Research Groups (Center for Artificial Intelligence Technology)

Computational Intelligence, Prof. Ivan Oseledets

Google Scholar, I.Oseledets@skoltech.ru.
Research focuses on the development breakthrough numerical techniques (matrix and tensor methods) for solving a broad range of high-dimensional problems and their combination with deep learning in order to reduce computational complexity and improve robustness of such algorithms.

Robustness of deep learning models: how to build adversarial attacks, how to detect adversarial attacks and how to train more robust and better certified models in real-world scenarios.Tensor methods in ML: the use of tensor approximations in building new classifiers and density estimators, especially for tabular data.
Non-euclidean geometry and deep learning: the usage of hyperbolic spaces to improve over state-of-the art models in deep learning
Improving explainability in DL models: how to enforce additional constrains on the hidden representations to make the resulting model less unpredictable
Recommender systems: building new efficient approaches for recommender systems
Physics-informed machine learning: development of efficient solvers for equations of mathematical physics based on machine learning
Different applied CV/NLP projects in different domains, including agriculture, design and recommender systems.

Courses delivered in Skoltech: Numerical Linear Algebra, Deep Learning, Research Methodology, Introduction to Recommender Systems.

Mobile Robotics, Prof. Gonzalo Ferrer

Google Scholar, G.Ferrer@skoltech.ru.

Research focuses on the autonomy for mobile robots, that is, being able to deploy robots in indoors, urban or outdoors environments and robots being able to execute any given task efficiently, reliably and safely. To achieve this, it is important to investigate on Perception (how to perceive and understand the world) and Action (how to actuate in succeed in a given task) and the combination of both topics into new solutions.

Show more:

Mapping and Localization with Tensor based TSDF
Neural Matchers for Visual Localization
Deep Learning Depth Prediction for SLAM
Human Gesture Detection and Motion Prediction
3D Mapping with Colored Point Clouds

Courses delivered in Skoltech: Planning Algorithms in Artificial Intelligence, Perception in Robotics.

Natural Language Processing, Prof. Alexander Panchenko

Google Scholar, A.Panchenko@skoltech.ru.

Main research interest is computational lexical semantics, including word sense embeddings, word sense induction, extraction of lexical resources, and other related topics. Research group members are interested in argument mining, neural and statistical natural language processing, information retrieval, knowledge bases, machine learning and intersections/interactions of these fields.

Lexical semantics (word sense induction and disambiguation, frame induction and disambiguation, semantic similarity and relatedness, sense embeddings, automated construction and completion of lexical resources such as WordNet and FrameNet)
Argument mining (comparative argument mining, and argument retrieval)
Learning representations of linguistic symbolic structures (graphs) such as knowledge bases and lexical resource
NLP for a better society: recognition of fake news, hate speech, and related phenomena
Textual style transfer
Tensors for NLP: learning representations based on tensor decompositions and tensor networks

Courses delivered in Skoltech: Introduction to Natural Language Processing, Deep Learning for Natural Language Processing.

Intelligent Signal and Image Processing, Prof. Anh Huy Phan

Google Scholar, A.Phan@skoltech.ru.

The group research interests include multilinear algebra, tensor computation, tensor networks, nonlinear system, blind source separation, and brain–computer interface.

Tensor structure for Light Convolution Neural Networks
Efficient architectures for GAN compression
Parallel estimation method for tensor decomposition based on rank-1 tensor approximation
Tensor network for intelligent image analysis
Tensor Networks for Machine Learning
Deep Learning Methods for Brain Computer Interface

Courses delivered in Skoltech: Convex Optimization and Applications, Tensor Decompositions and Tensor Networks in Artificial Intelligence.

Multiscale Neurodynamics for Intelligent Systems, Prof. Jun Wang

Publications, Jun.Wang@skoltech.ru.

Sparse Bayesian learning via neurodynamic optimization and its applications to nonlinear system identification, time series forecasting, stock index tracking etc
Neurodynamics-driven throughput optimization for simultaneous wireless information and power transfer in wireless sensor networks
Neurodynamics-driven Unit commitment and economic dispatch in electric power grids
Constrained supervised learning in neural ordinary differential equations via collaborative neurodynamic optimization
Neurodynamics-driven portfolio selection with various performance criteria and constraints

Mathematical Foundations of AI, Prof. Dmitry Yarotsky

Google Scholar, D.Yarotsky@skoltech.ru.

The research group has wide range of interests including theoretical and applied machine learning, approximation theory, optimization and mathematical physics.

Theoretical Results in Deep Learning
Machine Learning in Music Analysis

Courses delivered in Skoltech: Theoretical Methods of Deep Learning, Scientific Computing.

AI & Supercomputing, Prof. Sergey Rykovanov

Google Scholar, S.Rykovanov@skoltech.ru.

In Skoltech Sergey and his team develop a portfolio of HPC related educational courses and perform state-of-the-art research in HPC and its applications. Sergey has strong competence in HPC, especially in massively parallel computational plasma physics, and in design of novel particle accelerators and X-ray sources.

Development of the radiation calculation toolkit on GPUs
High Performance Computing for condensation driven aggregation problems
Virtualization of scientific workflow on supercomputers
Development of a High Performance Particle-In-Cell code for Plasma physics simulations
Virtualization and containerization in modern HPC environments (Kubernetes/Docker/Singularity)

Courses delivered in Skoltech: High Performance Computing and Modern Architectures, High Performance Python Lab, Parallel Computing in Mathematical Modeling and Data-Intensive Applications.

Quantum algorithms for machine learning and optimisation, Prof. Vladimir Palyulin

Google Scholar, V.Palyulin@skoltech.ru.

The group current research interests include various fields of statistical physics and stochastic processes research such as trajectory analysis, reinforcement learning for target search, anomalous diffusion and crowding problems.

Anomalous diffusion challenge: finding the best algorithm to classify trajectories with machine learning
Reinforcement learning for target search optimisation
ML for traffic optimisation problems
Non-backtracking random walks on multiplex networks
Optimisation of escape times by RL approaches

Courses delivered in Skoltech: Stochastic Methods in Mathematical Modeling, Soft Condensed Matter, Thermodynamics and Transport at Nanoscale.

Computational Imaging, Prof. Dmitry Dylov

Google Scholar, D.Dylov@skoltech.ru.

Research focuses mainly on the fundamental optics principals analytics to be interpreted and implemented along with the modern computer vision methods for wide range of tasks in imaging.

Computer vision. A new type of data augmentation for face detection. Unsupervised learning. Training frequency spectra instead of pixel neural networks. A series of successful projects on the application of algorithms in oncology and cardiology.
Visualization systems. Portable MRI device (ongoing). Speeding up clinical MRI setups 16 times. Super-resolution microscopy from a regular camera. Visualization device, autofocus, and detection of objects in the dark and under adverse weather conditions.
Nonlinear method of transmitting energy to signal (image) from noise. The theory of optimal transfer in reinforcement imitation learning, etc.

Courses delivered in Skoltech: Computational Imaging, Biomedical Imaging and Analytics.

AI for Materials Design, Prof. Alexander Shapeev

Google Scholar, A.Shapeev@skoltech.ru.

Modern industrial products are designed on a computer. However, modern materials science algorithms are not efficient for optimizing many materials properties of interest, such as alloy hardness or Li-ion battery power. This research group is working on applying machine learning (ML) to revolutionize computational materials algorithms and molecular modeling, just like ML revolutionized technologies such as computer vision.

Machine-Learning Interatomic Potentials
Machine Learning for Elastic Strain Engineering
Atomistic-to-Continuum Coupling

Courses delivered in Skoltech: Numerical Modeling, Advanced Materials Modeling.

AI-driven Modeling, Prof. Ekaterina Muravleva

Google Scholar, E.Muravleva@skoltech.ru.

Courses delivered in Skoltech: Introduction to Data Science.

Parallel algorithms for AI, Prof. Alexander Mikhalev

Publications, Al.Mikhalev@skoltech.ru.

Courses delivered in Skoltech: Foundations of Software Engineering.

Tensor Networks & Deep Learning, Prof. Andrzej Cichocki

Google Scholar, A.Cichocki@skoltech.ru.