Approximation of nonlinear functions using Spiking Neural Networks (SNN) based on the STDP algorithm

In this paper, we present a new approach to leveraging the spike-based neural network of Izhikevich model for the approximation of nonlinear functions. Our learning framework is grounded in reinforcement learning and utilizes the Spike-Timing-Dependent Plasticity (STDP) algorithm. This algorithm enables us to adjust the network’s performance by rewarding or penalizing it through modifications to the positive and negative values in the model’s single-neuron differential equations. By employing this innovative method, we can effectively control the behavior of spiking neurons and adjust their responses by modifying target values. This allows us to transform one nonlinear function into another with precision. Inspired by the architecture of the mammalian brain, our network consists of input neurons (sensory neurons) and output neurons (motor neurons). Throughout our research, we encountered various challenges, which we detail in the Network Training section. To assess the performance of our neural network models, we evaluate their accuracy using normalized Mean Squared Error (MSE) and Root Mean Squared Error (RMSE) as key performance indicators. For example, we obtained the values of RMSE and MSE for ۱/X function: ۱.۷ * ۱۰^-۳ and ۴.۱۴ * ۱۰^-۲, respectively.

کلیدواژه ها:

Spiking Neural Networks ، Reinforcement Learning ، dopamine modulated Spike Timing Depending Plasticity