Natural language is the most accessible and flexible source of data on human thinking. The result of the intellectual activity of many people, the text data, allows modern neural networks to generalize knowledge, acquire skills, and reproduce new texts for given tasks.

The NLP department combines research in the field of language modelling, benchmarks, and creating multilingual and multimodal models.
Multilingual Model Training
Models uniting texts, images, speech, sound
BlackboxNLP and low-resource languages problem solving
Large and reliable corpora collection in 60+ languages
Testing intellectual abilities of humans and models
There are several known problems that hinder the effective use of RL. Those are, the issue of effective exploration of environments with a rare reward, the complications in adapting the agents to changes in the environment and catastrophic forgetting when moving to a new environment.

There are multiple ways that partially solve these problems, but there is no singular approach that flexibly combine them into a general solution that would allow to effectively apply reinforcement learning to the previously mentioned complex problems.

As part of Synergy RL approach, we propose a new reinforcement learning method that would combine individual ideas into a universal framework.
Synergy RL
We are open for cooperation. Send us an e-mail, if you would like/want to work with us/to join us.
Team lead
Tatiana Shavrina
© 2021
Russia, Moscow
Nizhny Susalny lane 5 p. 19