STAGE Graph signals forecasting with application to transportation

A broad and diverse range of human and natural processes can be seen as time-evolving processes taking place over the nodes of a graph. Take for instance the spread of a virus or falsenews in a social network, measurements recorded by a network of weather stations, or traindelays observed in a railway system. Intuitively, the state of the system at timetdepends onwhat has happened in the past and nodes that are close in the graph tend to exhibit similarbehavior. Modelling such interplay between the behavior of the nodes in time as part of thespread from one side, and the graph structure, from the other, is a surprisingly recent topic. Inparticular, in such a context there is a lack of methods aiming to forecast the value at a givennodevat a given timettaking into account the full view of the problem. The goal of this internship is to make a step further in this direction, which is to design methods that incorporatesboth components and have good performance in practice. The main application of interest willbe the prediction of train delays in the SNCF railway system. This end will be facilitated bythe participation of the SNCF R&D department to the project, hence the student will have theopportunity to interact with both the academic and the industrial labs.