A holistic approach to optimizing echo state network through structural evolution and synaptic plasticity

Echo State Network (ESN) is a type of recurrent neural network (RNN) within the Reservoir Computing (RC) paradigm, where only the readout layer is trained. This has made it effective for modeling dynamic systems and time-dependent problems. However, its fixed topology and unmodifiable internal weights limit its further application in complex tasks. To address this, we propose an ensemble ESN model with multiple reservoirs that can adapt its network architecture and synaptic weights using genetic algorithms (GA) and synaptic plasticity. We evaluate various synaptic learning rules (Hebbian, decay-enhanced Hebbian, anti-Oja, and BCM) applied to the ensemble model for real-world tasks. Experimental results show that this approach significantly improves ESN’s learning capability and adaptability.