Nengo Implementation of an Unsupervised Oscillatory Neural Network for the Segregation of Auditory Signals

Abstract

In this thesis, we proposed a novel unsupervised oscillatory neural network model for the segregation of auditory signals. The proposed model is inspired by the Locally Excitatory Globally Inhibitory Oscillator Networks (LEGION) model presented in [1]. It consists of relaxation oscillators and a global inhibitor to mimic the neural oscillation. In order to maximize the model’s biological plausibility, we built the proposed model within Nengo (Neural Engineering Object), which is a Python neural simulator based upon the Neural Engineering Framework (NEF). At the end, our model is able to recognize the number of sound sources by analyzing a given correlogram. To ensure the correctness of the simulation results and to observe the proposed model’s cognitive process in the biological substrate, we also compare the simulation results of the proposed model with the ones of Wang’s LEGION model, which we built in MATLAB (Matrix Laboratory).