IReNNS: A recurrent neural network with independent neurons and its application in bioinformatics

We propose a simplified architecture for a recurrent neural network designed for learning from structures. We describe the architecture and the implementation and show the performances of the net. Two examples from the science domain are discussed: the first uses a synthetic data set and the second illustrates a chemical problem. We discuss about the results and compare them to other applications, in terms of performances and computational complexity.