The Neural Networks and Deep Learning Laboratory at MIPT conducts research in the field of deep neural network architectures for working with text in natural language, namely: creating algorithms capable of finding sequences of actions without a "teacher" to solve a task in an unknown environment. Since June 2017, the main activity of the laboratory is implementation of the DeepPavlov project. DeepPavlov is an open source library for text analysis and creation of dialog systems. The project is aimed at developing conversational neural network machine intelligence that can carry on a meaningful dialog with a person and achieve the goal set in the dialog. In the course of this work, the technological barrier in the field of machine intelligence algorithms will be overcome by consolidating knowledge about the mechanisms of the brain activity and the modern theory of deep machine learning.
The laboratory staff includes about 30 researchers and engineers, including 3 candidates of science, 7 postgraduates, and 2 students. The laboratory staff has published more than 50 scientific papers. The laboratory's hardware consists of a specialized computer for calculating neural network models (31st place in the Russian supercomputer rating), a mini cluster, and workstations with a total capacity of 125 GPUs.
Since 2016, every semester the laboratory conducts courses on "Applying Deep Learning of Neural Networks to Natural Language Processing Tasks" and "Reinforcement Learning". In the 2017–2018 academic year, the laboratory's employees were involved in the Machine Learning School for High School Students. In July 2018, the first CISS School on using the DeepPavlov library was held. In July 2019, the first international CISS 2 NLP School was held at Lowell University (USA).
The Neural Networks and Deep Learning Laboratory at MIPT held six international week-long scientific hackathon schools, which were attended by about 1000 people, and 40 leading researchers in the field of machine intelligence (Google Brain, Google DeepMind, openAI, Facebook AI Research, University of Oxford, New York University, Carnegie Mellon University, Stanford University, among others) gave public lectures, which were attended by more than 500 people. The lectures can be found on the laboratory's YouTube channel. This allowed us to form a pool of scientists and developers to implement the project and establish academic links with leading research centers in the field of artificial intelligence.
The laboratory of neural systems and deep learning also has industry recognition. In November 2015, it received a grant of $18,000 to use IBM cloud resources for research, and in January 2016 it received the status of NVIDIA GPU Research Center. In July 2016, the Lab won a Facebook grant in the form of GPU servers (one of 14 labs from 8 countries). In July 2019, a team of postgraduates from the laboratory made it to the final of the Amazon competition — Alexa Prize Socialbot Challenge 3. In 2019, the DeepPavlov project won the "Powered by TF Challenge" competition from Google for the best machine learning project that uses the TensorFlow library. In November 2020, a team of postgraduates from the laboratory made it to the final of the Amazon competition — Alexa Prize Socialbot Challenge 4.
We write for the popular IT and ML blogs and have our own blog on Medium.