Tilen Marc: Machine learning on graphs

Povzetek.  A typical machine learning model takes as an input a set of vectors and outputs a prediction about them. In many practical applications the input has also structural properties that can be captured as a graph. In the latest years a lot of effort has been made to create models that use these informations and make predictions about the vertices of the graph. In the talk we will introduce some models that aim to do that and present our recent work to create a model for semi-supervised learning on temporal, possibly heterogeneous graphs - a more complicated setting that captures more practical applications.