Scientific activity

The Center conducts research at the intersection of two rapidly growing areas of data analysis: deep learning and Bayesian methods of machine learning. Deep learning is a branch that involves the construction of very complex models (neural networks) for solving such tasks as image or music classification, style transfer from the image to the photo, following words prediction. Bayesian approach utilizes probability theory and mathematical statistics for solving the same problems. In recent years, Bayesian methods have been actively used in combination with neural network models to improve their predictive capacity, to adapt model to specific data more quickly and to solve a wider range of tasks. Research papers on this topic constitute important section of the world's leading conferences on machine learning (NeurIPS, ICML). The Centre's staff regularly publish works on these conferences and take part in conferences workshops.

Areas of research:

1.  Approximate Bayesian inference
2.  Variational Autoencoders
3.  Tensoring of neural network
4.  Accelerating of convolutional networks
5.  Stochastic optimization
6.  Diffusion Models.