Overview
Graphs are playing an ever increasing role in nowadays systems as a flexible tool to model complex systems. In addition, these systems generate a vast amount of data which can be modelled as signals or features over these graphs. This is for instance the case of infrastructure networks such as water, energy and transportation networks but also the case of wind farms, solar grids and IoTs. Consequently, developing and using machine learning tools to process these graph data is more important than ever. Such a tools need not only capture the graph structure of the data but also account for the dynamics of the topology as practical graphs change over time.
In this PhD project, we are looking for a candidate to work on one of the following fundamental areas:
• Physic-informed graph neural networks: The goal here is to leverage physic information from the medium (such as PDE models) to develop deep learning solutions for spatiotemporal data on graphs.
• Machine learning on higher-order… networks: We want here to use more than pairwise information to learn representations from network data. In particular, we seek to develop and analyze deep learning solutions for this setting.
• Adaptive spatiotemporal learning over graphs: In this area, we want to study machine learning approaches on dynamic graphs by taking into account the evolution of the topology and of the respective signals.
• Application to renewables power forecasting: the outcome of the research in the previous three pillars can potentially be applied on renewable power (e.g. wind) forecasting.
Candidates with other interest within the graph machine learning topic are also encouraged to apply by stating so in their application package. In whatever area, the candidate is expected to spend around 20% of their time on applying these techniques on renewables,
The project will be carried out in the research group of Dr. E. Isufi and co-supervied by Dr. H. Jamali-Rad from TU Delft / Shell. Dr. Isufi’s group at TU Delft works on fundamental research on graph signal processing and machine learning. We focus on both theoretical and applied research especially to recommender systems (in the Multimedia Computing Group) and water networks (in Aidrolab). Dr. Jamali-Rad’s group at TU Delft focuses on deep representation learning and self-supervised learning applied to variety of downstream tasks, including but not limited to computer vision. At Shell, Dr. Jamali-Rad leads a major portfolio of AI projects mostly focused on renewable power and biotechnology.
You will be offered quite a flexibility in the project, hence candidates able of working independently, eager to learn and grow as scientific researchers are most affiliated. You will also be collaborating with other senior PhD researchers in the group and will supervise master and bachelor theses
Company:
Delft University of Technology (TU Delft)
Qualifications:
Language requirements:
Specific requirements:
Educational level:
Level of experience (years):
Senior (5+ years of experience)
