prof. Ing. Filip Železný, Ph.D.

The student should develop new methods for machine learning about structured objects described by both real and discrete variables. Such objects are often found in bioinformatics, e.g. molecules are non-trivial structures containing both discrete (e.g. atom types) and real (e.g. charge, weigth) variables. The student should tackle the problems of how to represent such objects, construct relevant features (descriptors) and how to learn and represent hypotheses about such objects. The primary instruments will be mathematical logic, statistical inference, and calculus. http://cs.felk.cvut.cz/en/people/zelezny

The student will invent methods for learning from relational data containing both discrete and continuous variables.

Bioinformatics research has yielded vast amounts of annotated data about various molecules (DNA, proteins, small molecules, etc.). By finding patterns in such data through machine learning and data mining algorithms, one can discover new insights into the biological processes involving such molecules. Current machine learning algorithms are not well adapted to dealing with such data, mainly because the data is structured rather than flat (e.g. atom-atom bonds, etc.), the data instances are large (e.g. 1000's of atoms in a molecule), and the data is noisy. The student should develop novel machine learning methods effective at dealing with this special kind of data. The main instruments will include mathematical logic, graph methods, statistics, and heuristic methods from the field of constraint satisfaction problems (CSP). http://ida.felk.cvut.cz/zelezny/publications.html