Output details

Efficient online recurrent connectionist learning with the ensemble Kalman filter

<24> Paper published in the 2nd highest ranking neural networks journal, with an H-index of 65. This paper proposes a new approach to complexity reduction in online learning of RNNs through sequential Bayesian filtering. We develop a fast ensemble Kalman filter for derivative-free parameter estimation of recurrent neural networks (RNNs).Through forecasting experiments on observed data from nonlinear systems, it is shown that our system outperforms other RNN training algorithms in terms of real computational time and accuracy of forecasts. This paper has been cited by researchers reporting applications of our learning algorithm to digital watermarking and image fusion.