Modeling Neural Plasticity in Echo State Networks for Time Series Prediction

In this paper, we investigate the influence of neural plasticity on the learning performance of echo state networks (ESNs) and supervised learning algorithms in training readout connections for two time series prediction problems including the sunspot time series and the Mackey Glass chaotic system. We implement two different plasticity rules that are expected to improve the prediction performance, namely, anti-Oja learning rule and the Bienenstock-Cooper-Munro (BCM) learning rule combined with both offline and online learning of the read-out connections. Our experimental results have demonstrated that the neural plasticity can more significantly enhance the learning in offline learning than in online learning.