Bayesian Methods in Machine Learning

This is a weekly seminar run by the Bayesian methods research group. The seminar considers articles from the leading international conferences, hears presentations by group members about their research, carries out brainstorming sessions, and organizes lectures by leading Russian and international specialists. The seminar is open to all who are interested.

The seminars will focus on research and the application of research in line with the Bayesian approach to probability theory in machine learning and computer vision problem solving. The Bayesian approach has become particularly widespread, across the world, over the past 15 years. Its main features are:

• the ability to automatically establish the structural parameters of machine learning algorithms (choosing the number of clusters, setting the regularization coefficient, selecting relevant features and objects, defining the topology of the neural network etc);
• correct processing of uncertainty that makes it possible to expand classical Boolean logic to situations that contain significant degrees of unknown iformation, meaning that Bayesian methods can be applied to expert systems;
• the ability to account for structural and probability-based connections in Big Data, based on an actively developing graphic models in real time;
• representing data and creating parameters that make it possible to collate the results of observing indirect measures of unknown size with an a prioriview of its characteristic values;

Participants of this special seminar will play an active role in theoretical work to develop new approaches to creating structural parameters and algorithms for machine learning in non-standard problems.

Methodological support for the seminars will come from the Bayesian Methods in Machine Learning and Graphic Models courses given at the HSE’s Faculty of Computer Science and MSU’s Faculty of Computational Mathematics and Cybernetics.

See Seminar webpage for more detailed information.

Seminar moderators: Dmitry Vetrov, Dmitry Kropotov, Michael Figurnov.