Supervisors

Suitable for

Abstract

Graph neural networks (GNNs) are a popular class of models for learning over relational data. The goal of this project is to develop novel GNN models which can handle relational data evolving over time. While the applications are manifold, the incorporation of termporal information brings many additional challenges to GNNs. In this project, the student will explore the space of possible models (including existing ones) and develop a model which can learn adequate continuous time representations that can extrapolate from existing data. The student will drive a rigorous analysis of resulting model properties in comparison with existing literature, and conduct a detailed empirical evaluation.

Prerequisites: Good programming skills, experience with Pytorch/Tensorflow

References: A time-aware message-passing neural network [1], and an interesting way of incorporating time [2]

[1] Wu et al. TeMP: Temporal Message Passing for Temporal Knowledge Graph Completion. EMNLP 2020

[2] Han et al. Temporal Knowledge Graph Forecasting with Neural ODE, arXiv 2101.05151