Transformers architectures for time series forecasting

Time series forecasting is an important task that finds applications in different contexts, such as predictive maintenance, medicine or finance. A time series forecasting model aims at predicting future values of a target variable given the information available at a certain time step [1]. A time series is represented using a stochastic process. The available information can be for example, measurements from weather stations for climate modeling or sensor data on mechanical processes . Deep learning models have been used to tackle the time series forecasting problem, as these models have the ability to learn complex data representation. More recently, Transformer-based architectures have been proposed as a solution to time series forecasting. Transformers are networks based on attention mechanisms and represent the state-of-the-art in sequence modeling tasks like machine translation and language understanding [2]. Variants of these network architectures have been proposed as models for time series forecasting [3, 4].

The goal of this thesis is to study recent advances in time series forecasting, with particular emphasis on the use of Transformers architectures, and to apply these models to a real problem.