Nikolay Bogoychev

The University of Edinburgh
School of Informatics
Room 4.23, Informatics Forum
10 Crichton Street
Edinburgh
EH8 9AB
United Kingdom

Nikolay Bogoychev

About me

I am a PhD student at The University of Edinburgh Institute for Languages, Cognition and Computation, affiliate of the Pervasive Parallelism CDT. My supervisors are Adam Lopez and Kenneth Heafield.

I like languages, logographic writing systems, game theory and GPUs. I (try to) make things run faster and enjoy (premature) optimization. In my spare time I learn languages and play football.

Linkedin Github StackOverflow CV

PhD work

I am exploring opportunities for exploiting the raw computational power of GPUs in the domain of Machine translation.

Machine translation decoding is extremely resource intensive. For the tuning of a single model the decoder needs complete 25 iterations of translating the tuning set which is usually around 3000 sentences but if more data is available usually it is used. As such for more complex models it is frequently the case that the whole process will take couple of days. To make the matter worse frequently many models with subtle variations of each other are tuned to test different hypothesis and often researchers limit themselves on the amount of experiments they run because of the time requirements.

Machine translation decoders employ the "one thread, one sentence" computational model as decoding itself is embarrassingly parallel. Knowing that, could we map the problem to a GPU, resulting in a system that will perform on par with a computational cluster, but costing only a fraction of it?

Education

2014: Graduated from The University of Edinburgh with a First class Bachelor's degree in Artificial Intelligence & Computer science.

2015: Started a PhD in The University of Edinburgh ILCC.

Publications

Nikolay Bogoychev and Adam Lopez (2016). N-gram language models for massively parallel devices. In Proceedings of ACL, Berlin, Germany. [pdf] [bib]

Nikolay Bogoychev and Hieu Hoang (2016). Fast and highly parallelizable phrase table for statistical machine translation. In Proceedings of WMT, Berlin, Germany. [pdf] [bib]

Hieu Hoang, Nikolay Bogoychev, Lane Schwartz and Marcin Junczys-Dowmunt (2016). Fast, Scalable Phrase-Based SMT Decoding. In Proceedings of AMTA 2016, Austin, Texas [pdf] [bib]

Barry Haddow, Matthias Huck, Alexandra Birch, Nikolay Bogoychev, Philipp Koehn (2015). The Edinburgh/JHU Phrase-based Machine Translation Systems for WMT 2015. In Proceedings of WMT, Lisboa, Portugal. [pdf] [bib]

Alexandra Birch, Matthias Huck, Nadir Durrani, Nikolay Bogoychev, Philipp Koehn (2014). Edinburgh SLT and MT System Description for the IWSLT 2014 Evaluation. In Proceedings of IWSLT, Lake Tahoe, USA. [pdf] [bib]

The University of Edinburgh
School of Informatics
Institute for Languages, Cognition and Computation