Graph-Time Neural Networks
A Principled and Robust Approach to Learning from Relational Time Series
M. Sabbaqi (TU Delft - Electrical Engineering, Mathematics and Computer Science)
A. Hanjalic – Promotor (TU Delft - Electrical Engineering, Mathematics and Computer Science)
E. Isufi – Promotor (TU Delft - Electrical Engineering, Mathematics and Computer Science)
R. Taormina – Copromotor (TU Delft - Civil Engineering & Geosciences)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Learning representations from relational time series is a fundamental problem in many applications involving social networks, finance, and urban water systems, among others. In these domains, a wide variety of tasks require modeling relational time series for imputating missing ratings in a social network, forecasting of stock prices in a financial market, and anomaly detection in urban water systems. Such data upholds complex relatio-temporal dependencies that introduce multiple challenges per se. Data efficiency, robustness to structural perturbations, scalability to large-scale networks, interpretability, and transferability are some main challenges pertaining to the full scale deployment of machine learning solutions in this setting. However, the underlying fundamental challenge is to jointly model the temporal patterns and the relational dependencies simultaneously underpinning the aforementioned issues. This dissertation aims to study machine learning models for relational time series for this setup....