Modeling of Cortical Signals Using Optimized Echo State Networks with Leaky Integrator Neurons

Echo State Networks (ESNs) is a newly developed recurrent neural network model. It has a special echo state property which entitles it to model nonlinear dynamic systems whose outputs are determined by previous inputs and outputs. The ESN approach has so far been worked out almost exclusively using standard sigmoid networks. Here we will consider ESNs constructed by leaky integrator neurons, which incorporate a time constant and the dynamics can be slowed down. Furthermore, we optimized relevant parameters of the network by Particle Swarm Optimization (PSO) in order to get a higher modeling precision. Here the input signals are spikes distilled from the monkey's motor cortex in an experiment and the outputs are the moving trajectories of the wrist of a monkey in the experiment. The results show that this model can well translate the neuronal firing activities into the desired positions.