bí e ṣe ń align lẹ́yìn pretraining tó bẹ̀rẹ̀. Ẹgbẹ́ rẹ̀ wà ní ọ̀pọ̀ ibi káàkiri ayé, ó ń hàn bí iṣẹ́ náà tóbi tó àti bí e ṣe nira. Ṣáájú kí Google DeepMind dá, Petrov ti ń darí NLP research efforts tí ń wọlé taara sínú àwọn Google products bí Search, Ads, Translate, Assistant, àti Chrome.
Di record pùbìlìshìng rẹ̀ ti gbojá jù ọdún méjìlá (over a decade) ó sì bo ọ̀pọ̀ topic nínú natural language processing, títí kan machine translation, parsing, àti attribution nínú natural language generation. Ó jẹ́ contributor sí àwọn landmark papers tó jọmọ iṣẹ́ large model Google, pẹ̀lú iṣẹ́ lórí PaLM, PaLM 2, Gemini, Gemini 1.5, àti Gemma 2, àti pẹ̀lú Natural Questions benchmark dataset — ohun elo tí wọ́n ń lò púpọ̀ fún research question answering.
Research Contributions àti Recognition
Di research output Petrov ní contributions kọjá syntactic parsing, machine translation, àti bí wọ́n ṣe ń dá evaluation frameworks fún language generation. Ọ̀kan nínú area iṣẹ́ tuntun tó yàtọ̀ ni bí wọ́n ṣe ń wiwọn attribution nínú natural language generation models — ìwádìí bóyá a lè tọ́pa outputs model náà padà sí orísun tí a lè fìdí rẹ̀ múlẹ̀, ìbànújẹ́ tó ti ń pọ̀ sí i nígbà tí wọ́n bẹ̀rẹ̀ sí deploy large generative systems. Iṣẹ́ yìí, tí wọ́n pùbìlìshì sí Computational Linguistics ní 2023, ń hàn pé ó ní ìfẹ́ tó gbooro sí bí outputs language model ṣe lè gbẹ́kẹ̀lé jù, kí ó tún rọrùn láti túmọ̀.
Lẹ́yìn àwọn paper awards rẹ̀ ní ACL àti NAACL, Petrov gba 10-year Test-of-Time Award ní ACL 2023, tí ń mọ̀rírì research tí ó ṣẹlẹ̀ ní ọdún mẹ́wàá sẹ́yìn, tí ó fi hàn pé ipa rẹ̀ ṣi ń tẹ̀síwájú títí di òní nínú field náà. Di research areas rẹ̀, gẹgẹ bí a ṣe kà á sí Google research directory, bo Natural Language Processing, Machine Translation, Machine Intelligence, àti General Science.
Background àti Àwọn Achievement Miì
Petrov ti gbe ní Bulgaria, Germany, àti United States ní bí ó ṣe dà bíi pé gbogbo wọn tó dọ́gba, ó sì báyìí ó dá sí Berlin. Di achievement competitive rẹ̀ ní ìbẹ̀rẹ̀ ni World Championship ní RoboCup 2004 — ìdíje robotics international — tí ó ṣáájú karíyà rẹ̀ nínú NLP research. Ó pari Diplom rẹ̀ ní Freie Universität Berlin, ó sì dà bí ẹni pé ní Germany ni karíyà rẹ̀ bẹ̀rẹ̀ títí dé òpin, pẹ̀lú Berlin gẹgẹ bi base rẹ̀ ní báyìí, bó tilẹ̀ jẹ́ pé ó ń darí research fún ọ̀kan nínú àwọn AI organizations tó gbajúgbajà jù lọ ní ayé. Ó ti sọ pé ó ní ìdúróṣinṣin ti ara rẹ̀ sí ẹgbẹ́ bọọlu orilẹ-ede Bulgaria, pàápàá jù lọ nínú match lòdì sí Germany.