12/21/2017 15:33:50

MIPT will hold machine translation hackathon

As of Dec. 18, the DeepHack.Babel hackathon has opened registration for the qualification round. From Jan. 29 until Feb. 4, 50 participants that pass the test will have a chance to apply their knowledge to improve machine translation technology. The deadline for the qualification round is Jan. 8. To participate, sign up on the event’s website.

DeepHack.Babel is the fifth in a series of DeepHack events organized by MIPT’s Neural Networks and Deep Learning Lab. This year, the hackathon will focus on machine translation, with a special emphasis on neural machine translation (NMT) — a method that is gaining popularity among researchers and has already found commercial application. Contrary to the conventional statistical approach, NMT is based on artificial neural networks, which helps to achieve higher-quality translation. Companies such as Google, Yandex, and Microsoft are already using this technology in their translation services.

“The purpose of the MIPT hackathon is to develop learning techniques for machine translation that do not make use of parallel texts,” explains Mikhail Burtsev, the head of the laboratory. “The fact is, machine translation systems, like people, learn to translate by analyzing correct translations. However, what the hackathon participants will try to do is train a system so it could work without such references. The goal is for the program to learn to translate using two unrelated texts. This technology will broaden the scope of machine translation and improve automated translation from rare languages, for which not enough high-quality parallel data is available, since virtually any language has its own monolingual corpus. Moreover, if we succeed, it may even further improve machine translation models for languages that do have sufficient data.”

As usual, the DeepHack hackathon will also host a sci-hack school. Lectures on machine translation research will be given by the world’s leading experts from Unbabel, Yandex, Apple, Carnegie Mellon University (United States), the University of Leeds and the University of Sheffield (Great Britain), the Humboldt University (Germany), and Dublin City University (Ireland), as well as from other major research institutes. The lectures are open to the public but require registration. There will also be a live webcast of the lectures on the YouTube channel of DeepHack.

“DeepHack is probably Russia’s only hackathon platform of such scale and with a history of its own,” says Viktor Portnov, the head of the Data Science Department at Sberbank, who took part in the event last year. “We spent a week immersed in problem-solving, tackling cutting-edge data science and artificial intelligence issues. It was incredibly engaging, challenging, and intense up until the very last minute. Not only did we win, which was especially pleasing, but we also upgraded our NLP skills and met many interesting people. If there’s a chance to participate in the hackathon once more, we will definitely take it!”

The hackathon is organized by MIPT’s Neural Networks and Deep Learning Lab as part of the iPavlov project. For two and a half years, the researchers will collaborate with the largest machine learning scientific centers in order to develop conversational AI technology. The research findings will be published as an open-source library so that anyone who designs dialogue systems will be able to use the data for scientific or practical purposes.

The general partners of the hackathon include the National Technology Initiative Fund and Sberbank. The event has also been endorsed by the Open Data Science community, the NeuroNet Branch Union, and the Northern Biopharmaceutical Cluster.

If you have noticed a mistake on this page, select it and press Ctrl + Enter

Related posts


Remote jets are clearer now

Presidential prize awarded to head of MIPT’s Nanobiotechnology Lab
Macromolecular order in plastic kingdom
​Scientists develop a ‘third eye’ to examine stem cells