UnICORNN: A recurrent model for learning very long time dependencies

                     T. Konstantin Rusch 1 Siddhartha Mishra 1

              Abstract               such as in LSTMs (Hochreiter & Schmidhuber, 1997) and
                                GRUs (Cho et al., 2014), where the additive structure of
    The design of recurrent neural networks (RNNs)
                                the gates mitigates the vanishing gradient problem. How-
    to accurately process sequential inputs wi ...