When things happen, knowledge about the event and understanding of its importance in context propagates through a variety of world models, leading to patterns of behaviour that in turn affect the system. This task is to build representations and novel modelling techniques that allow these interactions to be instantiated, observed, and leveraged. The current work in this area concentrates on analysis among a restricted set of time-series and flattened event signals, using approaches such as Canonical Correlation Analysis (CCA) to identify potential dependencies. These approaches are prone to weak and conflicting signals and offer little transparency to study the underlying phenomena. Our objective is to look at naturally hierarchical, recurring, dependent structures such as graphs, trees, and a variety of clustering and dimensionality reduction techniques to quantify events, relationships, and allow for more natural exploration of the dynamic financial landscape.
We will provide frameworks for representation that bring transparency and understanding to interactions among events and a deep understanding of the state of the financial universe.
Supervisor: Doctor Marko Grobelnik, Jožef Stefan Institute / marko.grobelnik(at)ijs.si