ETD

• machine learning
• neural networks
• reti neurali

Recurrent Neural Networks (RNNs) are amongst the most powerful Machine Learning models to deal with sequential data. Many RNN architectures have been proposed over the years. We review three of the most used RNN architectures: the Standard Recurrent Network, the Long Short-Term Memory and the Gated Recurrent Unit. Furthermore, the Reservoir Computing (RC) has emerged as an alternative paradigm in the area of RNNs, whose the Echo State Network represents the most used model. In addition, minimum complexity RC-based networks have been developed, such as the Delay Line Reservoir and the Simple Cycle Reservoir.. We conduct several experiments on a varied set of problems, in order to compare their performances and analyze their correlation. Furthermore, we create a software framework to build RC models.