Graph Deep Learning for Time Series Forecasting

Beschreibung

Neural networks have been used to forecast time series for decades. One of the key elements enabling most of the field's recent achievements is the training of a single neural network on large collections of related time series. This approach, however, often considers each time series independently from the others and, consequently, discards dependencies that might be instrumental for accurate predictions. 

This book's purpose is to aim to address the shortcomings of the state of the art in correlated time series forecasting by relying on graph representations and graph deep learning methods. The author proposes graph-based predictors that model pairwise relationships among time series by conditioning forecasts on a (possibly dynamic) graph spanning the collection. The research introduces a comprehensive methodological framework characterizing the family of these predictive models and provides design principles for graph-based forecasting. Within this context, the book proposes methods to tackle the inherent challenges of the field, i.e., dealing with irregular observations, local effects, and latent relational dependencies. In addition, it addresses the computational scalability of the proposed framework, together with methodologies enabling transfer learning and hierarchical forecasting.

Extensive empirical results validate the introduced methodologies and place graph deep learning methods among the most important tools available in modern forecasting.

This book is a revised version of the PhD dissertation written by the author to receive his PhD from the Università della Svizzera italiana, Lugano, Switzerland. His current research focuses on machine learning methods for time series and graph processing, with applications in cyber-physical systems.  

Neural networks have been used to forecast time series for decades. One of the key elements enabling most of the field's recent achievements is the training of a single neural network on large collections of related time series. This approach, however, often considers each time series independently from the others and, consequently, discards dependencies that might be instrumental for accurate predictions. This book's purpose is to aim to address the shortcomings of the state of the art in correlated time series forecasting by relying on graph representations and graph deep learning methods.

The author proposes graph-based predictors that model pairwise relationships among time series by conditioning forecasts on a (possibly dynamic) graph spanning the collection. The research introduces a comprehensive methodological framework characterizing the family of these predictive models and provides design principles for graph-based forecasting. Within this context, the book proposes methods to tackle the inherent challenges of the field, i.e., dealing with irregular observations, local effects, and latent relational dependencies. In addition, it addresses the computational scalability of the proposed framework, together with methodologies enabling transfer learning and hierarchical forecasting. Extensive empirical results validate the introduced methodologies and place graph deep learning methods among the most important tools available in modern forecasting.

This book is a revised version of the PhD dissertation written by the author to receive his PhD from the Università della Svizzera italiana, Lugano, Switzerland. His current research focuses on machine learning methods for time series and graph processing, with applications in cyber-physical systems. 

Autor*in

Andrea Cini

Themen in »Graph Deep Learning for Time Series Forecasting«

Time series forecasting Graph-based predictor Predictive modelling Pairwise relationship Transfer learning Computational scalability Hierarchical forecasting

Stimmen zu »Graph Deep Learning for Time Series Forecasting«

Details

Link teilen