Talks and presentations

Determination of data complexity using a universal approximating model

October 27, 2019

Talk, Mathematical Methods for Pattern Recognition: the 19th Russian National Conference with International Participation, Moscow, Russia

The properties of the training dataset in classification and regression problems are investigated. We propose a method for determining the complexity of the training data using a two-layer fully connected neural network. The number of neurons in the inner layer of the neural network is used to determine the complexity. Book of abstracts (page 68).

Averaged Heavy Ball Method

October 23, 2019

Talk, 62th Scientific conference at MIPT, Section of Data analysis, recognition and prediction, Moscow, Russia

It is proposed a new approach to the classic Heavy ball method called the Averaged Heavy ball method for unconstrained optimization. This method reduces the peak effect, thereby providing a better convergence rate in the initial iterations. It is provided analysis for the quadratic case and conduct numerical experiments justifying the theory for a wider class of problems. Best talk Award.

Averaged Heavy Ball Method

June 19, 2019

Poster, Traditional Youth School: Control, Information and Optimization, Voronovo, Russia

It is proposed a new approach to the classic Heavy ball method called the Averaged Heavy ball method for unconstrained optimization. This method reduces the peak effect, thereby providing a better convergence rate in the initial iterations. Best poster Award.