A time series is a sequence of time-ordered data, and it is generally used to describe how a phenomenon evolves over time. Time series forecasting, estimating future values of time series, allows the implementation of decision-making strategies. Deep learning, the currently leading field of machine learning, applied to time series forecasting can cope with complex and high-dimensional time series that cannot be usually handled by other machine learning techniques. The aim of the work is to provide a review of state-of-the-art deep learning architectures for time series forecasting, underline recent advances and open problems, and also pay attention to benchmark data sets. Moreover, the work presents a clear distinction between deep learning architectures that are suitable for short-term and long-term forecasting. With respect to existing literature, the major advantage of the work consists in describing the most recent architectures for time series forecasting, such as Graph Neural Networks, Deep Gaussian Processes, Generative Adversarial Networks, Diffusion Models, and Transformers.
The raster is the most common type of spatio-temporal data, and it can be either regularly or irregularly spaced. Spatio-temporal prediction on regular raster data is crucial for modelling and understanding dynamics in disparate realms, such as environment, traffic, astronomy, remote sensing, gaming and video processing, to name a few. Historically, statistical and classical machine learning methods have been used to model spatio-temporal data, and, in recent years, deep learning has shown outstanding results in regular raster spatio-temporal prediction. This work provides a self-contained review about effective deep learning methods for the prediction of regular raster spatio-temporal data. Each deep learning technique is described in detail, underlining its advantages and drawbacks. Finally, a discussion of relevant aspects and further developments in deep learning for regular raster spatio-temporal prediction is presented.