A single-hidden-layer neural network for the classification of spike-waveforms

Abstract

The goal of this work is to use a simple neural network to classify and distinguish between pure signals of Gaussian white noise and eight different center-aligned spikes of different correlation levels. Using Python and Tensorflow, our network was trained with a supervised-learning approach. For small noise levels up to 10% (?_noise= 10%max(x)), where max(x) is the maximum spike-amplitude) classification accuracy, sensitivity and specificity are 100%. Higher noise levels, e.g., ?_noise=50%max(x), indicate that the network struggles with two things: distinguishing correlated data and reliably detecting pure noise.