Our team is a group of researchers from Moscow Institute of Physics and Technology (MIPT), Moscow Engineering Physics Institute (MEPhI), Russian Technological University (MIREA), Moscow State University (MSU), and other leading scientific and scholarly institutions. Under the scientific supervision of Dr. Alexander Sboev, we work on data analysis, machine learning (especially neural networks), and mathematical modelling, for various applied fields. Since 2011, our team has been doing research on the development and application of intelligent methods for analyzing data from various domains: texts, time series, audio recordings, images, and signals. Since then, notable experience in both fundamental and applied research has been gained.
In the field of text analysis, our team has developed and improved models for morpho-syntactic analysis, keyword extraction, search for nested topics in a given set of documents, evaluating the emotional coloring and aspect-based sentiment of messages, and profiling the author of a text, in particular determining the author’s gender and age. In recent years, a corpus of labelled Internet texts with feedback on medications has been prepared with markup suitable for recognizing named entities, establishing relations between them and relating them to a given dictionary (Entity linking or normalization). State-of-the-art models for these tasks have been developed.
In the field of time series, we have developed models for predicting optimal parameters of technical processes in metal rolling and oil and gas production, and epidemics forecasting models for the COVID-19 outbreak in different countries.
The direction of audio data analysis is motivated by the prospective possibility of implementing energy-efficient computing systems on base of neuromorphic processors and spiking neural networks (SNN). Our team has developed a few models of SNN learning based either on bio-inspired learning mechanisms or on converting pre-trained conventional neural networks.
The direction of image and signal processing is being explored as part of our work on an interface for controlling the motions of an agent in an environment via voice commands and the operator’s gaze. Modules have been created for analyzing the direction of the operator’s gaze on the image from the agent’s camera, and for analyzing operator’s commands (detecting mentions of objects, speed, direction of movement and other key attributes of the commands; reconstructing missed words; prioritizing commands, and so on). A prototype of a control interface has been developed. Based on the reinforcement learning approach, a neural network model of agent movement control has been developed for solving the follow-the-leader task.
Our team has performed and is currently performing a number of projects on analyzing textual data, creating medical diagnostic systems, analyzing industrial manufacturing processes, and developing efficient neuromorphic computing algorithms. We actively cooperate with Russian universities: our team members supervise bachelor, master and postgraduate theses in MIPT, MEPhI, MIREA, and MSU.
Collaboration is welcome!
The team has embarked on an RScF project Development of effective methods for training spiking neural networks for implementation on promising energy-efficient neuroprocessors
February 06–10, 2023Roman Rybka have a talk at 2-nd SCO Young Scientist’ Conclave, Bengaluru, India
May 2023The team has embarked on a joint RScF-NSFC project Nonlinear mathematical physics approaches for studying processes in fiber lasers and nonlinear control and excitation of novel soliton localized modes