Mathematical Institute
Topic
Sparse symmetric graphs in machine learning
PhD. program
9-1-9 Applied mathematics
Name of the supervisor
doc. Ondrej Šuch, PhD., M.Sc.
Contact:
Receiving school
Faculty of Mathematics, Physics and Informatics, Comenius University, Bratislava
Annotation
A popular approach to building complex classification models is to create an ensemble of simpler pairwise models. Combining pairwise models corresponds to a graph structure. Employing symmetric graphs allows us to prevent a potentially large prediction error, which occurs when a hard to distinguish class lacks sufficient coverage by a dense subset of the ensemble. The research topic combines deep neural networks, classical machine learning and theory of symmetric graphs (Cayley graphs).