Laboratory for Structural Methods of Data Analysis in Predictive Modelling

The Laboratory has proposed:

• a robust principal component method;
• a method of statistical estimation finite samples when noise and error are present in the model specification;
• a new boosting algorithms;
• a universal gradient method for problems of convex optimization;
• new methods of stochastic analysis and numerical modeling of stochastic processes. 

Implemented results of research: Proposed methods and approaches are actively used in solving problems in machine learning (predictive modeling, parameter evaluation), financial mathematics (evaluation of derivatives), transportation modeling, telecommunications and inn other fields.

Education and career development:

• Members of the academic staff of the Laboratory have provided training in statistics and optimization.
• The Laboratory has conducted conferences, summer schools, mini-course for students with participation of leading foreign scientists.

Collaborations: Joseph Fourier University (France), Weierstrass Institute (Germany), Center for Operations Research and Econometric(Belgium): joint research