From Graphs to Graph Neural Networks: Applications in Multi-Modal Heterogeneous Real-World Data

Abstract

In this talk, I will introduce the basic ideas behind graphs and graph neural networks (GNNs), and discuss why they have become powerful tools for modelling complex, structured, and heterogeneous real-world data. I will begin with an overview of graph-based representations and the core principles of message passing and graph representation learning, before moving to a range of AI applications where GNNs provide advantages over conventional learning approaches. In particular, I will discuss applications in heterogeneous and multimodal settings, including privacy-preserving federated learning, neuro-symbolic modelling, and emerging graph-based methods for spatial and multi-modal data integration. Overall, the talk aims to connect the foundations of graph learning with current AI challenges involving structure, heterogeneity, and multi-source data.

Short bio

Soheila Molaei is a senior researcher in the Department of Engineering Science at the University of Oxford. Her research focuses on artificial intelligence and machine learning, with a particular emphasis on graph neural networks, federated learning, neuro-symbolic AI, and learning from multimodal and heterogeneous data. Her work explores how graph-based methods can support robust, interpretable, and privacy-aware AI systems across complex real-world applications. She has also collaborated with researchers at the University of Bergen on graph-based and neuro-symbolic learning research.