on 28 November 2001
"Recurrent Neural Networks for Prediction: Learning Algorithms, Architectures and Stability," approaches the field of recurrent neural networks from both a practical and a theoretical perspective. Starting from the fundamentals, where unexpected insights are offered even at the level of the dynamical richness of simple neurons, the authors describe many existing algorithms and gradually introduce novel ones. The latter are convicingly shown to yield better prediction performances than traditional approaches, when applied to real-world data. They also dedicate a considerable amount of time on the (practical) issue of nonlinearity analysis of time series, which is or should be, indeed, the cradle of all proper modelling and/or filtering solutions: nonlinearity should be assessed prior to choosing the appropriate model and/or filters, since linear ones are to be preferred if sufficient for the problem. I would recommend this book to any researcher who is active in the field of recurrent neural networks and time series analysis, but also to researchers who are new in the field, since the book offers an extensive overview of the current state-of-the-art approaches.